Descriptions and Beyond - PDF Free Download

Descriptions and Beyond This page intentionally left blank Descriptions and Beyond edited by M A RG A R E I M E R ...

Author: Marga Reimer | Anne Bezuidenhout

83 downloads 1305 Views 3MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Descriptions and Beyond

This page intentionally left blank

Descriptions and Beyond edited by

M A RG A R E I M E R and

ANNE BEZUIDENHOUT

CLARENDON PRESS OXFORD

3

Great Clarendon Street, Oxford ox2 6dp Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Bangkok Buenos Aires Cape Town Chennai Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi Sa˜o Paulo Shanghai Taipei Tokyo Toronto Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York ß the several contributors 2004 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2004 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose this same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available ISBN 0–19–927051–1 ISBN 0–19–927052–x (pbk.) 10 9 8 7 6 5 4 3 2 1 Typeset by Kolam Information Services Pvt. Ltd, Pondicherry, India Printed in Great Britain on acid-free paper by Biddles Ltd., King’s Lynn, Norfolk

Acknowledgements

It is our hope that the publication of this volume will encourage a continuing dialogue between philosophers of language and linguists. There is a growing trend among researchers from these two traditionally distinct Welds to take note of one another’s work and to write in a way that is accessible to both sorts of audiences. This is surely a beneWcial trend, as an understanding of alternative perspectives on natural language can only increase one’s appreciation of the subtlety and complexity of the data. We would like to acknowledge all of our contributors, who have been immensely patient and understanding during the considerable time it has taken to bring this volume to press. Peter MomtchiloV and Rebecca Bryant at Oxford University Press smoothed the process and ensured that the Wnished product was worth all the time and eVort. Thanks also to Sid LittleWeld for editorial assistance. Kent Bach, David Chalmers, Michael Devitt, Bart Geurts, Michael Glanzberg, Robin Jeshion, Bruce Martin, Mark Sainsbury, Suzanna Siegel, Gretta Slabey, and Zolta´n Szabo´ gave us helpful advice regarding both the general and section introductions included in this volume. We are of course responsible for any errors that remain. Anne Bezuidenhout would like to thank her colleague in the Linguistics Program at the University of South Carolina, Stan Dubinsky, and the students in her philosophy of language seminar in Fall 2001, who read earlier drafts of many of these chapters and gave us excellent feedback. Thanks especially to Philippe Albert, Johnny Hancock, Changyong Liao, German Lopez-Hernandez, and Lan Zhang. A. B. M. R.

This page intentionally left blank

Contents

Notes on Contributors

ix

Introductory Remarks

1

I. Incomplete Descriptions 1. Descriptions and Situations

7 15

Franc¸ois Recanati

2. An Abuse of Context in Semantics: The Case of Incomplete DeWnite Descriptions

41

Ernie Lepore

3. This, That, and The Other

68

Stephen Neale

II. The Referential–Attributive Distinction 4. Descriptions: Points of Reference

183 189

Kent Bach

5. The Good, the Bad, and the Ugly

230

Nathan Salmon

6. Descriptive Indexicals and Indexical Descriptions

261

GeoVrey Nunberg

7. The Case for Referential Descriptions

280

Michael Devitt

III. Presupposition and Truth-Value Gaps 8. Would You Believe It? The King of France is Back! (Presuppositions and Truth-Value Intuitions)

307 315

Kai von Fintel

9. Descriptions, Linguistic Topic/Comment, and Negative Existentials Jay Atlas

342

Contents

viii

IV. Representation of DeWnites and IndeWnites in Semantic Theory 10. Referring Descriptions

361 369

R. M. Sainsbury

11. The Proper Form of Semantics

390

Joseph Almog

12. On a Unitary Semantical Analysis for DeWnite and IndeWnite Descriptions

420

Peter Ludlow and Gabriel Segal

V. Anaphoric Pronouns and Descriptions, IndeWnites, and Dynamic Semantics/Syntax 13. IndeWnites and Anaphoric Dependence: A Case for Dynamic Semantics or Pragmatics?

437 455

Richard Breheny

14. Grounding Dynamic Semantics

484

Paul Dekker

15. Pronouns as DeWnites

503

Craige Roberts

16. Dynamic DeWnite Descriptions, Implicit Arguments, and Familiarity

544

Alice G. B. ter Meulen

17. IndeWnites and Scope Choice

558

Ruth Kempson and Wilfried Meyer-Viol

VI. Names and Descriptions 18. Descriptive Descriptive Names

585 591

Robin Jeshion

19. Descriptively Introduced Names

613

Marga Reimer

References

630

Index

651

Notes on Contributors

Joseph Almog is a professor of philosophy at UCLA. His research interests are in metaphysics and philosophy of language and logic. Jay Atlas is the Peter W. Stanley Professor of Linguistics and Philosophy at Pomona College. He was educated at the Phillips Exeter Academy, Amhurst College, and Princeton University. His research interests are in philosophy of language, logic, and language and thought. Kent Bach, professor of philosophy at San Francisco State University, was educated at Harvard College and University of California, Berkeley. He has written extensively in philosophy of language, theory of knowledge, and philosophy of mind. His books include Thought and Reference (Oxford, 1987, expanded edition 1994) and, with Michael Harnish, Linguistic Communication and Speech Acts (MIT Press, 1979). Richard Breheny is a senior researcher at RCEAL, University of Cambridge. His research interests lie mostly in the Welds of semantics and pragmatics and include context dependence, theories of communication, language processing, and cognitive development. Paul Dekker is a post doctoral researcher leading a project on Formal Language Games in the ILLC/Department of Philosophy of the University of Amsterdam. His research interests are focused on the semantics/pragmatics interface. Michael Devitt is a Distinguished Professor of Philosophy at the Graduate Center of the City College of New York. His main research interests are in the philosophy of mind and language but he frequently does missionary work for the cause of realism. Kai von Fintel is an associate professor of linguistics in the Department of Linguistics and Philosophy at MIT. His research interests are in the intersections of semantics, pragmatics, and philosophy of language. Robin Jeshion is a professor of philosophy at Yale University. Her research falls primarily at the intersections of philosophy of language, philosophy of mind, and epistemology, and includes the nature of de re attitudes, reference, and a priori knowledge. She is currently working on analyses of demonstrative thought and mathematical intuition. Ruth Kempson is a professor of General Linguistics and a member of the Philosophy Department, King’s College London. She currently holds a Leverhulme Research Professorship. The focus of her current research is the development of the new

x

Notes on Contributors

Dynamic Syntax framework, a grammar formalism which characterizes syntactic/ semantic properties of natural language in terms of the left-to-right dynamics of processing language in context. Ernie Lepore is a professor of philosophy at Rutgers University. His research interests include philosophy of language and mind. Peter Ludlow is a professor in philosophy and in linguistics at the University of Michigan. His current research interests include the philosophy of linguistics, the semantics and logical form of noun phrases, the semantics of tense and the metaphysics of time, and the syntax and semantics of intensional environments in natural language. Alice G. B. ter Meulen is professor of Linguistics at the University of Groningen (NL), aYliated with the Center for Language and Cognition and the graduate school of Behavioral and Cognitive Neurosciences. Her primary research is concerned with the semantic representation of tense and aspect, and its relation to temporal reasoning in ordinary English. Her more general research program bears on information structure and dynamic semantics, analyzing how presupposed, entailed, or asserted information may diVer from what may be left implicit in discourse. Wilfried Meyer-Viol is on the faculty in the philosophy department at King’s College London. His research interests include tree logics, dynamic syntax, epsilon calculus, and formal properties of grammars. Stephen Neale is a professor of philosophy at Rutgers University in New Brunswick. His interests are in philosophy of language and mind, philosophical logic, and metaphysics. GeoV rey Nunberg is a senior researcher at the Center for the Study of Language and Information at Stanford University and a consulting professor in the Stanford Department of Linguistics. His research interests include pragmatics and lexical semantics. Franc¸ois Recanati is ‘directeur de recherche’ at the National Center for ScientiWc Research (CNRS) in Paris, and a member of Institut Jean Nicod. His books include Meaning and Force (CUP, 1988), Direct Reference (Blackwell, 1993), Oratio Obliqua, Oratio Recta (MIT Press/Bradford Books, 2000), and Literal Meaning (CUP, 2004). Marga Reimer is an Associate Professor in University of Arizona’s department of philosophy, where she has taught since 1992. Her current interests include the nature of linguistic convention and non-literal uses of language, especially metaphor. Craige Roberts is an associate professor of linguistics at the Ohio State University. She received her Ph.D. from the University of Massachusetts at Amherst in 1987. She studies formal semantics and pragmatics, with particular interests in anaphora and

Notes on Contributors

xi

deWnites, the semantics of modals and tense, information structure, and the semantics and pragmatics of prosody. R. M. Sainsbury is professor of philosophy at the University of Texas at Austin and Susan Stebbing Professor of Philosophy at King’s College London. He is the author of Russell (1979), Paradoxes (1995), Logical Forms (2000) and Departing From Frege (2002). He is currently working on a book entitled Reference Without Referents. Nathan Salmon is professor of philosophy at the University of California, Santa Barbara. He has published numerous articles in metaphysics and the philosophy of language, most of which are slated to appear in his collected papers, Existence, Essence, and Reference. He has written two previous books, Frege’s Puzzle (Ridgeview Press) and Reference and Essence (Prometheus Books), which was awarded the Arlt Award in the Humanities by the Council of Graduate Schools in the United States. Gabriel Segal is professor of philosophy at King’s College London. His research covers various aspects of philosophy of language and philosophy of psychology, including semantics for natural languages and philosophy of linguistics.

This page intentionally left blank

Introductory Remarks

This collection contains chapters on six general topics: 1. 2. 3. 4. 5.

Incomplete descriptions The referential-attributive distinction Presupposition and truth-value gaps Representation of deWnites and indeWnites in semantic theory Anaphoric pronouns and descriptions, indeWnites, and dynamic syntax/ semantics 6. Names and descriptions

What motivates this particular selection of topics? There is a sense in which the selection of topics is self-motivating, as it reXects some of the current interests of philosophers of language and linguists. Contributors (all of them philosophers or linguists) were invited to write on any issue concerning descriptions and/or related phenomena that they found to be of interest, and which they believed would be of interest to both philosophers of language and linguists. Introductions accompany each of the six parts that comprise this collection. These introductions are intended to provide the background relevant to the speciWc issues addressed in the respective sections. The purpose of the present remarks is to unify, to the extent possible, what might initially appear to be six disparate topics. Brevity will be sought, so as to minimize repetition of material covered in the section introductions. It is natural to begin thinking about descriptions with Russell’s (1905) Theory of Descriptions and Strawson’s (1950) reply to that theory. Indeed, beginning with Russell and Strawson may well be inevitable, as much (although not all) of contemporary work on descriptions has been inXuenced by the debate between these two philosophers. Russell did a brilliant job of motivating his quantiWcational account of descriptions by showing how it was able to solve logical puzzles that seemed intractable on alternative accounts. Strawson’s response was as trenchant as Russell’s theory was brilliant. Russell’s theory was based (according to Strawson) on certain fundamental confusions, including a meaning/reference confusion and a related sentence/statement confusion. These (and other) confusions led Russell to overlook the importance of context in determining what is said (stated, asserted). So Strawson claimed. Alleged

2

Introductory Remarks

counter-examples to Russell’s theory were cited to bear this out. Strawson famously appealed to what are now known as ‘incomplete’ descriptions, descriptions that fail to denote uniquely. Consider, for instance, an utterance of ‘The table is covered with books’. Here, the description used is incomplete, as there is more than one table in existence. Russell’s theory would seem to predict (incorrectly) that the utterance in question would have to be false, given the plurality of tables. For Russell’s theory claims that the proposition expressed by such an utterance would be one to the eVect that there exists a unique table and that whatever is a table is covered with books. But according to Strawson, there is only a presupposition to the eVect that there is a unique table to which the speaker is referring. In assertively uttering the sentence in question, the speaker is saying of a particular table (the table to which he is referring) that it is covered with books. The debate as to whether incomplete descriptions pose a genuine threat to Russell’s theory continues unabated over Wfty years later. All three chapters in the section on incomplete descriptions take the position that such expressions pose no threat to Russell’s Theory of Descriptions. The chapters by Recanati and Neale attempt to show that Russell’s theory (properly reWned) does indeed accommodate the fact that it is possible to ‘say something’ true via the utterance of a sentence containing an incomplete description. Both philosophers regard the proper incorporation of contextual elements as the key to solving the incompleteness problem. Lepore’s chapter takes a truly radical stance, rejecting as semantically irrelevant the very notion of ‘what is said’. Lepore is thus led to claim that whatever might be ‘said’ by way of sentences containing incomplete descriptions is simply irrelevant to a semantic theory like Russell’s. Some opponents to Russell’s Theory of Descriptions (including Devitt, this volume) feel that incomplete descriptions pose an intractable problem for that theory, one that can be handled only by viewing some uses of descriptions as referential. In a ground-breaking paper, Donnellan (1966) drew attention to a phenomenon that he believed undermined Russell’s Theory of Descriptions. He argued on behalf of what has come to be known as the ‘referential/attributive distinction’. To use a description ‘attributively’ is to use it to say something about whoever or whatever happens to be the (unique) F . The statement made will be true just in case the (unique) F is G. To use a description ‘referentially’ is to use it to say something about a particular entity (x) to which the speaker intends to draw the audience’s attention—an entity which may, or may not, be an F . The statement made will be true just in case x is G. While Russell’s theory captures the attributive use, it fails to capture the referential use. So argued Donnellan. As the chapters in the section on the referential/attributive distinction make clear, the debate over whether there is a semantically signiWcant referential use that undermines Russell’s theory is far from over. Two of the chapters (Bach’s and Salmon’s) put forth novel methodological arguments in favor of a unitary Russellian account of descriptions. A third chapter (Nunberg’s) presents an empirically based argument for a unitary Russellian account, and a fourth chapter (Devitt’s) presents a multifaceted defense of a semantically signiWcant referential use.

Introductory Remarks

3

In looking at alleged counter-examples to Russell’s theory, including incomplete descriptions and referential uses of descriptions, it is important not to lose sight of the arguments that seem to favor a Russellian account of descriptions. As noted above, Russell defended his theory by claiming that it solved a variety of logical puzzles. One of these puzzles involved sentences containing nondenoting descriptions: descriptions that are satisWed by nothing. How is it possible to utter such a sentence, and to thereby say something meaningful ? An utterance of ‘The present king of France is bald’ would be perfectly meaningful—even in the absence of any (unique) king of France. Strawson thought that the sentence uttered would be meaningful, although the utterance would be without a truth value. Russell’s Theory of Descriptions predicted that the sentence uttered (more precisely, the proposition it expressed) would not only be meaningful, but false. Russell thought the prediction was accurate: indeed, he claimed that the proposition expressed was ‘plainly false’. On the face of it, we have an argument based on a single intuition—an intuition that is surely not universally shared. Indeed, Strawson thought it was obvious that the utterance in question would be neither true nor false. He was thereby taken by many to suggest that any utterance containing a nondenoting description would lack a truth-value. Yet Strawson (1964) later admitted that certain uses of nondenoting descriptions could indeed be used to make false statements. Consider, for instance, an assertive utterance of ‘I saw the king of France at the University of Arizona this morning’, made on March 18, 2003. Intuitively, such an utterance would be false, not without a truth-value. So, what is one to do when faced with intuitions that seem to point in diVerent directions? How is one to decide which intuitions are accurate? Is there any way out of the ‘intuition-mongering’ that seems inevitably to be generated by debates over the truth-value of particular utterances? These concerns (in addition to others) are reXected in the chapters dealing with presupposition and truth-value gaps. Both von Fintel and Atlas argue that there is indeed a principled way of dealing with these truth-value intuitions, although the principles they oVer are quite diVerent, and yield diVerent results in a number of key cases. Although many philosophers of language accept either Russell’s unitary theory of descriptions or a hybrid theory of the sort proposed by Donnellan, some see the need for a genuinely novel account of descriptions. These philosophers have rejected the idea that utterances of ‘The F is G’ ever have the logical form Russell attributed to them. All three of the chapters in the section on the representation of deWnites and indeWnites in semantic theory propose novel accounts of descriptions, breaking away from the Russellian paradigm. Sainsbury argues that deWnite descriptions are in fact referring expressions, rather than quantiWers. Nevertheless, Russell’s theory captures the truth conditions (although not the logical form) of sentences of the form ‘The F is G’. Almog takes a more radical view, rejecting the ‘translational approach’ to semantics, which aims to assign to each sentence of a natural language the proposition it expresses. Ludlow and Segal argue that deWnite descriptions are semantically equivalent to indeWnite descriptions. They further contend that any deWniteness or uniqueness

4

Introductory Remarks

implication associated with deWnite descriptions is derived pragmatically, by means of Gricean conversational principles and maxims. IndeWnite descriptions have received a considerable amount of attention from both linguists and philosophers. Philosophers tend to be interested in the logical relations between deWnite and indeWnite descriptions. Are both existential quantiWers, as argued by Ludlow and Neale (1991)? Are both the same quantiWer, as argued by Ludlow and Segal (this volume)? Some philosophers, inspired by the pioneering work of Geach (1962) and Evans (1977, 1980), are interested in the relation between indeWnites and anaphora. Indeed, both Neale (1990) and Stalnaker (1998) have done inXuential work on this particular topic. An interest in indeWnites and anaphora is shared by many linguists. Linguists writing on the topic today show the inXuence of fellow linguists, including Kamp (1981) and Heim (1990a). However, these linguists also show an appreciation and understanding of the work of philosophers like Evans, Neale, and Stalnaker. The linguist’s interest in the relation between descriptions and anaphora is evident from the chapters in the section dealing with anaphoric pronouns and descriptions, indeWnites and dynamic syntax/semantics. Here, a central issue concerns the interpretation of indeWnite descriptions and the pronouns and deWnite descriptions that are anaphoric on them. Three linguists give three diVerent analyses. Breheny argues that such pronouns are to be treated as quantiWers, whereas Dekker treats such pronouns as referring expressions. Roberts argues that all pronouns are deWnites whose proVered contents are discourse referents and whose interpretations are constrained by various sorts of presuppositions. A second issue (addressed by ter Meulen) concerns the distinction between indeWnites that are asserted (as in ‘One of the marbles is not in the bag’) and those that are inferred (as in ‘Nine of the ten marbles are in the bag’). Ter Meulen notes and characterizes the diVering potential of these two types of indeWnites to act as antecedents for pronouns. A third issue concerns the connections between the scopal properties of indeWnites and anaphoric pronoun resolution. Kempson and Meyer-Viol draw attention to a variety of examples that suggest parallels between the way in which anaphoric pronouns depend on their antecedents and the scopal properties of indeWnites. Philosophers of language, like linguists, often look to other linguistic phenomena when theorizing about deWnite descriptions. Philosophers have a long-standing interest in the relation between descriptions and proper names, an interest matched by the linguist’s interest in the relation between descriptions and anaphora. Indeed, Russell’s ‘On Denoting’ was in large part a diatribe against the view that deWnite descriptions and proper names are semantically of a piece. The idea that descriptions could ever function as logically proper names was soundly rejected by Russell. Interestingly, Russell defended the idea that ordinary proper names function as descriptions. This idea was much criticized by Kripke (1980) and others, although it still has its advocates. There is a sense among many that, if there are proper names that function as descriptions, then these are proper names whose reference is Wxed by description. ‘Neptune’ is a case in point, having been stipulated by nineteenth-century astronomer Leverrier to

Introductory Remarks

5

refer to whatever was the planetary cause of Uranus’s orbital perturbations. The case for a descriptive interpretation of such expressions was powerfully made by Evans (1979), although it has been criticized by Millians like Donnellan (1979), Salmon (1987), and Soames (1998a). As the two chapters in the section dealing with names and descriptions suggest, the status of these expressions remains unresolved. While Jeshion argues that the descriptive account of descriptive names is insuYciently motivated, Reimer argues otherwise. The particular selection of topics included in this volume does not purport to be exhaustive. It does, however, purport to be representative of a number of issues, concerning descriptions and related phenomena, that currently occupy the minds of both philosophers of language and linguists.

This page intentionally left blank

PA RT I

Incomplete Descriptions

In ‘On Denoting’ (1905), Russell famously argued that deWnite descriptions are quantiWer expressions. More precisely, Russell claimed that the proposition expressed by an assertive utterance of a sentence of the form The F is G, is logically equivalent to: 9x(Fx & 8y(Fy ! y ¼ x) & Gx).1 Less formally, the proposition expressed is a conjunctive one to the eVect that there is exactly one F and whatever is F is G. Thus, consider Russell’s well-known example of (1)

The present king of France is bald.

According to Russell’s Theory of Descriptions, the proposition expressed by any assertive utterance of this sentence is one to the eVect that there exists exactly one present king of France and whatever is presently king of France is bald. Although Russell was aware that his proposed analysis of deWnite descriptions might appear unintuitive because of its complexity, he argued that it was the only analysis of such expressions equipped to solve a host of logical puzzles—puzzles concerning identity statements, propositional attitude attributions, nondenoting descriptions (including negative existentials), and the law of excluded middle. Russell insisted that the view against which he was inveighing—the view that deWnite descriptions mean what (if anything) they denote2—was eVectively refuted by these same puzzles. To see this, consider the following six sentences: (2) (3) (4) (5) (6) (7)

The current US president is George W. Bush. The present king of France does not exist. Fred believes that George is the current US president. Fred believes that George is George. The present king of France is bald. The present king of France is not bald.

1 Russell’s own (1905) rendering employs diVerent notation and is considerably less elegant than this now standard rendering. 2 Although this was Russell’s primary target in ‘On Denoting’, he was also concerned to undermine Frege’s (1892/1970) view, according to which terms that refer to ‘deWnite objects’ exhibit a sense/reference distinction. On Frege’s view, sentences containing nondenoting descriptions in the position of grammatical subject express propositions that are without truth-value; on Russell’s view, the propositions expressed by such sentences are false.

8

I. Incomplete Descriptions

Suppose we adopt the view that deWnite descriptions (as well as ordinary proper names) are logically proper names: that their meaning is nothing more nor less than their reference. Suppose further that the year is 2003. Then (2), which is intuitively both true and informative, will emerge as expressing a trivial truth of identity.3 Sentence (3), which is intuitively true, will emerge as meaningless (if not self-contradictory) on account of the meaninglessness of the subject term.4 If descriptions are logically proper names, then (4) should be semantically equivalent to (5), since the sentences are identical save for diVerent, though co-referring, terms. But suppose we make the plausible assumption that semantic equivalence requires identity of truth conditions. In that case, (4) is clearly not semantically equivalent to (5): the latter might be true while the former is false. Sentences like (6), sentences containing nondenoting descriptions, descriptions that fail to describe anything real, pose yet another sort of puzzle. In particular, such sentences appear to generate violations of the law of excluded middle. If there is no present king of France, then (6) is not true. Does this mean that (7) is true? Intuitively not; (7) would appear to be true just in case there exists a unique king of France and he is not bald. But if that is correct, we would appear to have a violation of excluded middle: either p or not-p. None of (2) through (7) are problematic on a Russellian account of deWnite descriptions. In particular, (2) gets interpreted as expressing an informative proposition to the eVect that there exists a unique current president of the US and that he is identical to George W. Bush. (3) gets interpreted as expressing a true proposition to the eVect that it is not the case that there exists a unique present king of France. (4) gets interpreted as attributing to Fred a belief about George to the eVect that he is the unique current US president, while (5) might be used to attribute to Fred a belief in George’s self-identity. (7) emerges as ambiguous: true on one reading and false on another, thereby preserving the law of excluded middle. In particular, the sentence expresses a truth when the negation is given wide scope (it is false that there exists . . . ) and expresses a falsehood when it is given narrow scope (there exists a unique . . . and it is false that . . . ). There is thus no violation of excluded middle. No doubt, the ability of Russell’s Theory of Descriptions to solve such puzzles is largely responsible for its continued popularity among philosophers of language. Another likely reason for its popularity is the fact that deWnite descriptions appear to behave in many ways like other expressions standardly treated as quantiWers.5 This lends credence to the view that deWnite descriptions are quantiWers rather than logically proper names (or ‘referring terms’).

3 We are assuming here and elsewhere a compositionality principle to the eVect that the meaning of a complex expression is a function of the meaning of its parts, plus syntax. 4 Alternatively, one might claim that the description is meaningful because it denotes the nonexistent present king of France. But most would agree with Russell that one should avoid positing such ‘unreal’ entities if it is possible to do without them. 5 See, for instance, Neale (1990).

I. Incomplete Descriptions

9

Forty-Wve years after the publication of ‘On Denoting’, Strawson (1950) argued that Russell’s analysis was subject to some rather obvious counter-examples. One sort of counter-example involves deWnite descriptions that fail to uniquely denote: that fail to be ‘satisWed’ by exactly one F . Such descriptions are often called ‘incomplete’. The diYculty, for the Russellian, is to explain the apparent fact that utterances of sentences of the form The F is G can be true even in the absence of a unique F . Thus, consider Strawson’s example of an assertive utterance of: (8)

The table is covered with books.

Although there are many tables in existence, it seems absurd to suppose that this fact would preclude any utterance of (8) from emerging as true. As Strawson emphasizes, what seems relevant to the felicity of such an utterance is not that there be just one table, but that there be just one table to which the speaker is referring. At any rate, if Russell’s theory is accepted as true, then any utterance of a sentence containing an incomplete description (in the position of grammatical subject) would be false on the grounds of a failed uniqueness condition. This has struck many as highly unintuitive. Call the resultant task of accounting for this unintuitive consequence ‘the problem of incomplete descriptions’. Some philosophers have concluded (with Strawson) that the problem of incomplete descriptions is insurmountable for the Russellian. Some of these philosophers have concluded further that descriptions can at least sometimes function semantically as referring expressions.6 But many other philosophers believe that the phenomenon of incomplete descriptions is easily accommodated within a Russellian framework. Over the past several decades, several quantiWcational approaches to the problem of incomplete descriptions have emerged. Two of these have stood out in the literature. Following Lepore (this volume), these might be called the ‘semantic’ and ‘pragmatic’ approaches to incomplete descriptions. The semantic approach treats incomplete descriptions as contextually sensitive quantiWer phrases that are somehow ‘completed’ by the context of utterance.7 There are two ways in which such completion might be eVected. First, one might view the context of utterance as somehow constraining the domain of quantiWcation in such a way that, in the case of a felicitous utterance of The F is G, the domain contains exactly one F . Following Neale (1990), this might be referred to as the ‘implicit’ approach to incompleteness. Second, one might view the context of utterance as somehow supplementing the descriptive content of the incomplete description, so that the resultant description, the F which is H, is complete, at least where the utterance is felicitous.8 What is said on such an interpretation is what would have been said had the speaker uttered ‘The F which is H is G.’ Following Neale (1990), this might be referred to as the ‘explicit’ approach to incompleteness. To get a sense of 6 See, for instance, Donnellan (1966), Wettstein (1981, 1983), and Devitt (1981b, this volume). 7 See Neale (1990) for a defense of this approach. 8 This approach is criticized by Recanati, Lepore, and Devitt in their contributions to this volume, the latter two of whom criticize the implicit approach as well.

10

I. Incomplete Descriptions

how these approaches work, consider again an assertive utterance of (8). If we adopt the implicit approach, we might imagine that the contextually constrained domain of quantiWcation includes all and only those objects within the shared perceptual Weld of the speaker and hearer. If the utterance is felicitous, there will be exactly one table in this domain. Alternatively, if we adopt the ‘explicit’ approach, the proper interpretation of the description ‘the table’ might be something like ‘the table over there’, where the latter description is uniquely denoting (in cases where the utterance is felicitous). The identity and individuation of the proper ‘completion’ would depend upon various details of the context of utterance, details accessible to the hearer and exploited by the speaker in his communicative endeavors. Despite the superWcial diVerences between these two approaches, the underlying strategy is ‘semantic’ in so far as it aims to show how seemingly incomplete descriptions are in fact complete and thus not at odds with the predictions of Russell’s Theory of Descriptions. The pragmatic approach, in contrast to the semantic approach, denies that incomplete descriptions are completed by the context of utterance. They are precisely what they appear to be: incomplete descriptions, descriptions that fail to denote uniquely. The Russellian pragmatist thus concedes that utterances of sentences containing such expressions are in fact false on account of a failed uniqueness condition.9 Intuitions to the contrary are explained by claiming that, although what is strictly speaking said is false, the speaker might nevertheless manage to communicate something true, something to the eVect that some particular F is G. Thus, while what is literally said by an assertive utterance of (8) would be false, given the plurality of tables, such an utterance might well communicate something true—something like the table over there is covered with books. The misleading intuitions result when we misidentify what is merely communicated—something ‘pragmatic’—with what is actually said—something ‘semantic’. In their contributions to this volume, Recanati and Lepore agree that neither the explicit version of the semantic approach nor the pragmatic approach is plausible. Yet they take opposing positions on how to handle incomplete descriptions. Recanati oVers a novel version of the semantic approach to incompleteness, a variant of the ‘implicit’ approach. Lepore rejects all versions of both the semantic and pragmatic approaches to incomplete descriptions. Neale proposes a variant of the ‘explicit’ approach, which he develops within a framework of what he calls ‘linguistic pragmatics’. Let us begin with Recanati’s account of incomplete descriptions. Recanati’s account is embedded in a more general theory, according to which nominal predicates are evaluated with respect to ‘situations’, and are associated with a variable standing for the situation of evaluation.10 The ‘situation variable’ may be assigned a value contextually, but it may also be bound by a higher quantiWer. In this framework, domain restriction 9 See Bach (1987/1994) for a sentence-based version of this approach. 10 See Recanati (1986, 1996) for earlier versions of this view.

I. Incomplete Descriptions

11

proceeds through the assignment of values to the situation variable. The descriptive condition corresponding to the predicate ‘F ’ in a deWnite description ‘the F ’ becomes uniquely identifying when that predicate is evaluated with respect to the situation that is the value of the associated variable. The uniqueness constraint which the determiner ‘the’ imposes on the predicate ‘F ’ is what guides the assignment of a value to the situation variable associated with it: ‘the’ in ‘the F ’ tells us that the predicate ‘F ’ must denote a singleton-set in the relevant situation, and the situation variable associated with ‘F ’ is assigned a value accordingly. To get a sense of how variables are assigned values in Recanati’s theory, consider the following Wve sentences: (8) (9) (10) (11) (12)

The table is covered with books. The murderer is insane. I met a child and a woman in Lyon the other day. The woman gave me her newspaper. Whenever I go to Lyon by train, the controller is a woman. The President lives in the White House.

(All of the foregoing are discussed in Recanati’s contribution.) Let us consider these in turn, beginning with (8). Suppose that (8) is uttered by the speaker while conversing with the hearer about the furniture in the room where the conversation is taking place. In such a case, the relevant situation—the ‘focus’ situation—would be demonstratively given, as it is perceptually accessible to both the speaker and the hearer. (9) is a bit more complex. Suppose the sentence is uttered upon the discovery of a mutilated corpse, where the speaker’s intention is to make an assertion about whoever it was who murdered the individual in question. Here, a situation s1 is demonstratively given: the speaker and hearer are perceptually confronted with a situation in which the crime victim is dead and mutilated. The focus situation is not, however, s1 : for the murderer is not to be found in s1 . The focus situation is instead an extension of s1 incorporating the event s2 , the murder, that brought about s1 . In both (8) and (9), there is implicit reference to the focus situation, but such reference can also be explicit as in an utterance of (10). Here, the relevant situation is given in the preceding sentence: an event that occurred some time earlier, the event of the speaker’s meeting a child and a woman. (11) and (12) illustrate the idea that situations can be quantiWed over, either explicitly or implicitly. In (11) the quantiWcation is explicit. Here, the situation with respect to which the description ‘the controller’ is to be evaluated is not singled out uniquely. Instead, the speaker universally quantiWes over situations in which she goes to Lyon by train. Each particular assignment of value to the quantiWed variable (each train trip to Lyon) provides a domain of objects in which there is a unique controller, but none is singled out in particular. Now let us consider (12). On one reading of (12), there is an implicit universal quantiWcation over a certain sort of situation that might be called ‘US situations’. In all such situations s, whoever is president in s, lives in the White House. If, instead, the speaker refers to the actual US situation in which Bush is president, (12) gets another reading, yet it need not be a ‘referential’ one about Bush

12

I. Incomplete Descriptions

himself.11 We may implicitly refer to the current situation and say of whoever is the president in that situation that he lives in the White House. The motivation behind Recanati’s account of incomplete descriptions appears largely methodological. First, it is embedded in a more general account of nominal predication, and so accommodates other phrases standardly treated as quantiWer phrases. Second, it allows for a uniform characterization of a variety of distinctions associated with deWnite descriptions. In particular, Recanati argues that his account of nominal predication can unify the distinctions between: complete and incomplete descriptions, ordinary and anaphoric descriptions, referentially and attributively used descriptions, and descriptions involving implicit and explicit reference to, or quantiWcation over, situations of evaluation. Lepore’s Russellian approach to incomplete descriptions is a radical one. Lepore considers and rejects both semantic and pragmatic solutions to the problem of incomplete descriptions. SpeciWcally, he charges that semantic strategies suVer from an ‘over generation’ problem: that is, all such strategies permit the use of sentences containing incomplete descriptions to express propositions that they cannot reasonably be held to express. Thus, consider again an utterance of (8), (8)

The table is covered with books.

Suppose that the proposed analysis is: (13)

The table there is covered with books.

Then, consider (14): (14)

The table is there.

Intuitively, no utterance of (14) expresses a necessary truth. But it is diYcult to see how this is to be prevented on the sort of approach in question. For on the semantic approach to incomplete descriptions, what is to prevent an utterance of (14) from being analyzed as in (15)? (15)

The table there is there.

In addition, Lepore argues that semantic strategies saddle deWnite descriptions with more ambiguity than they actually exhibit. To see Lepore’s point, consider an incomplete description like ‘the table’ and consider the view that the felicitous use of such an expression will yield a completed description: one containing a contextually speciWed object(s)/individual(s) referred to by way of a hidden indexical expression. On such a view, the relation of the hidden indexical to the object selected will vary as distinct relational nouns or adjuncts are invoked, contingent upon the context of utterance. To see this, consider an utterance of (8), which might (on the view in question) express any of (16), (17), or (18): 11 See the introduction to Part II for details on the ‘referential’ use of descriptions.

I. Incomplete Descriptions (16) (17) (18)

13

The table there is covered with books. His table is covered with books. That table is covered with books.

Suppose that ‘the table there’ is correctly interpreted as the table over there; suppose that ‘his table’ is correctly interpreted as the table of him; and suppose that ‘that table’ is correctly interpreted as the table identical to that. Since in any context, any of these (or untold other) sorts of completions may be operative, it follows that ‘the table’ (or ‘table’) is multiply ambiguous. For the semantics of the expression must be able to accommodate the fact that there are any number of ways that the predicate ‘table’ might combine with the hidden indexical expression (via ‘over’, ‘of ’, ‘identical to’ etc.) so as to yield a complete description. Positing so much ambiguity would appear costly, given the apparent lack of motivation for such positing. The moral of this, and the preceding ‘over generation’ objection, is signiWcant, according to Lepore. It is to practise temperance with respect to how much work we allot to context in semantics. Lepore then rejects the pragmatic approach (together with the semantic approach) on the grounds that it is based on a misguided notion of ‘what is said’, according to which ‘what is said’ is interchangeable with the genuinely semantic notion of the ‘proposition expressed’. On this wrong-headed view, what is said by a single utterance is a single proposition. On this picture, a speaker says that p just in case he assertively utters a sentence that expresses the proposition that p. But this is simply false: what a speaker says is determined by an enormous variety of non-semantic considerations. In particular, Lepore claims that, (a) The truth conditions of a sentence need not correspond to what is said by an utterance of that sentence. (b) What is said by the utterance of a sentence can be true even though the truth conditions of the sentence are not satisWed (and vice versa). (c) What is said by the utterance of a sentence can be about something even though the truth conditions of the sentence make no mention of that thing. (d ) Because of (a)–(c), intuitions about utterances of sentences can in no simple and direct way be used as guides to the truth conditions for these sentences. Lepore contends that eliminating this misguided picture of ‘what is said’, a picture that underpins both the semantic and pragmatic approaches to incomplete descriptions, completely dissolves the so-called ‘problem’ of incomplete descriptions. Neale does not share Lepore’s pessimistic attitude toward the theoretical importance of the notion of ‘what is said’. On the contrary, Neale develops and defends a notion of ‘what is said’, locating it within an approach he refers to as ‘linguistic pragmatics’. This view revolves around two central ideas. First, what a speaker says in uttering X on a given occasion is under-determined by the information encoded in X itself; and second,

14

I. Incomplete Descriptions

identifying what a speaker is saying by uttering X on a given occasion involves pragmatic inference as well as linguistic decoding. He then goes on to employ that framework in defending a Russellian account of incomplete descriptions used referentially. Neale prefaces his analysis by pointing out that no problem of incompleteness arises for a Go¨delian description (ix)(Fx : x ¼ a), a description of the sort used in explicating Go¨del’s slingshot argument. (See Neale 2001, for details.) According to Neale, this suggests a foolproof way of interpreting utterances of incomplete descriptions used referentially. The point is illustrated with an example from Schiffer (1995). Ferdinand Pergola is known to us to uniquely satisfy the matrices of a variety of distinct descriptions (‘the author of Smells and Tickles’, ‘the man staggering up to the podium in front of us’, etc.). Seeing Pergola trip, I say to you ‘the guy is drunk’ and you understand what I have said. According to Schiffer, no single uniquely satisﬁed matrix, however complex, is salient enough to underlie both my intention and your understanding. Neale suggests that we appeal to a simple, rather than complex, matrix. In particular, you understand me as saying something whose truth conditions are captured by (19), where a is being used to refer to Pergola: (19)

[thex : guy(x) : x ¼ a] x is drunk

The basic idea, says Neale, is both simple and intuitive: completion of an incomplete description used referentially is effected non-descriptively. Neale views his analysis of the referential use of incomplete descriptions as allowing for a rapprochement between those who have argued for a semantically signiﬁcant referential use and those who have argued for a unitary Russellian account. Whenever a description is used referentially, it is being used in such a way that there is an obvious Go¨delian completion. Do we then say that it is part of the meaning or semantics of ‘the’ that it suggests such a completion? Or should we view such completions simply as abstractions over regular interpretations? This question, Neale contends, will be answerable only once we have a better understanding of the cognitive mechanisms that underlie utterance interpretation. However, whatever the nature of such mechanisms turns out to be, the philosophical point will remain unscathed: descriptions are Russellian, and the phenomenon of referential use is but a special case of the phenomenon of incompleteness.

1 Descriptions and Situations Franc¸ois Recanati

1. Referentiality and Incompleteness Genuine singular terms, such as proper names or (presumably) demonstrative pronouns, are referential; by which I mean that their conventional use is to refer. For every such term t there is a linguistic convention in virtue of which, whenever t is used, there must be a unique object x such that the speaker uses t to refer to x. The modal ‘must’ here indicates that the condition is a felicity condition: a condition which must be satisWed for the use of the expression to be felicitous. Note that, in stating the convention, I have taken the notion of ‘reference’ as primitive. The notion of reference at stake here is the pragmatic notion: speaker’s reference, as it is sometimes called. To refer in that sense (the pragmatic sense) is, very roughly, to draw the hearer’s attention upon some object for the purpose of predicating something of that object, and to do so in the overt manner characteristic of human communication. The linguistic convention governing the use of a referential term does not merely specify that there must be a unique object referred to by the speaker (pragmatic condition); it also speciWes that the object in question must possess whatever properties the referential expression encodes as part of its linguistic meaning (descriptive condition). For example, the referent of ‘I’ must be the speaker; the referent of ‘she’ must be a female person; etc. In the case of proper names, I have argued that the relevant descriptive condition is metalinguistic: it involves the property of being called by that name (in the contextually relevant community).1 But the proper treatment of names is a controverted matter which I will keep away from in this chapter. Various distinctions can be made with respect to the descriptive condition encoded by a referential term. In the case of ‘I’ the descriptive condition happens to be uniquely identifying (insofar as an utterance typically has a unique utterer); but the descriptive condition associated with the third person pronoun ‘she’ is satisWed by every female I am indebted to Anne Bezuidenhout (and her colleagues), Herman Cappelen, Eros Corazza, Stephen Neale, Philippe Schlenker, and Isidora Stojanovic for helpful comments or discussions. 1 See Recanati (1993: 135–67).

16

I. Incomplete Descriptions

person, hence it is far from uniquely identifying. Another interesting diVerence between the descriptive conditions respectively associated with ‘I’ and ‘she’ is this. In the case of ‘I’ the descriptive condition is token-reXexive. The referent is the utterer of this token. In the case of ‘she’ the condition is not token-reXexive: it is the general condition that the referent be a female person.2 Note that, contrary to what the ‘I’/‘she’ contrast suggests, the descriptive condition may be token-reXexive without being uniquely identifying. In the plural, the Wrst person pronoun (‘we’) arguably refers to a group which must contain the speaker among its members. That descriptive condition is tokenreXexive—the speaker is the person who utters the relevant token of ‘we’—but it is no more uniquely identifying than the descriptive condition associated with ‘she’. There is an indeWnite number of groups containing the speaker among their members. Whether or not the associated descriptive condition is uniquely identifying, a referential expression denotes, or semantically refers to, a single object in a given context. The denotation or semantic referent is that object which, in the context at hand, satisWes the complex pragmatic-cum-descriptive condition. For example, the referent of a use of ‘she’ is that female person whom the speaker refers to in uttering that word. If no object, or more than one object, satisWes the complex pragmatic-cum-descriptive condition, the expression fails to denote anything and carries no semantic value, in that context. From what I have said it follows that the semantic referent is parasitic on the speaker’s referent, in the sense that the expression does not refer unless the speaker herself refers to some object. Still we make room for a distinction between speaker’s reference and semantic reference. The semantic referent must satisfy the descriptive condition associated with the expression by the rules of the language. If the speaker’s referent does not satisfy the descriptive condition, it will not acquire the status of semantic referent. That is not to say that something else will be the semantic referent, in such circumstances. That is ruled out in virtue of the pragmatic condition. In a context in which the speaker mistakenly says ‘she’ while pointing to a man, the speaker’s referent (the man pointed to) does not acquire the status of semantic referent, but even if there is a unique female around, she does not become the semantic referent either. Something cannot become the semantic referent unless it satisWes the pragmatic condition, that is, unless it is also the speaker’s referent. Two further properties of referential terms must be mentioned, one semantic and the other epistemic. First, referential terms are ‘directly referential’. The semantic referent is the referential term’s semantic value—what it contributes to the truth conditions of the utterance in which it occurs. The complex pragmatic-cum-descriptive condition associated with the referential term is not itself among the truth conditions 2 Many theorists hold that the condition associated with demonstrative pronouns is token-reXexive: thus a token of ‘she’ is said to denote the female the speaker is pointing at while uttering that token, or the most relevant female in the context of utterance of the token. As far as I am concerned, I admit that the complex pragmaticcum-descriptive condition associated with demonstrative pronouns is both uniquely identifying and tokenreXexive (see below), but I deny that the descriptive condition per se has either of these properties.

Recanati, Descriptions and Situations

17

of the utterance, but among its felicity conditions: it constrains the context in which the expression can felicitously occur. A referential expression is felicitously used only if there is one and only one object x such that (i) x is F (descriptive condition) and (ii) x is referred to by the speaker (pragmatic condition). Only in a context meeting those conditions will the referential expression contribute something to the content of the utterance in which it occurs. What it contributes, as we have seen, is its semantic referent: the object which contextually satisWes the complex pragmatic-cumdescriptive condition. Hence the content of an utterance G(t) in which t is a referential term is a singular proposition; a proposition true iV the semantic referent of t satisWes G( ), and false otherwise. Second, since the content of a referential utterance G(t) is singular, to understand the utterance—to grasp its content—one must entertain a de re thought involving the referent of t. A thought to the eVect that whatever contextually satisWes the complex condition is G will not do: for the content of such a thought is irreducibly general. Proper understanding requires being en rapport with the object, being acquainted with it so as to be able to entertain de re thoughts about it. This means that the interpreter must possess a dossier of information concerning the object, a dossier which the use of the referential term will evoke and activate, thus making ‘re-identiWcation’ possible. Re-identiWcation occurs, Strawson tells us, if and only if the singular term used establishes for the hearer an identity, and the right identity, between the thought of what-is-being-spoken-of-by-the-speaker and the thought of some object already within the reach of the hearer’s own knowledge, experience, or perception, some object, that is, which the hearer could, in one way or another, pick out or identify for himself, from his own resources. (Strawson 1961: 402)

Let us now turn to deWnite descriptions. They can be treated as referential terms, as Strawson proposed (Strawson 1950, 1952). On that treatment, the linguistic material following the deWnite article (e.g. the phrase ‘man with a brown hat’ in the deWnite description ‘the man with a brown hat’) encodes the descriptive condition which the referent has to satisfy; but there is no implication that one and only one object satisWes that condition. Uniqueness implication there is, but it pertains to the pragmatic condition: what is implied by the use of the deWnite article is that there is a unique object such that the speaker is referring to that object. The descriptive condition encoded by the nominal expression following the deWnite article is meant to help the hearer identify the relevant object. The object in question—the speaker’s referent—becomes the description’s semantic referent (or denotatum, as I will henceforth call it) just in case it actually satisWes the descriptive condition. There are two basic arguments in support of a Strawsonian, referentialist account of deWnite descriptions: the similarity-with-pronouns argument, and the incompleteness argument. The similarity-with-pronouns argument goes like this. Third-person pronouns, when deictically used, are commonly treated as referential terms. The speaker, looking at a

18

I. Incomplete Descriptions

woman, says ‘She is vulgar’, thereby referring to the woman. The direction of the speaker’s gaze and the descriptive condition associated with the pronoun ‘she’ enable the hearer to Wgure out that the speaker is referring to that woman whom he also can perceive. All that is straightforward. Now exactly the same story can be told if the speaker, instead of using the third person pronoun, uses a deWnite description and says: ‘The woman is vulgar’. As Stephen SchiVer puts it, ‘the two cases have exact psychological parity’ (SchiVer 1997: 263). An opponent of the referential account can argue that ‘the woman’ is a special case, namely, an incomplete deWnite description—a type of description which, precisely because it is incomplete, requires completion by means of an act of demonstrative reference. In such cases, the descriptive condition is not uniquely identifying and we need to rely on the context to Wx the denotatum. In normal cases, however, the descriptive condition is identifying by itself and we do not have to rely on pragmatic factors. There, for the anti-referentialist, lies the diVerence between descriptions and pronouns: it is in the nature of the (demonstrative) pronoun ‘she’ that we need an act of reference on the speaker’s part to Wx the denotatum. Not so with deWnite descriptions, except in the special cases in which they are ‘incomplete’. Insofar as they are special, such cases should be left aside in theorizing about descriptions. There are two distinguishable elements in this anti-referentialist response to the similarity-with-pronouns argument. First, there is the implicit observation that only certain uses of deWnite descriptions are intuitively ‘referential’ (i.e. involve an act of reference on the speaker’s part and an act of re-identiWcation on the hearer’s part). I will deal with that observation below. It raises a serious problem for the referentialist, though perhaps not an insuperable one. The other aspect of the anti-referentialist reply is the idea that incomplete deWnite descriptions are too ‘special’ to be considered in theorizing about descriptions. That part of the anti-referentialist reply is clearly not acceptable. There is nothing abnormal about incomplete descriptions; indeed most descriptions are incomplete. Just as the descriptive condition associated with a referential term may or may not be uniquely identifying (recall the diVerence between ‘I’ and ‘she’), the descriptive condition associated with a deWnite description may or may not be uniquely identifying. It would be question-begging to argue that incomplete descriptions (i.e. those descriptions whose descriptive condition is not uniquely identifying) are somehow deviant, and that only complete descriptions matter to semantic theory. The main argument in favor of a referential treatment of (certain uses of ) deWnite descriptions is precisely the availability of an account of incomplete deWnite descriptions within that framework (Kripke 1977; Wettstein 1981). That is the incompleteness argument. If deWnite descriptions are referential, it does not matter that sometimes the descriptive condition per se is not uniquely identifying; for we can rely on the speaker’s act of reference to provide a unique candidate for the status of semantic referent. To say that a description is referential is to say that its denotation is Wxed by the pragmaticcum-descriptive condition, rather than by the descriptive condition alone. On the

Recanati, Descriptions and Situations

19

other hand, if we do not treat the description as referential, we are left in a quandary. An incomplete description is such that the descriptive condition, by itself, is not uniquely identifying. How, then, can its denotation be determined? That is the problem which incomplete descriptions raise for nonreferentialist accounts. Various solutions to this problem will be discussed below (section 2). Before evaluating these two arguments in favor of the referential analysis, let us brieXy consider the main argument against it—an argument I have already alluded to. We may call it the ‘lack-of-generality argument’: Lack-of-generality argument The referential analysis imposes an act of reference to the denotatum on the speaker’s part, and a correlative act of re-identiWcation of the denotatum on the hearer’s part. Yet there are many cases in which a description is used but no such acts take place (nor are required). As Donnellan pointed out, descriptions may be used attributively, in such a way that a general rather than a singular proposition is expressed. Such nonreferential uses of deWnite descriptions are clear counter-examples to the referential analysis. Is this argument compelling? That is unclear. After all, pronouns themselves have distinct uses: they can be used referentially (deictically), but they also have bound uses and anaphoric uses. Why not, then, accept for descriptions what we accept for pronouns, namely, that they have a deictic use, distinct from their other uses? If we are prepared to endorse an ambiguity thesis regarding pronouns, we should be prepared to endorse it also regarding descriptions. Note that Strawson himself acknowledged the existence of nonreferential uses of descriptions (as in ‘The whale is a mammal’); but he did not feel compelled to give up the referential treatment, because he questioned the need for a genuinely uniWed semantic analysis. Still, I think it should be conceded to the anti-referentialist that a uniWed analysis of descriptions is preferable, ceteris paribus, to a non-uniWed analysis. The same thing holds for pronouns, of course. On the other hand, the referentialist should be granted his Wrst point: An analysis which does justice to the striking similarities between pronouns and descriptions is preferable, ceteris paribus, to an analysis which treats them as illusory.3 The problem is that these two desiderata may (and actually do) enter into conXict. Many philosophers hold that a uniWed, nonreferential analysis of descriptions is available, typically an analysis along Russellian lines completed by Gricean considerations. In virtue of the Wrst desideratum, such an analysis should be preferred to the sort of view defended by either Strawson or Donnellan. But it does not satisfy the second desideratum: the referentiality of descriptions is treated as a pragmatic illusion, 3 This applies to the bound uses of descriptions, as much as to their referential uses. Just as the referential use of a description resembles the deictic use of a pronoun, the ‘bound’ uses of descriptions resemble the bound uses of pronouns. A theory which takes these similarities at face value is preferable, ceteris paribus, to a theory which does not. As George Wilson says (1984: 11), ‘the uses of pronouns and descriptions need to be understood together’.

20

I. Incomplete Descriptions

while the referentiality of pronouns is treated as a genuine semantic fact. As SchiVer pointed out, this diVerence of treatment is worrisome, given the similarity between the cases. Let us now evaluate the two referentialist arguments. I have just said that I grant the referentialist his Wrst point: deictic pronouns and referential descriptions should be treated more or less on a par. This argument cuts both ways, however. The analogy between pronouns and descriptions can be interpreted as supporting a descriptive treatment of pronouns, as much as a referential treatment of descriptions. For example, we might equate the pronoun ‘she’ with an incomplete description such as: ‘the female’, and analyze that description within the uniWed, Russellian-cum-Gricean framework (or whichever uniWed framework is used for the analysis of deWnite descriptions). This immediately leads us back to the main argument in favor of a referential analysis of descriptions: the argument from incompleteness. How are we to treat an incomplete description such as ‘the female’ if we do not accept the referential analysis? Everything turns out to hinge upon that argument. If we analyze the argument, however, we see that it involves two distinct claims, one of which can be disputed: First claim Incompleteness raises a deep problem for descriptive (nonreferential) approaches. Second claim The problem of incompleteness does not arise in a referentialist framework. The Wrst claim seems to me undoubtedly true, but I reject the second one as false. The problem of incompleteness still arises within a referentialist framework (Neale 1990: 98–100). It follows that we cannot legitimately use the incompleteness problem as an argument in favor of the referential analysis. There are two reasons why the problem of incompleteness still arises in a referentialist framework. First, among incomplete deWnite descriptions, some are clearly attributive, as when the speaker discovers Smith’s horribly mutilated body and exclaims: ‘The murderer is insane!’ Faced with this apparent counter-example to the referentialist approach, Wettstein maintains that, in this as in the other cases, the incomplete description is completed by an act of demonstrative reference (to the corpse). That is an interesting claim yet, appearances notwithstanding, it is irrelevant to the issue at stake. Even if Wettstein is right and an incomplete description is always completed through an act of demonstrative reference, that need not be an act of reference to the description’s referent. In the murderer example, the speaker refers to Smith’s mutilated body. What makes that occurrence of the description attributive is precisely the fact that the speaker does not refer to but merely describes the denotatum. The fact that the speaker refers to something else is irrelevant. (The sort of position which Wettstein’s observation actually supports I call ‘quasi-referentialism’. According to quasireferentialism, what enables us to complete a deWnite description is an act of

Recanati, Descriptions and Situations

21

demonstrative reference, though not necessarily an act of reference to the denotatum. A particular version of quasi-referentialism, namely situational referentialism, will be discussed in section 3.) The second reason why referentialism does not solve the incompleteness problem is that, as we all know, the phenomenon of incompleteness concerns not only descriptions, but also quantiWers. When I say ‘Everybody failed’, the quantiWer is obviously incomplete. (I do not mean that everybody in the world failed.) The referentialist view does not apply in a case like this. The quantiWer must be completed, and that may involve reference to, for example, a particular group of students, such that the speaker means that everybody in that group failed; but the speaker does not refer to the actual persons who failed. I conclude that the two arguments in favor of referentialism—the similarity-withpronouns argument and the incompleteness argument—actually fail to support that position. What we need is something which referentialism does not provide, namely, an account meeting the following three desiderata: (i) the account must be semantically uniWed, (ii) it must capture the striking similarities between pronouns and descriptions, and (iii) it must solve the incompleteness problem. In this chapter I am mainly concerned with (iii), but I will keep an eye on (i) and (ii).

2. Incompleteness and Indeterminacy If we give up the referentialist analysis, how can we account for description incompleteness? In a referentialist framework uniqueness pertains to the act of reference performed by the speaker. The semantic referent is not the unique F simpliciter, but the unique F referred to by the speaker. In a non-referentialist framework, we must say that the descriptive condition F , by itself, determines the semantic referent. The semantic referent will be the unique satisWer of that condition, as in Frege’s theory. The problem is that, whenever the description is incomplete, there is no unique satisWer (by deWnition). Three main solutions to this problem have been put forward in the literature: (i) We can bite the bullet and treat all incomplete descriptions as defective (nondenoting). (ii) We can treat incomplete descriptions as elliptical for the appropriately completed descriptions. For example, ‘the table’ will be understood as: the table of the living room, the table which Aunt Martha gave us, or possibly the table there. (iii) We can insist that the descriptive condition associated with incomplete deWnite descriptions is uniquely identifying. I said above that it is non-uniquely identifying ‘by deWnition’. Still it can be argued that it becomes uniquely identifying if we appropriately restrict the domain of discourse. What is true by deWnition is only this: the descriptive condition associated with incomplete deWnite

22

I. Incomplete Descriptions descriptions is non-uniquely identifying in the total world. But it may well be uniquely identifying in a partial situation.

If we choose the Wrst option, we face a problem: we must account for the clear intuition that utterances such as ‘The table is covered with books’ can be true. The problem can be put in general terms. By treating incomplete descriptions as nondenoting, we introduce a gap between the deliverances of semantic theory and ordinary truth-conditional intuitions; a gap which it is then incumbent upon pragmatic theory to reduce. In previous writings I argued that the gap in question should be kept as narrow as possible. That means that, other things being equal, we should prefer semantic theories which minimize the gap over theories which widen it. It follows that the option I have just mentioned should be seriously considered only if every other reasonable attempt fails. (It will not be considered in the rest of this chapter.) Option (iii) I Wnd most promising. The descriptive condition associated with an incomplete description such as ‘the woman’ becomes uniquely identifying if we restrict the domain, for example, if we focus on the perceptual scene. That is very similar to the phenomenon of demonstrative reference: we look at the scene before us, and are thereby related to various perceived individuals—say, a man and a woman—whose behavior we monitor through our continuous perception of the scene. The nonuniquely identifying descriptive condition associated with the description ‘the woman’ suYces, within the domain of objects thus singled out, to make the hearer understand which object is in question. Domain restriction, in such a case, proceeds through something like demonstrative reference to a global situation. Option (iii) is therefore compatible with a quasi-referentialist approach, according to which incomplete descriptions (or incomplete quantiWers) are completed via an act of reference on the speaker’s part (and a correlative act of identiWcation on the hearer’s part). On the particular version I call situational referentialism, the entity tacitly referred to is a partial situation with respect to which the descriptive condition F happens to be uniquely identifying (section 3). Option (ii) is probably the most popular and it too is compatible with a quasireferentialist approach. In the rest of this section I will discuss that option and try to make clear why I think option (iii) is, on the whole, preferable. Option (ii) can be interpreted in several ways. The ‘ellipsis’ at issue can be construed either as syntactic ellipsis in the strict, grammatical sense or as ellipsis only in a loose, semantic sense. In previous papers (Recanati 1989a, 1996) I argued against the ellipsis approach construed in the strict, grammatical sense. I have the feeling that most theorists nowadays consider such a theory as preposterous and hardly worthy of a serious reply.4 Let us therefore ignore it and consider only a semantic, non-syntactic construal of the ellipsis theory. On such a construal, the sentence ‘The table is covered 4 See the collection of papers in Mind and Language 15/2–3 (2000). ‘The syntactic ellipsis approach . . . strikes me as a nonstarter’ (Bach 2000: 267 n.); ‘I Wnd it hard to believe anyone has ever proposed such a clumsy and bizarre view’ (Neale 2000: 292).

Recanati, Descriptions and Situations

23

with books’ is not syntactically elliptical for another, more complete sentence, but the statement which the speaker makes by uttering this sentence involves a contextually provided property of the table, which property enriches the matrix of the description (analyzed a` la Russell–Neale): ‘the x: Fx’ is contextually enriched into ‘the x: Fx & Hx’.5 A well-known problem with the ellipsis theory is that it is indeterminate which completion the speaker has in mind. That problem arises at several levels. To see that let us distinguish three things: • o, the reference or denotatum (an object); • P, some uniquely identifying property of o through which it is singled out, and which provides the (contextual) sense of the description; • a, some linguistic expression explicitly expressing P. If one opts for a syntactic version of the ellipsis theory, it will be indeterminate which among many diVerent, possibly synonymous sentences the elliptical sentence is elliptical for. Even if we Wx the property P which is used to uniquely identify o, still it is indeterminate which expression a expressing P the incomplete description is elliptical for. Is ‘the table’ elliptical for ‘the table which Aunt Martha bequeathed us’ or ‘the table we inherited from Martha’? By opting for a semantic version, we suppress that source of indeterminacy: the same property P is involved in both cases, and that is all that matters. But there remains a major source of indeterminacy: a great number of distinct, uniquely identifying properties P1 , P2 , P3 , etc. can be invoked to render the description complete. Which one is the right one? As Wettstein argued, there is no principled answer to that question, hence the ellipsis theory cannot be right: When one says, for example, ‘The table is covered with books,’ the table the speaker has in mind can be more fully described in any number of ways, by the use of any number of nonsynonymous, uniquely denoting descriptions (for example, ‘the table in room 209 of Camden Hall at t1 ’, ‘the table at which the author of The Persistence of Objects is sitting at t1 ’, etc.) . . . It might be supposed that we could decide on one of these Russellian descriptions as the correct one by reference to the intentions of the speaker. In many cases, however, the speaker will have no such determinate intention. (Wettstein 1981: 41–2)

In defense of the ellipsis theory, several philosophers have argued as follows.6 Let us use a referential completer instead of a descriptive completer. For example, in the 5 There are two versions of the theory, depending on the exact status of the contextually provided property H. According to one account the contextually provided property is an unarticulated constituent of the statement made by uttering the sentence, just as the location of rain is an unarticulated constituent of the statement made by saying ‘It’s raining’ (Perry 1986). Thus SchiVer says that a possible meaning rule for ‘The F is G’ is: ‘Utter ‘‘the F is G’’ only if there is a property H such that you mean that the F and H is G’ (1997: 256). Other theorists deny that there are unarticulated constituents and think such constituents are always the value contextually assigned to a free variable in logical form (Stanley 2000). If one takes this line, one will say that every description ‘the F ’ comes with a free variable for the extra property H, which variable is assigned a deWnite value in context. 6 See Soames (1986: 352); Neale (1990: 101).

24

I. Incomplete Descriptions

murderer case, let us follow Wettstein and say that the description is completed by a demonstratively given individual (the victim). The description ‘the murderer’ encodes a two-place relation, ‘murderer-of ’, the second argument-place of which is Wlled by the demonstrated individual, namely Smith. That individual is not described, he is demonstratively given; hence the problem of the plurality of non-synonymous, co-denoting descriptions does not arise. The trick consists in moving from the level of sense to the level of reference so as to bypass the problem of the multiplicity of potential senses. When it comes to other, nonrelational examples (‘the woman’, ‘the table’), the same sort of solution can be appealed to—or so it is argued. All the completions mentioned by Wettstein in connection with the incomplete description ‘the table’ (e.g. ‘in room 209 of Camden Hall at t1 ’, ‘at which the author of The Persistence of Objects is sitting at t1 ’) are descriptive, but we can also think of demonstrative completions like: ‘over there’. Thus ‘The x: x is a table’ can be enriched into: ‘The x: x is a table and x is there’, or into ‘The x: x is a table and x ¼ that’. Since demonstratives are directly referential, what completes the content of the description in such cases is a place in egocentric space or an object (the table itself ). In this way, it is claimed, we avoid the indeterminacy problem. The completion takes place at the level of reference, not at the level of sense; hence the problem of the plurality of potential senses does not arise. This alleged solution to the indeterminacy problem rests on a confusion, due to an ambiguity in the very notion of a ‘referential completer’. A referential completer is meant to be a worldly entity—for example, an object or a place. That is required for the argument to go through. Such completing entities are indeed involved in what I called demonstrative completions; but they do not do the completing by themselves. If by ‘referential completer’ we mean what really does the completing, then a referential completer is not an object, but an object-dependent property. Once we see that, however, we realize that the indeterminacy problem has not been solved. Except perhaps in special cases where there arguably is an ‘implicit argument’ (as in ‘the murderer’), we need a relation to bridge the gap between the incomplete description and the completing entity. In the table example, we need the located_at relation if the completing entity is a place, and we need the identity relation if it is the table itself. Like the completing entity, the bridging relation is contextually provided. Together with the completing entity, it determines a (demonstrative) property of the denotatum which is what putatively completes the incomplete description: the property of being identical with that (where that ¼ the table in question) or the property of being located there (where there ¼ the location of the table in question). Now there is absolutely no reason to think that we can get rid of indeterminacy by using such completers. For it is still indeterminate which completing demonstrative property H is contextually provided: the property of being there, the property of being this object, the property of being in front of me, or whatnot. I conclude that the appeal to demonstrative completions does not meet the indeterminacy objection which besets the ellipsis approach. We have not really moved from the

Recanati, Descriptions and Situations

25

level of sense to the level of reference. Demonstrative, object-involving properties such as those mentioned in connection with the table example determine possible contextual senses for the incomplete description, and there are many potential senses of that sort, just as there are many potential descriptive senses. Another solution to the indeterminacy problem has been put forward in the literature. It consists in biting the bullet and accepting the indeterminate nature of the contextual completion (W. Blackburn 1988; SchiVer 1997). It is a fact that, in a normal situation of utterance, there will often be a collection of potential and equally legitimate completers for a given incomplete description. Insofar as the speaker does not have one of them in mind to the exclusion of the others, the proposition expressed by the utterance, ‘The F is G’, will be somewhat indeterminate; but this does not prevent us from ascribing deWnite truth conditions to it, using supervaluation techniques. We can say that the utterance is true iV all the potential completions are true, false iV all the potential completions are false, and unevaluable otherwise. For that solution to work we must be given a set of potential completers. At this point, an advocate of the quasi-referential account may argue as follows. Such a set, hence the truth conditions of the utterance, can be established only as a side eVect of an act of reference. Typically, the potential completers will correspond to salient properties of the denotatum in the context at hand. The properties in question are made available through the speaker’s act of demonstrative reference to the denotatum, hence the referentialist viewpoint is vindicated: what completes the description and enables the hearer to overcome the non-uniquely identifying character of the descriptive condition is the speaker’s act of reference which gives us access to the denotatum and its properties, hence to several potential completers.7 If the description is not used referentially, still, as Wettstein pointed out, something is referred to, which determines a set of potential completers. In the murderer example, what is demonstratively given is a certain object—Smith’s body—in a certain state, which state presumably results from someone’s action. The action itself has several facets and can be described both as a killing and as a mutilating. The object, the state, and its presumed etiology are among the aspects of a complex, holistic situation which determines the set of potential completers for the incomplete description ‘the murderer’. The murderer is the person who did the killing; but he is also, presumably, the person who did the mutilating. The description can be understood not only as ‘the person who murdered Smith’ but also, still more explicitly, as ‘the person who murdered Smith and savagely mutilated his body’, or perhaps as ‘the person who broke into the house, murdered Smith, and savagely mutilated his body’. As soon as we are allowed to enrich the description with contextually provided properties, several, equally legitimate enrichments spring to mind. To meet the indeterminacy objection, we can supplement the ellipsis theory with an appeal to vagueness and supervaluation; but again, this presupposes that a set 7 As Wiggins puts it (in a diVerent context), to know which entity a given expression refers to is ‘a single piece of knowledge which can be given in countless diVerent ways by countless diVerent descriptions’ (1975: 11).

26

I. Incomplete Descriptions

of potential completers is somehow given, and it is given only through some form of reference to a complex, holistic situation. In the next section we shall see that the quasi-referentialist view cannot be maintained. What matters to my present purposes is not the quasi-referentialist view, however, but rather the need to appeal to something in order to account for the provision of a set of completers. The ellipsis-cum-vagueness theory presupposes that a set of completers is somehow given. My claim has been that such a set is given only against a certain background, namely, with respect to a situation in which the denotatum is involved. In the relevant situation the denotatum has properties and stands in relation to other objects. Those features of the denotatum which happen to be instantiated in the situation at issue determine the set of the potential completers for the description. The plurality of potential completers therefore corresponds to the internal complexity and the holistic character of the focus situation. (On this analysis the referential use of a description falls out as a special case: the case where the focus situation is demonstratively given and the denotatum itself is demonstratively given as part of the situation in question. Thus when I see a man and a woman quarreling and I say ‘The woman is vulgar’, the set of completers for the description is determined by the perceived situation. In such a case, there is demonstrative reference both to the deictically given situation and to the denotatum who centrally features in it.) Now the reason we have for preferring option (iii) to option (ii) can be stated quite simply: we need a focus situation to determine the set of completers which the ellipsiscum-vagueness theory requires in order to run its supervaluations. Once available, however, the situation in question can be used directly to complete the description, by providing a restricted domain with respect to which the descriptive condition happens to be uniquely identifying. Hence there is no need to go into the complications of the ellipsis-cum-vagueness theory.8 8 As Herman Cappelen pointed out to me, giving up the ellipsis theory may not be suYcient to get rid of indeterminacy. There is no reason why there should not be indeterminacy also with respect to the focus situation: for a given utterance involving a deWnite description, several distinct completing situations may turn out to be compatible with all relevant aspects of the context. Indeed, according to Stephen Neale, indeterminacy cannot be eliminated, whichever framework one opts for. ‘To the extent that there are aspects of what is said that are not directly traceable to particular semantic features of [the uttered sentence], indeterminacy is going be inevitable—at least if what is meant by ‘‘indeterminacy’’ is that there are competing characterizations of what U said among which no principled choice can be made’ (Neale, 2002). If Cappelen and Neale are right, the case against the ellipsis theory is admittedly weakened. But how much? ‘One consequence of the ubiquity of indeterminacy’, Neale says in the same passage, ‘is that all versions of a type of argument used by Wettstein, Recanati, Reimer, SchiVer, and others against traditional explicit (or ellipsis-based) accounts of what U said by uttering X , where X contains a so-called incomplete description, such as ‘‘the book’’, are discredited’. This goes a bit too far, I think. Even if it is ultimately ineliminable, indeterminacy should not be multiplied without necessity. For that reason the ‘semantic’ version of the ellipsis theory is preferable to the syntactic version, and the semantic version in terms of ‘referential completers’ is preferable to the semantic version in terms of ‘descriptive completers’. For the very same reason, the approach in terms of situations seems to me preferable to the ellipsis theory. Still, I agree with Cappelen and Neale that considerations of indeterminacy cannot be decisive, given the pervasiveness of the phenomenon. Hence I do not take myself to have refuted the ellipsis theory. Rather, I have tried to provide reasons for exploring the alternative approach in terms of situations. A systematic comparison of the two approaches will be possible only after each one has been suYciently elaborated.

Recanati, Descriptions and Situations

27

3. Situational Referentialism From what I have said it follows that deWnite descriptions do not constrain the context (the situation of utterance), as referential terms do; rather, they constrain an arbitrary focus situation which may but need not be the situation of utterance. If I say ‘I am bald’, the denotatum (the person who is said to be bald) must be speaking in the situation of utterance. But if I say ‘the president is bald’, the denotatum need not be president in the situation of utterance. He may be president in a remote situation (e.g. a distant country) which simply happens to be the topic of the current conversation. According to what I call situational referentialism (a particular version of the quasi-referential approach), the denotatum need not be referred to, but at least the situation in which it is to be found (the focus situation) must be referred to. Is that right? Undoubtedly, there are cases in which the relevant situation is demonstratively given. The speaker and the hearer witness a scene in which there is a man and a woman. The situation is perceptually accessible, and the reference of the description ‘the woman’, qua constituent in that situation, is also perceptually accessible. That is characteristic of referential uses of deWnite descriptions. In other cases, a situation is demonstratively given, but we need to extend it beyond what is demonstratively given in order to secure a referent for the description. Thus, in the murderer example, we need to go beyond the demonstratively given situation s1 (a situation in which Smith lies dead and horribly mutilated) and extend it by considering another, temporally prior situation s2 having s1 as one of its eVects. Situation s2 is the killing and mutilating of Smith by someone—the murderer. As the murderer is to be found in s2 , but not in s1 , we must go from the demonstratively given s1 to its hidden cause s2 to evaluate the description. So there is a form of demonstrative reference in that example, as Wettstein stressed, but it can be argued that it is neither a reference to the denotatum, nor even a reference to the situation in which the denotatum is to be found, namely s2. It is a reference to another, causally related situation, namely s1. This provides a prima-facie counter-example to situational referentialism (s-referentialism, for short). On the basis of that and similar counter-examples a refutation of situational referentialism can be attempted. The argument runs as follows: 1. One cannot demonstratively refer to a situation without demonstratively referring to its constituents. (Assumption) 2. The denotatum is a constituent of the focus situation. (Since the focus situation is, by deWnition, the situation where the denotatum is to be found.) 3. It follows that one cannot refer to the focus situation without referring to the denotatum. 4. Therefore, if the focus situation is always referred to, as situational referentialism claims, the denotatum also is always referred to, and no attributive (nonreferential) use of incomplete descriptions is possible.

28

I. Incomplete Descriptions 5. But it is a fact that there are attributive uses of incomplete descriptions (as in the murderer example). 6. It follows that the focus situation is not always referred to: hence situational referentialism is false.

Faced with the alleged counter-example and the alleged refutation, the situational referentialist has an easy reply. Assumption 1 can and should be denied. Just as one can perceive a complex object without perceiving all its parts (e.g. I can perceive a cow without perceiving its tail), one can perceive or demonstratively refer to a situation without perceiving or demonstratively referring to all its constituents. That is arguably what happens in the alleged counter-example to situational referentialism—the murderer example. Let us reanalyze that example as follows. The speaker and hearer are perceptually confronted with a situation in which Smith is dead and mutilated. That is situation s1 . The relevant focus situation is not s1 but an extension of s1 incorporating the event s2 that caused s1 . Let us dub that extension s3 . Situation s3 contains both s1 and s2 as proper parts. It is demonstratively given since its proper part s1 is, just as the cow is perceptually available as soon as a relevant portion of the cow is. And just as the perceptual availability of the cow does not entail the perceptual availability of all its parts (e.g. its tail), there are aspects of the demonstratively given situation s3 which are not demonstratively given. In particular, the temporally anterior portion of s3 , namely s2 , is not demonstratively given; and it is in that portion that the denotatum (the murderer) is to be found. Of course, it may be deemed arbitrary to claim that the focus situation is s3 rather than s2 . But that option is not ruled out, and that is suYcient to dispose of the alleged counter-example to situational referentialism. In another important class of alleged counter-examples, the relevant situation is mentioned in the discourse—it is given linguistically rather than extralinguistically. Anaphoric uses of deWnite descriptions such as (1) fall in that category. (1)

I met a child and a woman in Lyon the other day. The woman gave me her newspaper.

The situation relevant to the evaluation of the description ‘the woman’ in the second sentence is the situation described by the Wrst sentence: a particular event in the life of the speaker, namely his meeting a child and a woman in Lyon the other day. ‘The woman’ denotes the unique woman in that situation. Since the focus situation is described rather than demonstratively referred to, this type of case constitutes another alleged counter-example to situational referentialism.9 But that type of counter-example is not convincing either. According to John Austin (1950), when we say things like ‘I met a child and a woman in Lyon the other day’, we do two things: we refer to a historic situation via the demonstrative components of the 9 According to a Davidsonian, event-based analysis, the Wrst sentence of (1) should be understood as saying that there is a situation of a certain type, namely a past event consisting of my meeting a child and a woman in Lyon the other day. The situation is no more referred to than the child and the woman are.

Recanati, Descriptions and Situations

29

sentence (e.g. the past tense, the name ‘Lyon’, etc.), and we describe that situation as a situation of a certain type (involving a man and a woman, etc.). There are various ways of implementing Austin’s proposal (Recanati 1999: 113–15), but as soon as we accept the basic framework we can no longer say that the situation which serves as focus situation for the next sentence is merely described. In the Austinian framework the Wrst sentence both describes and refers to the situation in question. Hence anaphoric uses of descriptions do not constitute a counter-example to situational referentialism. The Austinian reply itself can be found unconvincing. If the speaker refers to a historic, real situation, how can we be sure that that situation did not involve several women, even if the speaker, in his description of the situation, mentions only one? Assume, for example, that the speaker met the child and the woman in the train station. Presumably, there were other women around, which the speaker may or may not have noticed. The uniqueness implication conveyed by the description ‘the woman’ in the subsequent sentence therefore suggests that the completing situation is not the real situation (with all its complexity, including passing women in the background), but rather the situation as described by the previous sentence. Only in that situation will the woman be suitably unique (since the speaker mentioned only one woman). But if we so construe the focus situation, the Austinian response on behalf of situational referentialism is no longer available: we can no longer say that the focus situation is referred to rather than merely described. The situational referentialist can make the following reply. The historic situation s1 in which the meeting took place is complex and diVerent sub-situations can be discerned within it. Arguably, the focus situation tacitly referred to by the speaker when he utters the incomplete description ‘the woman’ is not s1 in its entirety but a sub-situation s2 contained in s1 : a minimal sub-situation satisfying the description the speaker makes of s1 , that is, a minimal sub-situation s2 s1 in which the speaker meets a child and a woman. In that minimal sub-situation s2 , there is only one woman.10 That is not the end of the discussion. Suppose the speaker is a woman. In the minimal sub-situation s2 there will be two women, namely the speaker and the woman she met. Yet the deWnite description ‘the woman’ still carries a suggestion of uniqueness. What are we to do to solve this diYculty? I have no deWnite answer to that question, but I am conWdent that ingenious solutions have been, are being or will be devised, and I conclude that anaphoric cases such as (1) provide no decisive counterexample to s-referentialism. That is not to say that there are no decisive counter-examples. I think there are plenty of them. Often, the relevant situation is not ‘given’ at all, whether linguistically or extralinguistically. Rather, the speaker quantiWes over situations of the relevant type. Thus she can say: 10 is the part-of relation between situations. A minimal situation such that p is a situation s such that (i) in s, p, and (ii) there is no s0 such that (a) in s0 , p and (b) s0 is a proper part of s. (See Heim 1990a: 146. Borrowing ideas from S. Berman, Heim deWnes ‘minimal’ as follows: min S ¼ {s 2 S: :9s 0 2 S[s 0 s & s 0 6¼ s]}, where S is a set of situations.)

30

I. Incomplete Descriptions (2)

Whenever I go to Lyon by train, the controller is a woman.

Here the situation with respect to which the description ‘the controller’ is to be evaluated is not singled out uniquely, by whatever means. The speaker universally quantiWes over situations in which she goes to Lyon by train. Each particular assignment of value to the quantiWed variable (each train trip to Lyon) provides a domain of objects in which there is a unique controller, but none is singled out in particular. This sort of case constitutes a decisive counter-example to situational referentialism and, presumably, to the quasi-referentialist project of showing that incomplete descriptions are always completed through an act of reference. Just as there may be tacit as well as explicit reference to the relevant situation, there may be tacit as well as explicit quantiWcation over situations. In one reading of (3)

The president lives in the White House.

there is implicit universal quantiWcation over a certain sort of situations which we may call ‘US situations’. In all such situations s, whoever is the president in s lives in the White House. (Below I shall represent such cases by enclosing the implicit quantiWer within angle brackets instead of square brackets.) If, instead, the speaker refers to the actual US situation, in which Bush is president, (3) gets another reading. Note that the ‘referential’ interpretation is not mandatory, even on that reading. We may tacitly refer to the current situation and say of whoever is the president in that situation that he lives in the White House. Although nonreferential (as far as the denotatum is concerned) that interpretation is quite diVerent from the ‘generic’ reading, in which we tacitly quantify over situations.11 In an important class of examples, often discussed in the recent literature (Stanley and Szabo´ 2000a; Bach 2000, Lepore this volume), there is explicit quantiWcation over objects, and a correlative, albeit implicit, quantiWcation over situations. Thus Kuroda gives this example from Japanese (1982: 48–9): (4)

subete no kyoozyu-ga gakusei-o minna rakudai-saseta all professor-SUBJ student-OBJ all Xunked

In French, this would be translated as: ‘Tous les professeurs ont recale´ tous les e´tudiants’ (literally: all the professors Xunked all the students). Like the Japanese sentence, the French sentence is ambiguous. Among its possible readings there is the following: every professor Xunked all of his/her students. Here each professor x determines a 11 As this brief discussion shows, a simple sentence such as (3) has a number of possible interpretations. One may or may not refer to the denotatum; and one may or may not refer to the situation with respect to which the description is to be evaluated. In the generic reading of the description in (3) we do not refer but quantify over situations; in the referential reading we refer to both the situation and the denotatum which features in it; in the attributive reading, we do not refer to the denotatum, but we may or may not refer to the situation with respect to which the description is to be evaluated. (Sentence (3) can be interpreted as saying that in all US situations s, the president in s, whoever he is, lives in the White House; or as saying that in this particular situation, s1 , the president in s1 , whoever he is, lives in the White House.) Moreover, when there is implicit quantiWcation over situations, the implicit universal quantiWer can be variously restricted.

Recanati, Descriptions and Situations

31

situation f (x): the teaching situation involving that professor and his or her students. It is the teaching situation f (x) which serves to evaluate the quantiWer ‘all the students’: every professor x Xunked all the students in the situation f (x). The situations in question are (implicitly) quantiWed over, as a result of (explicitly) quantifying over the arguments to the (unarticulated) function f from professors to situations. Finally, we should take notice of a group of cases which possess features in common with all the examples discussed so far: (5)

Each time I go to Lyon, I meet a woman and a child. The woman gives me her newspaper and I thank her for her kindness.

Let us focus on the interpretation in which the indeWnite descriptions in the Wrst sentence take narrow scope, in such a way that, like the controller in example (2), the woman–child pair can be diVerent for each trip to Lyon. As in (1), the deWnite description ‘the woman’ in the second sentence of (5) is evaluated with respect to some situation(s) introduced in the Wrst sentence, namely the (minimal) situation(s) in which the speaker goes to Lyon and meets a woman and a child. Here, however, it is clear that the speaker does not refer to the historic situation in which he meets a woman and a child; he quantiWes over such situations without singling out any particular one of them. That is similar to (2), except for the following feature: in (2), the description ‘the controller’ is evaluated with respect to a range of situations introduced by the universal quantiWer ‘Everytime I go to Lyon by train’, and it occurs within the scope of that quantiWer. Hence it can be suggested that some form of ‘binding’ occurs (see section 4 below). But in (5) the description does not occur within the scope of the quantiWer. If, therefore, we think that some form of binding takes place even in (5), we must posit an implicit quantiWer in that example just as we did for examples (3) and (4). I will argue that, in the second sentence of (5), there is an implicit quantiWcation over situations which is parasitic on the explicit quantiWcation over situations that takes place in the Wrst sentence.

4. Explicit and Implicit Binding of Situational Variables It is often said that the completion problem concerns quantiWers in general, and descriptions as a special case (since descriptions themselves can be construed as quantiWers—a view which I accept). The treatment in terms of situations seems to support such a view. In Recanati 2000 I oVered the following picture: a situation is a portion of the world which determines a set of facts—the facts which hold in the situation. Each fact consists of an n-place relation and a sequence of n arguments. The domain of a situation is the set of objects that are constituents of some fact holding in the situation. Often the domain of discourse is restricted to the domain of the situation which happens to be in focus. When we say ‘Everybody came’, we mean that everybody in the

32

I. Incomplete Descriptions

relevant situation s came. (‘Everybody in situation s’ means: every person who belongs to the domain of s, that is, every person who is a constituent of some fact in the ‘factual set’ of s.) Situations thus provide a restricted domain of discourse, over which quantiWers can range. Still, I think it is a mistake to view situational completion as primarily concerning quantiWers. Situational completion primarily concerns predicates. In a standard conception, predicates denote sets of objects, and quantiWers denote relations between sets of objects. In ‘Every F is G’, ‘every’denotes that relation which holds between the set of F s and the set of Gs just in case the former is included in the latter. In ‘The F is G’, ‘the’ similarly denotes a certain relation between the set of F s and the set of Gs, namely the relation which holds just in case j{F }j ¼ 1 and {F } {G}. In both cases the completion problem arises because predicates denote sets of objects only relative to situations; hence before we can evaluate a quantiWcational statement we must evaluate the predicates with respect to situations so as to determine the sets of objects which serve as arguments to the quantiWer. When I say that predicates denote sets of objects only relative to situations, I intend this as a rather trivial point. Some objects are red in a given situation, which may no longer be red in a diVerent (say, temporally posterior) situation. So the set of red objects is variable and depends upon the situation at stake, even if we do not vary the domain from one situation to the next. There may be properties which stick to their objects in the sense that, if an object has them in a situation, it must have them in any situation to the domain of which that object belongs. Even if there are such sticking properties, still the predicates which correspond to them will possibly denote diVerent sets of objects in diVerent situations because the domain of objects itself can vary from one situation to another. So predicates require situations for their evaluation. The same thing holds for sentences. Whether a sentence is true or false depends upon how things are in the relevant circumstance of evaluation. When a sentence is asserted in isolation, the circumstance of evaluation is determined pragmatically. That pragmatically determined circumstance of evaluation I call the exercised situation—the situation with respect to which the speaker intends his utterance to be evaluated. The circumstance of evaluation for a given sentence may also be determined linguistically, by preWxing that sentence with a situation-indicator. In such a case, the circumstance of evaluation for the sentence is determined by the nearest situation-indicator above it. For example, if we analyze the sentence ‘It will rain’ as ‘It will be the case that þ it rains’, we shall say that that sentence is true in a situation s iV there is a situation s 0 temporally posterior to s such that the embedded sentence ‘it rains’ is true in s 0 . The situations relevant to the evaluation of the embedded sentence ‘it rains’ are determined by the temporal operator ‘it will be the case that’ right above that sentence. Whether or not a sentence occurs in isolation, the main predicate in that sentence— the predicate which corresponds to the topmost verb-phrase—is always evaluated with respect to the circumstance of evaluation for the sentence in question. Consider the

Recanati, Descriptions and Situations

33

simple sentence: ‘Every student laughs’. There is simply no possibility of a divergence between the situation with respect to which the sentence is evaluated and the situation with respect to which the main predicate, ‘laughs’, is evaluated. That means that, if the sentence is asserted in isolation and evaluated with respect to some exercised situation s, the set of laughers which serves as second argument to the quantiWer ‘every’ will be the set of laughers-in-s. In contrast, the predicate which occurs as part of the nounphrase, ‘student’, can be evaluated with respect to any situation: it may be the exercised situation s (in which case every student-in-s is said to be among the laughers-in-s), it may be the situation of utterance c (in which case every student-in-c is said to be among the laughers-in-s), or it may be any auxiliary situation which happens to be suYciently salient for either linguistic or extra-linguistic reasons.12 In line with what has been said, we can adopt a notation inspired from Kuroda’s ‘indexed predicate calculus’ (Kuroda 1982) and associate a free situational variable with each noun-phrase predicate. Such a variable can be pragmatically assigned a value in context, but it can also be semantically bound, as in the examples I gave in the previous section: (2)

Whenever I go to Lyon by train, the controller is a woman.

Here the main clause ‘the controller is a woman’ is evaluated with respect to the situations introduced by the situation-indicator ‘whenever I go to Lyon’ (construed as a universal quantiWer over situations of a certain type). The noun-phrase predicate, ‘controller’, is associated with a situational variable which can be assigned any value, but which can also be bound by the quantiWer, as in the most natural reading of (2): [Every s: in s, I go to Lyon by train] [the x: controllers (x)] (woman (x))13 As I pointed out, situational binding can also be implicit. Thus I represent example (3), on its generic interpretation, as follows. (Implicit quantiWers occur within angle brackets instead of square brackets.) <Every s: US-situation (s)> [the x: presidents (x)] (x lives in the White House) The most interesting cases of that sort are those in which the implicit quantiWcation over situations is parasitic on some explicit quantiWcation, as in (one reading of ) Kuroda’s Japanese example: 12 See Recanati 1996: 454–7, where I discuss, inter alia, Kempson’s example: ‘The hostages were welcomed home by the President’. As Kempson points out, ‘at the point in time in the past at which a set of people is welcomed home by their president, they are transparently no longer hostages’, hence we need to evaluate the noun-phrase with respect to a situation distinct from that with respect to which the sentence is evaluated. 13 It is possible to explicitly represent the situation with respect to which the main clause ‘the controller is a woman’ is evaluated, by introducing another situational variable bound by the situation-indicator: ‘[Every s: in s, I go to Lyon by train] [the x: controllers (x)] (woman (x))s . Such a variable is redundant, however, since the situation of evaluation can only be determined by the situation-indicator right above the sentence. This feature makes the situation-indicator similar to a modal operator, and I emphasize that similarity by omitting the redundant situational variable in the nuclear scope.

34

I. Incomplete Descriptions (4)

Every professor Xunked all the students [i.e. all of his or her students]

While the situational variable associated with the noun ‘professor’ remains free and must be pragmatically assigned a value, that associated with the noun ‘student’ is indirectly bound by the quantiWer ‘every professor’: the situations which Wll the extra argument place associated with the predicate ‘student’ are the values of the implicit function f from professors to teaching situations, hence they covary with the professors. Kuroda represents this reading of (4) straightforwardly as (4a), where a functional expression ‘f (x)’ is used instead of a situational variable to complete the predicate ‘student’: (4a) [Every x: professors (x)] [Every y: student f (x) ( y)] (Xunked (x, y)) In order to provide a uniWed semantic analysis, Stanley and Szabo´ (2000) generalize that sort of move: they systematically replace situational variables by ordered pairs of an objectual variable and a higher level function variable. Relative to a context, the higher level function variable is assigned a function from objects to quantiWer domains (sets), while the other variable is assigned an object. In this framework (4) will be analyzed as [Every x: professorg(z) (x)] [Every y: studentf (x) (y)] (Xunked (x, y)) Here, the object variable z will be contextually assigned a certain school, and the function variable ‘g’ the function mapping schools to the set of people working for them. The set denoted by ‘professorg(z) ’ under this assignment is the set of professors in school z. As for the other noun, ‘student’, the associated function variable will be assigned, as in Kuroda’s treatment, the function mapping professors to the set of students in their class. Since the object variable ‘x’ is bound by the initial quantiWer ‘every x: professorg(z) (x)’, diVerent sets of students will be denoted by ‘student f (x) ’ relative to diVerent assignments of values to the quantiWed variable ‘x’: for each professor x, ‘student f (x) ’ will denote the set of students in that professor’s class. Stanley’s and Szabo´’s proposal complicates the picture by forcing us to introduce a pair of an objectual variable and a higher-level function variable even when we could directly relativize a predicate to a given situation. This is OK if the complication is necessary to achieve a uniform analysis, but in the present case the complication seems to me unnecessary. If we acknowledge the phenomenon of implicit quantiWcation, as we clearly should, then we can use straight situational variables and represent (4) as follows. [Every x: professors (x)] < Every s0 : s 0 ¼ f (x) > [Every y: students0 ( y)] (Xunked (x, y)) The situational variable associated with the noun ‘student’ is bound by the implicit quantiWer ‘every s 0 : s 0 ¼ f (x)’, which itself contains an occurrence of the objectual variable ‘x’ bound by the initial quantiWer ‘every x: professors (x)’. In this way the indirect binding of situational variables by objectual quantiWers is accounted for.

Recanati, Descriptions and Situations

35

In the previous section I sketched an analysis of (5) along similar lines and I want to pursue it here. There are two sentences in (5): (5)

a. Each time I go to Lyon, I meet a woman and a child. b. The woman gives me her newspaper.

The Wrst sentence, (5a), explicitly quantiWes over situations: it says that every situation in which I go to Lyon is a situation in which I meet a woman and a child (not necessarily the same each time). The second sentence contains the deWnite description ‘the woman’, which is intuitively evaluated with respect to the situations introduced into the discourse by the preceding universal quantiWer, ‘each time I go to Lyon’. (For qualiWcations, see below.) Since the description in the second sentence falls outside the scope of that quantiWer, we must handle the dependence of the description’s domain upon the preceding quantiWer by positing an intermediate, implicit quantiWer over situations. How can we do it? The situations that are implicitly quantiWed over in the second sentence cannot be exactly the same as those that are quantiWed over in the Wrst sentence, viz. the situations in which I go to Lyon. They must contain a unique woman (since they will serve for the evaluation of the deWnite description ‘the woman’), hence they should be thought of as the minimal situations which satisfy the conditions set up in the previous sentence. By ‘the conditions set up in the previous sentence’ I mean the conditions conveyed by (i) the restriction of the explicit quantiWer, and (ii) the matrix clause. If the Wrst sentence says that for d situations such that p (restriction) it is the case that q (matrix), then an anaphoric description ‘the F ’ in the following sentence will be evaluated with respect to the minimal situations in which it is the case that p and q. Along those lines, example (5) can be tentatively represented as follows: Each time I go to Lyon, I meet a woman and a child. [Every s: in s, I go to Lyon] [an x: womans (x)] [a y: childs (y)] (I meet x and y) The woman gives me her newspaper. < Every s0 : s 0 2 min {s: in s, I go to Lyon & I meet a woman and a child } > [the z: womans0 (z)] (z gives me her newspaper) That is not quite satisfactory, however. Just as, in the Wrst sentence, the clause ‘I meet a woman and a child’ is evaluated with respect to the situations introduced by the quantiWer ‘each time I go to Lyon’, in the second sentence the clause ‘the woman gives me her newspaper’ is to be evaluated with respect to the situations introduced by the implicit quantiWer. If we take those situations to be minimal situations in which I go to Lyon and I meet a child and a woman, as in the foregoing analysis, then we are prevented from accounting for simple variants of (5) like the following: (6)

Each time I go to Lyon, I meet a woman and a child. The woman introduces me to another woman.

36

I. Incomplete Descriptions

Since the second sentence, ‘the woman introduces me to another woman’, is to be evaluated with respect to the situations introduced by the implicit quantiWer, those situations cannot be the minimal situations mentioned above (i.e. situations in which there is a unique woman). They must be non-minimal extensions possibly containing several women—since there must be at least two women in the relevant situations for the second sentence of (6) to come out true.14 What this shows is that we need to draw a distinction between the situation(s) with respect to which the second sentence of either (5) or (6) is evaluated, and the situation(s) with respect to which the deWnite description in that sentence is evaluated. The latter situations must be minimal so as to contain a unique woman, but the former must be non-minimal (extended) so as to possibly contain several women. The analysis of (5) must be revised accordingly. I suggest that we add another implicit quantiWer, in the scope of the implicit universal quantiWer over minimal situations, so as to introduce the extended situations with respect to which the second sentence of the example is to be evaluated. (A situation s1 extends a situation s2 just in case s2 s1 & s1 6¼ s2 .) The analysis we arrive at is the following: Each time I go to Lyon, I meet a woman and a child. [Every s : in s, I go to Lyon] ([an x: womans (x)] [a y: childs (y)] (I meet x and y))s The woman gives me her newspaper. < Every s 0 : s 0 2 min {s: in s, I go to Lyon & I meet a woman and a child} >< 9s 00 : Ext(s 00 , s 0 ) > ([the z: womans0 (z)] (z gives me her newspaper))s00 For the sake of perspicuousness I have marked the situations with respect to which the clauses ‘I meet a woman and a child’ and ‘the woman gives me her newspaper’ are evaluated. (The situations in question are determined by the explicit or implicit quantiWer right above the clause.) As we can see, the clause ‘the woman gives me her newspaper’ is now evaluated with respect to a non-minimal situation s 00 , possibly containing several women, while the minimal situation s 0 of which it is an extension serves for the evaluation of the deWnite description ‘the woman’. Another class of examples which can be handled in this sort of way are the so-called ‘bound’ uses of deWnite descriptions. George Wilson (1984: 23) gives the following example: (7)

Every Bulgarian scientist who was Wred from the observatory was consoled by someone who had known the Bulgarian scientist as a youth.

According to Wilson, ‘the Bulgarian scientist’ in this sentence functions like a pronoun bound by the quantiWer ‘every Bulgarian scientist’. Such a use is not amenable to Russellian analysis, he says; for it carries ‘no implication or presupposition that the 14 Here again, I am indebted to Heim’s discussion of related issues in Heim 1990a.

Recanati, Descriptions and Situations

37

descriptor is satisWed uniquely, and there is no suggestion at all that the speaker or the audience could supply a qualiWcation to the descriptor in virtue of which it would uniquely apply’ (G. Wilson 1984: 23). In his contribution to this volume, however, Stephen Neale insists that even that type of example can be accounted for in a Russellian framework. He shows this by appealing to the approach in terms of semantic ellipsis (the ‘explicit approach’, as he calls it): he analyzes the description as ‘the y: Bulgarian-scientist y & y ¼ x’, where (i) ‘x’ is a variable bound by the higher quantiWer ‘every x: Bulgarian scientist x’, and (ii) the entire clause ‘y ¼ x’ is unarticulated and results from contextually enriching the matrix of the description. Neale concludes that the description the Bulgarian scientist’ ‘is not, pace Wilson, a variable, but just another description that is elliptical for a fuller description that contains a variable.15 Now the same sort of solution is available if we opt for the approach in terms of situational completion (the ‘implicit’ approach, in Neale’s terminology). Instead of enriching the matrix of the description with an unarticulated clause, as Neale does, we can appeal, once again, to an implicit quantiWer over situations (or, rather, to a pair of quantiWers, as in the previous example). Wilson’s example (7) can be given the following analysis: [Every x: Bulgarian-scientists (x) & Wred-from-the-observatory s (x)] < Every s 0 : s0 2 min {s: in s, Bulgarian-scientist (x) & Wred-from-the-observatory (x)} > < 9s 00 : Ext(s 00 , s 0 ) > ([The y: Bulgarian-scientists0 (y)] (x was consoled by someone who had known y as a youth))s00 The description ‘the Bulgarian scientist’ in the second relative clause is evaluated with respect to a situation introduced by an implicit universal quantiWer. (The relative clause itself is evaluated with respect to yet another situation, introduced by an implicit existential quantiWer in the scope of the implicit universal quantiWer.) The dependence of the description upon the higher quantiWer ‘every Bulgarian scientist’ is accounted for by having the matrix of the implicit universal quantiWer contain variables bound by the higher quantiWer. On this analysis (7) is very much like (4): in both cases a situational variable is indirectly bound by an overt quantiWer, via an implicit quantiWer which binds that variable and itself contains a variable bound by the overt quantiWer.

5. Conclusion In this chapter I have argued that neither the referential nor even the quasi-referential analysis solves the incompleteness problem. Instead I have put forward a theory 15 Note that this provides us with the means for reducing pronouns to descriptions if we want to. E-type pronouns are classically treated as descriptions; deictic pronouns can be treated as incomplete descriptions (‘he’ ¼ the male, etc.). The main diYculty raised for descriptive theories of pronouns comes from bound pronouns. Now, following Neale’s suggestion, we can treat bound pronouns as not themselves variables, but rather as descriptions containing (bound) variables.

38

I. Incomplete Descriptions

according to which every nominal predicate is associated with a situation variable. (See Recanati (1986, 1996) for earlier versions of this view). That variable may be contextually assigned a value, but it may also be bound by a higher quantiWer, explicitly or implicitly. In this framework domain restriction proceeds through the assignment of values to the situational variable. The descriptive condition corresponding to the predicate ‘F ’ in a deWnite description ‘the F ’ becomes uniquely identifying (it is satisWed by a single object) when that predicate is evaluated with respect to the situation which is the value of the associated variable. The uniqueness constraint which the determiner ‘the’ imposes on the predicate ‘F ’ is what guides the assignment of value to the situation variable associated with it: ‘the’ in ‘the F ’ tells us that the predicate ‘F ’ must denote a singleton-set in the relevant situation, and the situation variable associated with ‘F ’ is assigned a value accordingly. In this framework one can unify a number of the uses of deWnite descriptions. A deWnite description is analyzed as a restricted quantiWer ‘the x: Fx’. What distinguishes the families of use that have been recorded in the typological literature is the situation with respect to which the predicate ‘F ’ is evaluated. Four distinctions emerge: • Complete versus incomplete. The diVerence between ‘complete’ and ‘incomplete’ descriptions pertains to the situation with respect to which such descriptions are meant to be evaluated—a partial situation for incomplete descriptions, the total world for complete descriptions (assuming there are any). • Ordinary versus anaphoric. What characterizes the so-called anaphoric uses of descriptions is the fact that the description is evaluated with respect to situations introduced in the discourse, instead of situations made salient in some other way. Anaphora may be intrasentential or intersentential, depending on where the ‘antecedent’ is found. • Situationally singular versus general. That is the distinction between cases in which the relevant situation is referred to and cases in which there is quantiWcation over situations. Both anaphoric cases (whether intra- or intersentential) and ordinary, non-anaphoric cases may be either situationally singular or situationally general. • Explicit versus implicit. Whether the situations of evaluation are referred to (s-singular uses) or quantiWed over (s-general uses), this can be done explicitly or implicitly. Several diYcult cases, such as the bound uses of deWnite descriptions mentioned by George Wilson, can be handled in terms of implicit quantiWcation over situations. Even though the four distinctions are independent, there seems to be a strong connection between anaphoricity and explicitness. Non-anaphoric cases are, by deWnition, cases in which the relevant situation is not given in the discourse, but in some other way. It follows that the relevant situation cannot be ‘explicit’ (i.e. linguistically articulated) in such cases. But the connection between the explicit/implicit distinction and the anaphoric/non-anaphoric distinction is not as strong as it may seem. The situations with respect to which a given description is evaluated may remain implicit even though

Recanati, Descriptions and Situations

39

the description counts as ‘anaphoric’ in a loose sense. That is so, for example, whenever the relevant situations are implicitly quantiWed over but the course of values of the situational variable is parasitic upon that of an explicitly quantiWed variable found elsewhere in the discourse, as in examples (4) and (7). Before closing, I would like to say something about the referential uses of deWnite descriptions with which our inquiry started. That use falls out as the special case in which the completing situation—the value of the situational variable—has the following properties: (i) it is demonstratively given, and (ii) it bears a speciWc relation to the denotatum, namely that relation which holds between a situation s and an object o just in case one cannot demonstrate s without demonstrating o, because o occupies a central position in s. (When that is so, I say that the situation s highlights the object o.) To say that the completing situation is ‘demonstratively given’ is to say that it is an aspect of c, the situation of utterance. By focusing exclusively on that feature of referential descriptions, one may be tempted to oversimplify and to treat referential descriptions as nothing but descriptions evaluated with respect to c. That, I think, is a mistake: the second feature of referential descriptions is no less important than the Wrst one. To see that, let us consider an example discussed by Heim (1991). Heim considers a view very similar to that which I put forward regarding the completion of nominal predicates. On the view she considers, nouns and verbs possess extra argument positions for worlds and times, but nouns are special in that their world and time arguments can be chosen freely. (In contrast, the world and time arguments of a verb are always locally bound, i.e. bound by the proposition abstractor or tense or modal operator at the clause boundary right above it.) Heim argues that the extra argument positions enable us to obtain something very much like a referential reading for deWnite descriptions even in contexts where the description takes narrow scope. She gives the following example: (8)

It is always the case that the player on the left wins.

Because of syntactic scope barriers, the description cannot take wide scope over the temporal operator ‘it is always the case that’. Still, we Wnd that (8) has a reading in which the actual player-on-the-left is said to win in all situations (whether or not he plays on the left in those situations). The existence of such a reading seems to support the view that descriptions have a genuine referential use which is irreducible to a matter of scope. As Heim points out, we can account for that reading in terms of the extra argument positions associated with nouns: we can consider that the situation (world– time pair) with respect to which the nominal predicate ‘player on the left’ is evaluated is not the circumstance of evaluation for the sentence (i.e. the class of situations introduced by the operator ‘it is always the case that’) but the situation of utterance c: a situation in which a certain man happens to be playing on the left. Using situation variables, (8) can be represented as [Every s: s is a game of the relevant sort] ([the x: player-on-the-leftc (x)] (x wins))s

40

I. Incomplete Descriptions

This suggests that the so-called referential reading much emphasized by Strawson, Donnellan, and others can be accounted for within the classical approach—say, in the Russellian framework—even though, as (8) shows, it cannot be reduced to a matter of scope. I agree with Heim that the relevant reading can be accounted for in this way. But I want to emphasize that this is deWnitely not the ‘referential’ reading—the reading talked about by Strawson, Donnellan, and others. Even if we assume that the completing situation invoked by the speaker is demonstratively given in the context of utterance, still this does not, by itself, guarantee that the denotatum also is demonstratively given. That will be guaranteed only if the situation highlights the denotatum. But there are many situations which involve an object without highlighting it. Thus we can construct the following variant of Heim’s example: (9)

It is always the case that the player on the left, whoever he is, wins at the last minute.

(I have added ‘at the last minute’ for purely stylistic reasons.) This has a natural interpretation on which ‘player on the left’ is evaluated with respect to the situations s introduced by the operator ‘it will always be the case that’—situations across which the identity of the player-on-the-left may vary. As Heim pointed out in connection with (8), another, ‘rigid’ reading is available, where only the actual ‘player on the left’ is at issue. In this case the speaker says of the man who happens to be the player on the left in the current situation that he always wins. Still, as the qualiWcation ‘whoever he is’ in (9) makes clear, the speaker is not in a position to refer to or identify the person in question; she can only describe him as ‘the player on the left’. For example, we may imagine that the player in question is hidden from view, so that the speaker can demonstrate the global situation, but not the player on the left who is a (hidden) constituent of that situation. Thus construed, the situation in question does not highlight the player on the left. The description is not used referentially, even though the completing situation is demonstrated. So we need two further distinctions within the category of ‘singular uses’: • Deictic versus nondeictic. A situationally singular use of a description (i.e. a use in which the situation of evaluation is tacitly or explicitly referred to) counts as deictic whenever the focus situation is (an aspect of ) the situation of utterance. That is the case in the relevant readings of (8) and (9). • Referential versus nonreferential. A deictic use of a description counts as referential (in the sense of Strawson–Donnellan) iV the focus situation highlights the denotatum. That is deWnitely not the case in (9).

2 An Abuse of Context in Semantics: The Case of Incomplete DeWnite Descriptions Ernie Lepore

1. Introduction According to Russell (1905, 1919), any sentence of the form d The F is Ge is true just in case there is exactly one F and it is G. And so ‘The president of the United States of America in 2000 is happy’ is true just in case there is a unique US president in 2000, and he is happy. So understood, this sentence expresses a perfectly general proposition and not a singular one about Bill Clinton. Critics and champions alike have fussed and fretted for well over half a century about whether Russell’s treatment is compatible with acceptable uses of incomplete deWnite descriptions,1 where a description d the Fe is incomplete just in case more than one object satisWes its nominal F, as in (1). (1)

The table is covered with books.

If Russell is right, it follows that unless every table but one is destroyed, (1), and so all of its tokens, is false. Yet utterances of (1) are often taken to be communicating something true, even though everyone knows the world is table abundant. Since using incomplete descriptions need not compromise eVective conversational exchange, how could Russell be right? This chapter was presented at various universities and conferences and I’d like to thank all these audiences for putting up with me as I tried to make sense of this debate. Special thanks to my colleague Stephen Neale for provoking my interest in this topic, and also to Mark Baker, Emma Borg, Ray Elugardo, Jerry Fodor, Lou Goble, Bob Hale, Kent Johnson, Brian Loar, Murali Ramachandan, Paul Pietroski, Barry Smith, Agustin Rayo, Franccn o¸ is Recanati, Nathan Salmon, Stephen SchiVer, Zolta´n Szabo´, Peter van Inwagen, Alan Weir, Howard Wettstein, Tim Williamson, Crispin Wright, ZsoWa Zvolenszky, and most especially, Herman Cappelen, JeV King, and Jason Stanley for extensive and often exhausting exchanges about these and other topics. 1 Quine (1940); Reichenbach (1947); Strawson (1950); Sellars (1954); Vendler (1967); Kripke (1977); Hornsby (1977); Peacocke (1975); Lewis (1979); Sainsbury (1979); Grice (1981); McCawley (1981); Davies (1981); Devitt (1981b); Wettstein (1981, 1983); Evans (1982); Salmon (1982, 1991); Barwise and Perry (1983); Recanati (1986); Soames (1986); Bach (1987); W. Blackburn (1988); Reimer (1992); Neale (1990); SchiVer (1995); Larson and Segal (1995); to name just a few of the contributors to this debate.

42

I. Incomplete Descriptions

Some authors conclude he cannot (e.g. Strawson 1950; Donnellan 1968; Devitt 1981b; Wettstein 1981), arguing that an utterance of (1) says something true only when ‘the table’ is a singular term that refers to (and does not quantify over) a speciWc table.2 And, more likely than not, were someone to utter (1) directly in front of the sole table in a room, without any other table having already been rendered salient, his auditors would take him quite naturally to be talking about this table. Of course, unlike a proper name, not every token of ‘the table’ picks out the same table, so deWnite descriptions are at best context-sensitive singular terms. The presumption is that if incomplete deWnite descriptions are singular terms why then not unite them, and treat complete ones as singular terms as well. But incompleteness by itself surely could not establish singularity. For even if this line is right about some cases, it cannot accommodate every unobtrusive use of an incomplete deWnite description, since some known incomplete descriptions are used without anything being potentially referenced. For example, Wnding a gruesomely mangled body, a detective exclaims (2). (2)

The murderer is insane.

The description ‘the murderer’ is incomplete, since murder is rampant. Yet our speaker need not have any particular murderer in mind. His attitude might be that, regardless of whoever committed this crime, what was expressed by his utterance of (2) is true, having based his conclusion entirely on the state of a victim’s body, and not on any particulars about the identity of whoever committed this crime. Explaining how (these so-called attributive) uses of incomplete descriptions (Peacocke, 1975: 117; Davies, 1981: 150; Recanati, 1986: 67; Soames, 1986: 278) can be used to say something true is one of the ultimate aims of this chapter.3 2 Many authors recommend that some uses of incomplete descriptions are complex demonstratives: ‘it seems to me to be likely that [‘‘incomplete’’] descriptions such as ‘‘the table’’ present diYculties for a Russellian analysis. It is somewhat tempting to assimilate such descriptions to the corresponding demonstratives (for example, ‘‘that table’’), replacing the ‘‘the’’, say, in ‘‘the table’’, with ‘‘that’’ as in ‘‘that table’’ ’ (Kripke 1977: 22; see also Peacocke 1975: 209; Hornsby 1977: 33; Brinton 1977: 400–2; Devitt 1981b: 517; Wettstein 1983: 52– 8; Lycan 1984b: 190; Millican 1990: 179–80; SchiVer 1995: 43). Pace these authors, in addition to not being semantically innocent, such assimilation does not advance a case for even some uses of incomplete descriptions being (demonstrative) singular terms, since expressions of the form ‘That F ’ are themselves not singular terms; but rather (restricted) quantiWer expressions. See Lepore and Ludwig (2000). 3 This sort of example is often invoked to separate referential from attributive uses of incomplete deWnite descriptions. But, unfortunately, it relies essentially on an unexplicated notion of ‘what a speaker has in mind’. However, it is easy enough to devise examples that do not so rely. Suppose someone utters ‘Every US president is married to the woman to his right’, in a context where every living past and current US president is standing directly to the left of his wife. Knowledgeable people will take the speaker to have said something true. But suppose that everyone on stage is lined up so that the president Wrst in line has several women to his right, only one of whom is his wife. This alignment renders ‘the woman to his right’ incomplete. In this case, does it even make sense to posit a singular reference? A speaker might not know how many presidents are assembled, basing his utterance on what he knows about protocol on such occasions. To claim he referred to, say, Wve women with a single utterance leaves reference thoroughly unintuitive and theoretically useless. (Indeed, had he prefaced his utterance with ‘it is possible that’, assigning it wide scope, he would, on this account, have referred to indeWnitely many possibilia!)

Lepore, Abuse of Context in Semantics

43

In short, unexceptional uses of incomplete deWnite descriptions that are not singular terms are endemic. But the traditional view that all deWnite descriptions are quantiWers contributing only general uniqueness conditions to propositions expressed by their use is jeopardized by incomplete ones if they on occasion denote. So, what are we do in the face of these commonplace linguistic facts? Giving up Russell’s achievement has proven enormously diYcult. The deWnite article ‘the’ behaves grammatically and in at least some cases uncontroversially semantically like standard quantiWer expressions, so much so that it is hard to see how a semantic theory of complex noun-phrases could proceed systematically were we to deny deWnite descriptions quantiWcational status. In response, two sorts of strategies have emerged in the literature for protecting standard quantiWcational treatments of deWnite descriptions from the phenomena of incompleteness. One strategy is to treat deWnite descriptions as harboring hidden indexical expressions, so that whatever descriptive meaning alone leaves unWnished its context of use can complete.4 The other strategy concedes that incomplete deWnite descriptions never denote, and so (1) and (2), for example, are not true. But, just the same, our psychologies compensate for the shortcomings of our language and an over-abundant world to determine what true proposition a speaker is trying to get at when he uses an incomplete description. The former strategy is semantic, in so far as it aims to show how seemingly incomplete descriptions can be both quantiWers and denoting, and so utterances of (1) and (2) can be true. (In this regard, this strategy shares a goal with the strategy that tries to treat incomplete deWnite descriptions as singular terms. They all aim to render utterances of sentences like (1) and (2) true.) The latter strategy, however, is pragmatic, inasmuch as, though it endorses Russell’s account, it concedes that nothing is ever denoted (or referenced) by an incomplete description—and so sentences like (1) and (2) are false. Still something true might get conveyed (or ‘speaker meant’) by a particular use of one of them. In what follows I will Wrst present schematically various semantic strategies, charging that each suVers from what I shall call an over-generation problem, namely, each permits uses of sentences with incomplete descriptions to express propositions that they cannot reasonably be held to express. In addition, I will argue that semantic strategies saddle deWnite descriptions with more ambiguity than they actually exhibit. The moral of these two criticisms is signiWcant: it is to practice temperance with respect to how much work we allot to context in semantics. Unfortunately, this moral is ignored by a great deal of what goes on in contemporary philosophy, with the study of incomplete deWnite descriptions being but one instance.5 And so I proVer this chapter, 4 ‘Hidden’ because they are not pronounced, even though they are real components of uttered sentences. 5 An appeal to hidden indexicals can be found e.g. in the literature on the semantics of propositional attitude verbs like believe and know (with hidden indexicals referring to modes of presentation or standards of evaluation: Richard 1990; Cohen 1999; DeRose 1992, 1995; Lewis 1996), comparative adjectives (with hidden indexicals referring to comparison classes: Partee 1973), vague predicates (with hidden indexicals referring to precise properties or classes: Lewis 1983b; Soames 1998b), in moral theory (with hidden indexicals referring to moral standards or set of preferences: Drier 1990; Davies and Humberstone 1980).

44

I. Incomplete Descriptions

in this regard, as a caution against positing indexicals in order to resolve philosophical debates. Next I will turn to so-called pragmatic approaches. Pragmatic strategies agree with semantic ones in two chief respects: they all hold that when someone speaks, something unique is (literally) said and further that whatever is said (and even what is conveyed) by an utterance of a sentence is determined exclusively by the context of utterance. Both claims, I will argue, are mistakes and exposing them will render transparent that unobtrusive uses of incomplete deWnite descriptions pose no threat whatsoever to the traditional quantiWcational treatment of deWnite descriptions. So, in eVect, my ‘solution’ to the phenomenon of incompleteness is radical. I agree with the pragmatist that no incomplete deWnite description is denoting, but neither pragmatics nor semantics oVers a workable solution to the problem of incompleteness, and practitioners of both share a mistaken picture about what is said. Eliminating that picture, I will argue, also eliminates the alleged problem of incomplete deWnite descriptions.

2. Semantic Proposals I begin with several semantic proposals that try to explain how someone can denote a` la Russell with an incomplete deWnite description. The Wrst group tries to render incomplete deWnite descriptions denoting by showing how the nominal F in d The Fe can be interpreted relationally, in context, as elliptical for a richer (complete) description. The other semantic proposal I will consider subsumes incompleteness under a more general problem of domain selection—for example, how can an utterance of ‘Everyone ate potatoes yesterday’ be true, since at least one person did not eat potatoes yesterday?— and then tries to show how in context the domain of an apparently under-speciWed quantiWer can be restricted with ‘the’ being a limiting case.6 I do not pretend my survey of semantic proposals in response to incompleteness is exhaustive. That would be to a priori limit philosophical imagination.7 It is, however, intended to be representative of what is currently on oVer in semantics (and pragmatics), and if I am right, they all suVer from excessively high expectations of both semantics and pragmatics.

Indexicality and Completion A natural and old reaction to the phenomenon of incomplete deWnite descriptions is that a speaker who knowingly uses one is speaking elliptically. She is denoting what she 6 Absorbing the problem of incomplete descriptions into the more general problem of under-speciWed quantiWcation, if legitimate, provides another reason for resisting any inference from incompleteness to singular reference. See Sainsbury (1979); Davies (1981); Westerstahl (1985a); and Neale (1990). 7 However, the only other semantic proposal for accommodating the problem of incompleteness I am aware of is discussed and criticized in n. 25. So, my discussion is as far as I know de facto complete.

Lepore, Abuse of Context in Semantics

45

would denote were she to have uttered a diVerent description, whose linguistic meaning together with the way the world is uniquely denotes (see e.g. Quine 1940: 146; Vendler 1967: 45–6, 52, 55). So, for example, it is natural to understand someone who tokens (3) and (4) in sequence, (3) (4)

John saw exactly one man in Mary’s kitchen last week. The man wore a hat.

as denoting with ‘the man’ in (4) what ‘the man whom John saw in Mary’s kitchen last week’ does, a description with a recovered ‘restricted adjunct based upon a previous occurrence of the same noun in an identifying context’ (Vendler 1967: 45–6). However, since (1) and (2) can initiate a conversation, this linguistic rule cannot govern every potential completion. Alternatively, when linguistic ellipsis alone fails to determine an unarticulated completion, it is natural to think that other sorts of contextual cues do (e.g. Quine 1940: 146; Vendler 1967: 55 V.). The question then becomes how such context-sensitivity gets triggered. Russell himself was aware of the role that overt (or explicit) indexicality can play in determining a denotation for a deWnite description, as with indexical descriptions like ‘the present king of France’, ‘the table here’, and ‘my father’ (Russell 1959: 239; see also Reichenbach 1947: 258).8 The next two semantic proposals contend that contextualization can be also eVected though covert (or hidden) indexicality. On one proposal, a contextually salient item gets indexed; on the other, it is a contextually salient restriction on a domain of quantiWcation.9 I will examine each proposal in turn.

An Implicit Argument Place Suppose, for example, someone who utters (2) is saying what he would have said had he instead uttered (200 ), where the speaker’s token of ‘her’ picks out the victim. (200 )

Her murderer is insane.

Likewise, in uttering (1), a speaker might have said what he would have said had he instead uttered (10 ) or (100 ). (10 ) The table there is covered with books. (100 ) Her table is covered with books.

8 By an ‘indexical’ here and throughout is meant any expression-type whose semantic value can vary across use, i.e. can vary from context to context. 9 I do not here want to limit how these completions get developed. Some proponents have developed hypotheses about the relationship between logical form and indexicality; others have not. My criticisms are intended to stick however these details get worked out. Elsewhere I have developed criticism of so-called ‘unarticulated constituent’ approaches, where the semantics is run not on sentences but on utterances, and so contextualization is eVected by some non-linguistic means (cf. Cappelen and Lepore 2004).

46

I. Incomplete Descriptions

The general story from which these instances are drawn is that a token of the nominal in each description can somehow latch onto the most contextually salient object that completes the description (Evans 1982: 324; Soames 1986: 279; Neale 1990: 100–1). In the present cases, it might be a location (as in (10 )), or a person (as in (100 ) or (200 )). If no such item is available, the utterance is without a truth-value, much as (10 ) and (200 ) are in a context where no female is demonstrated. We cannot evaluate this proposal without being told how such contextualization is achieved with uses of apparently non-indexical incomplete deWnite descriptions. In the sizeable literature where this sort of suggestion is recommended, or at least hinted at as an option, no proponent (or critic) actually develops an account of how an item in a context is secured when someone uses an incomplete deWnite description. Rather, recommendations are alleged in speciWc cases. In some writings tokens of incomplete deWnite descriptions are treated as contributing to what is said or expressed locations, as with some uses of ‘the table’ mapping onto the table there (Sellars 1954: 200–1; cf. also Grice 1969: 142; Vendler 1967: 55; Sainsbury 1979: 122; Evans 1982: 324). In others, tokens are treated as contributing to what is said or expressed at times, as with some uses of ‘the president’ mapping onto the present president (Recanati 1986: 61); some as introducing people, as with some uses of ‘the car’ onto our car (Soames 1986: 301), ‘the murderer’ onto his murderer (Wettstein 1981: 250–2; Salmon 1982: 44). None of these recommendations is contextually invariant. It is easy to imagine circumstances where any incomplete description can be unobtrusively used but the recommended contextualization fails. For example, when a parking attendant tells a customer, ‘The car is out front’, the latter would not take him to be denoting their jointly owned car. Or, when a detective wonders out loud, ‘Did the murderer also murder him?’, no listener would take him to be wondering whether his murderer murdered him—understood coreferentially. So by what means do speakers cancel potential completions and favor others? All of the familiar examples suggest a general proposal according to which sentence-types (1) or (2) share truth conditions with (1 ) or (2 ), (1 ) (2 )

The table (i) is covered with books. The murderer (i) is insane.

where ‘i’ is an indexical (hidden but just the same there in (1) and (2) respectively), the semantic value of which, in a context of use, is whatever most contextually salient item combines with the linguistic meaning of an incomplete description to render it denoting. If no such item exists, that use of the incomplete deWnite description is nondenoting. In short, what actually gets introduced to eVect a completion is contingent upon what gets (covertly) indexed from context to context. And what gets indexed in a context of use depends on what happens to be relevantly most contextually salient in that context. Tantalizing metaphysical, conceptual, and psychological questions remain about what determines which item in any context is most contextually relevant for rendering

Lepore, Abuse of Context in Semantics

47

a token of an incomplete deWnite description denoting, but this proposal has kept up its end of the semantic bargain by sketching how an incomplete description type might denote on an occasion of use. Before evaluating this proposal, I want to look at a variation on it for exploiting context-sensitivity, one that tries to cast the phenomena of incomplete descriptions as subsumable under the more general problem of under-speciWed quantiWcation.

Contextually Selected Domains Teachers who assert (5) are rarely challenged, and are almost always taken to have said something true. (5)

Every student must hand in homework.

Yet it is unlikely that anyone who utters (5) expects every student in every course in every school in each school district in every country to hand in homework. How can this intuition be squared with the ordinary semantics for quantiWers? On an occasion when a speaker utters (5) can she succeed in asserting (6)? (6)

Every student here must hand in homework.

If so, the question remains as to how a restriction to every student here is eVected? Standard semantic accounts of quantiWcation assume that a domain for a quantiWer expression is determined by its nominal alone, say, students for the noun-phrase ‘every student’. Relevant uses of (5) are supposed to show that a more restricted domain of quantiWcation can be determined by context as well. With an utterance of (5), so it is claimed, the quantiWer ‘every’ need not quantify over all students, only discourse relevant ones. Of course, a restriction on ‘every’ need not be as narrow as in (6). Context is supposed to determine just how broadly or narrowly a domain of quantiWcation is to be restricted. Accordingly, paralleling the last suggestion, a narrower quantiWer domain might get selected, say, a contextually salient set that intersects with the extension of ‘student’ to provide the domain of discourse for ‘every’. It might be every student present, or every student enrolled in the speaker’s class; and so on. Truth conditions for (5) might be explicitly represented by (5 ), (5 )

Every student (i) must hand in homework.

where the explicit indexical ‘i’ is a covert indexical in (5), and its semantic role is to pick out not a single item (as in the last contextual proposal) but rather a set10 that further restricts the domain of ‘every’. (5 ) can be represented only slightly more rigorously by (5 ). (5 )

[Every x: x is a student & x is in i](x must hand in homework).

10 Consider ‘If there were more than 30 bottles, John would be happy’ (Stanley and Szabo´ 2000a). This example shows we might need to invoke possible sets as well as actual ones, i.e. properties.

48

I. Incomplete Descriptions

In short, in some contexts, an indexed domain of quantiWcation might be the set of individuals enrolled in that class; in another it might include auditors. In uttering (5), a speaker succeeds in expressing something true, because in that context (5) itself succeeds in indexing a set of objects that serves to restrict ‘every student’ to a narrower contextually deWned domain. An alleged bonus of appealing to indexicality in accounting for domain selection is its explanation for how distinct tokens of the same quantiWer expression-type even within a single sentence might range over distinct domains. Suppose there are two groups of sailors, one on deck and one on shore, and all the sailors on deck waved to all the sailors on shore.11 In such circumstances, one might conclude that (7) can be used to express something true. (7)

Every sailor waved to every sailor.

Restricted to a Wxed set of sailors, an utterance of (7) asserts that every sailor in this set waved to every sailor in this same set, including each to him or herself. Alternatively, the two quantiWers in (7) might index distinct restricted domains, expressing in that context what (8) would in that same context of utterance. (8)

Every sailor (here) waved to every sailor (there).

Accordingly, truth conditions for (7) might be better represented as (7 ), (7 )

Every sailor (i) waved to every sailor (k),

where indexicals ‘i’ and ‘k’, hidden in (7), can index disjoint sets. This hidden indexical proposal for quantiWcation can obviously be extended to the article ‘the’ to explain the phenomenon surrounding ordinary uses of incomplete descriptions. A number of authors have suggested as much (Sainsbury 1979: 114; Davies 1981: 160; Lycan 1984b: 190; Recanati 1986: 68; Higginbotham 1988: 39; Neale 1990: 94–5; Salmon 1991: 89–90, 95; Stanley and Williamson 1995).12 Extended to the article ‘the’, the truth conditions for (1) or (2) can be represented as (1þ ) and (2þ ) (1þ ) (2þ )

[The x: table (x) & i (x)](x is covered with books) [The x: murderer (x) & i (x)](x is insane)

where the indexical ‘i’ attaches to the nominal expressions ‘table’ and ‘murderer’; and in a context C , whatever set (property) this indexical picks out restricts the domain of quantiWcation for ‘the’. If the intersection of the extension of ‘table’ with whatever set is indexed has one table in it and it is covered with books, (1) is true in that context; if it is not covered with books or if there is no unique table is that set, (1) is false in that context. 11 Adapted from Stanley and Williamson (1995). 12 See Larson and Segal (1995: 333–4) for reasons not to assimilate incompleteness to quantiWer underspeciWcation.

Lepore, Abuse of Context in Semantics

49

Problems for Context Determining Domain Selection Does this hidden indexical proposal for domain speciWcation rule out someone who utters (1) in a home with three tables from expressing a truth? Or, suppose he utters it in a single room in which three tables are all lined up in a row, can he still express something true? Any notion of context that excludes two of these tables demands clariWcation (Reimer 1992: 355–7; 1998a: 99–100; Larson and Segal 1995: 332). We want to know by which means context can be so Wne-grained. But this is not a semantic problem, more a metaphysical one about how to individuate contexts, and so, I will defer. There are also general worries about indexical treatments of domain selection. For example, requiring nominals to harbor hidden indexicals has odd logical consequences, at least odd to my ears (see, however, Recanati 1986: 72). (9) looks logically false, and (10) logically true, but neither is on the current strategy. (9) (10)

Some man died and it is not the case that some man died. If a man arrives, then a man arrives.

Since nothing in the semantics guarantees that the distinct pair of hidden indexicals in (9) and (10) are co-referential in every context, (9) is not contradictory and (10) is not logically true. This account of domain speciWcation also spurns certain standard equivalences, for example deWnite descriptions as equivalent to an existential and universal. On a standard Russellian account, (11) is supposed to paraphrase (12). (11) (12)

[The x: x is a man](x is tall) (9x)(x is a man & (8y)(y is a man x ¼ y) & x is tall)

But since distinct hidden indexes attach to two quantiWers in the expansion of (11) to (12), logical equivalence is not guaranteed.13 Finally, evaluating inferences can get hairy. Consider an inference from (7) and (13) to (14). (7) (13) (14)

Every sailor waved to every sailor. Popeye and Bluto are sailors. So, Popeye waved to Bluto.

According to the current proposal, this inference is not valid, since nothing in the argument guarantees which restricted domain Popeye and Bluto are in. There is a standard worry about evaluating arguments with context-sensitive expressions. So, consider the argument (15)–(16). (15)

That horse is brown and heavy.

13 Stanley and Williamson (1995: 294) embrace this conclusion that a Russellian expansion is not equivalent to its deWnite description counterpart, because the former is more context-dependent, as it allows more slots for contextual eVects than are intuitively possible.

50

I. Incomplete Descriptions (16)

Therefore, that horse is brown.

Since this argument has overt indexicals in it, in order to show it is valid requires Wxing co-demonstrations across all contexts in every model. But even with an analogous Wxing for (7) the inference from (7) and (13) to (14) is still obscured. These complaints are but pleas for elaboration. Instead of pursuing them, I want to argue that all hidden indexical accounts impose interpretations on sentences with deWnite descriptions (and quantiWers) they simply do not have.

3. Pseudo-Ambiguity If deWnite descriptions harbor an indexical in their underlying syntactic form at that level of analysis which feeds into the interpretive component of the grammar, then the best linguistic theory should justify positing such indexical elements. My Wrst complaint is that these strategies posit more ambiguity than is tolerable.14 I will begin with the strategy that says that ‘the table’ can denote because it harbors a hidden indexical that in a context of use singles out an item that anchors its descriptive content so as to render the description denoting. In diVerent contexts a diVerent item can get indexed, contingent on which item is more contextually salient vis-a`-vis the use of that description. In uttering (1), (10 ), (100 ), or something else gets expressed (see, however, Reimer 1992: 353–4).15 (1) (10 ) (100 )

The table is covered with books. The table there is covered with books. His table is covered with books.

Truth conditions for (10 ) and (100 ) are respectively, something like (17) and (18). (17) (18)

The table over there is covered with books. The table of him is covered with books.

The expressions I want to highlight are the prepositions ‘over’ and ‘of ’. These particular choices need not be correct or unique, but some such expression is needed to explain how the semantic value of ‘table’ composes with the contextually determined semantic values of ‘him’ or ‘there’ to determine a unique denotation for ‘the table’; otherwise, we have a sort of third man predicament with a set of tables and a place (or a person) and no semantic relation connecting them. (What is unclear is whether it is 14 Elsewhere I have argued that even the best syntax will not support positing hidden indexicals, e.g. such indexicals fail to support anaphoric relations; they violate weak cross over eVects; and that the alleged syntactic binding argument in support of positing a hidden indexical (Stanley 2000) is unsound (Cappelen and Lepore 2002). 15 Under-determination about which object is most contextually salient may result from normal epistemological under-determination.

Lepore, Abuse of Context in Semantics

51

best to describe the compositionality problem by appeal to adjuncts like ‘over there’ or ‘of him’ or whether it is not more apt to describe it as positing two distinct relational nouns ‘table of ’ and ‘table over’ with distinct relata—him and there (Neale 1990: 101). It does not matter how it is characterized to my main critical point.) The relation of the indexical to the object selected may vary as distinct relational nouns or adjuncts are summoned, contingent on context. This means that the account treats the noun-phrase ‘the table’ or the simple noun ‘table’ as many ways ambiguous as there are distinct ways of combining an item with the meaning to ‘table’. Suppose ‘his table’ is correctly interpreted as denoting the table of him; ‘the table there’ as the table over there; ‘that table’ as the table identical to that (see Lepore and Ludwig 2000), and ‘the topical table’ as the table under discussion.16 Since in any context any of these (or untold other) sorts of completions may be in play, contingent partly on which item is contextually salient, it follows that ‘the table’ (or ‘table’) is multiply ambiguous. That is a lot of ambiguity to have to swallow to protect from incompleteness. Positing this much ambiguity is not only uneconomical, it also gets the semantics backwards. For it erroneously aYrms that we do not understand the contribution a deWnite description makes to what is expressed or said until we have Wrst Wgured out which item its hidden indexical is picking out on an occasion of use.17 These circumstances are diVerent from normal instances of ambiguity. Ordinarily, we disambiguate an expression by appeal to various rules of thumb. The complex noun-phrase ‘Xying planes’ is ambiguous, taking one meaning if ‘Xying’ is a gerund and another if it is an adjective. Or, because the noun ‘bank’ is ambiguous, in order to understand its tokens we look to context to help determine which of its meanings is in play. But the way we exploit context with ordinary ambiguous expressions is not by seeing which objects are semantically referenced and then returning to interpret the sentence; rather, we have views about what sorts of things happen on the sides of rivers versus what sorts of things happen in Wnancial institutions, or how someone’s Xying a plane versus how a plane’s Xying overhead can be dangerous, and then we make our choice based on this background information. With incomplete descriptions, we are being asked to disambiguate (an unobvious ambiguous expression) by 16 Recanati worries that even if an object implicitly indexed adds no descriptive content to the proposition expressed by an utterance, ‘it’s still totally indeterminate which particular sentence expressing that content the uttered sentence is elliptical for—it’s indeterminate which particular ‘‘relationally descriptive term’’ the incomplete description is elliptical for’ (1996: 449–50). For polemical purposes, I shall assume it is determinate which mode of composition is invoked. I care little if it is elliptic, and assume only that it is determinate how an indexed object composes with the meaning of the nominal the indexical attaches to. 17 Thanks to Jason Stanley for helping me to see how variation in objects indexed together with variation in modes of composition, contingent on which object gets indexed, combine to generate rampant ambiguity. I am not, however, committing him to endorsing this objection. Indeed, Stanley believes the objection can be avoided by positing two indexicals, so that the logical form of (1) is ‘The table (R)(i) is covered with books’, where ‘R’ indexes a relation and ‘i’ an object. This strategy fails. First of all, it permits indexing a genitive relation, as in ‘His table’, with ‘R’ while indexing a location, with ‘i’, so that, with an utterance of (1), a speaker could assert whatever ‘There table is covered with books’ means—something one cannot do. Also, it is hard to see how the account does not saddle itself with a sort of third man argument, with both a salient relation and a salient item indexed but no statement of how they relate to each other or to the semantic value of the nominal.

52

I. Incomplete Descriptions

Wrst seeing which sort of item is being indexed and then asking which sort of compositional meaning can relate this sort of item to the extension of the nominal. That is awfully bad news for this hidden indexical strategy. In order to accommodate within the semantics every alleged case of completion for incomplete deWnite descriptions, these strategies must be suYciently liberal in what they say about what can be contextually relevant. But the more liberal they are the more implausible they become, until they wind up treating relatively simple expressions as potentially vastly ambiguous.18

QuantiWer Dependency, Ambiguity, and Hidden Indexicals A Wrst inclination might be to conclude that the ambiguity charge does not extend in the domain speciWcation strategy, because the compositional rule invoked for composing the semantic value of the hidden indexical and the semantic value of the nominal to which it attaches is uniformly set intersection. Recall that (5) is true just in case every thing that is a student and is in (i) must hand in homework, and (1) is true just in case the thing that is a table and is in (i) is covered with books, where in context ‘i’ always has as its semantic value the most contextually salient domain. Despite these claims I will argue that the compositional rule is not conjunction in certain cases. In order to run my argument I must Wrst discuss so-called scopal ambiguities involving sentences with nested quantiWers. (19) could mean that some woman is such that she is loved by every man. (19)

Every man loves some woman.

On this reading ‘some woman’ takes wider scope than ‘every man’. (19) can also mean every man is such that he loves some woman or other, perhaps a diVerent one for each man; on this reading, ‘every man’ takes wider scope over ‘some woman’. (20) is likewise ambiguous. (20)

In every photo, the red sign stands out.

(20) might mean that there is a unique red sign that stands out in every single photo. Or, it might mean that in every photo, there is always one and no more than one red sign, perhaps a diVerent one in each photo that stands out. In order to distinguish these two readings an assignment of scope alone is insuYcient. For, suppose we tried to represent these two readings of (20) along the lines of (20 ) and (20 ). (20 ) [Every x: photo x][the y: red sign y]( y stands out in x) (20 ) [The y: red sign y][every x: photo x]( y stands out in x)

18 One way to avoid ambiguity is to sacriWce compositionality: so meaning alone, even appropriately linguistically contextualized, fails to determine the proposition expressed. This route, I have argued elsewhere, is too expensive, and indeed, incoherent. Cf. Cappelen and Lepore 2004.

Lepore, Abuse of Context in Semantics

53

(20 ) fails to capture the second reading because it fails to guarantee that the range of the interior quantiWer depends on the range of the exterior one. (20 ) is logically equivalent to (20 ).19 (20 ) and (20 ) fail to acknowledge that there is a quantiWer dependency relationship between the two quantiWers in (20) in this sense: the range of the inner quantiWer varies according to the values introduced by a variable-binding operator. Once scopes are assigned, only red signs in photos are relevant. But how is this restriction to be eVected? According to hidden indexicalists, the article ‘the’ in (20) quantiWes over a domain determined both by the linguistic meaning of ‘red sign’ and some indexed set; and ‘every’ quantiWes over a domain determined by the linguistic meaning of ‘photo’ and another indexed set. On the wide scope reading of ‘the red sign’, a token of (20) is true just in case the unique red sign in the Wrst domain indexed stands out in every member of the second domain indexed. In an eVort to defend positing hidden indexicals, some authors have suggested20 that the relevant intuitive quantiWer dependency can be fostered between the two quantiWers in (20) if the hidden indexical attached to ‘red sign’ is so interpreted as to be bound by the higher quantiWer ‘every photo’, represented in (21). (21)

[Every i: photo (k) i] [the y: red sign in (i) y] ( y stands out)

Unlike (20 ), ‘[the y: red sign in (i) y]’ in (21) cannot be assigned wide scope, because of the restriction created by the Wrst token of ‘in (i)’. So, (21), unlike (20 ), is distinct in truth conditions from the wide scope reading of ‘the red sign’ in (20). Whatever domain is associated with ‘the red sign’ varies as a function of the values introduced by ‘every photo’. Assuming binding is a syntactic phenomenon,21 examples like (20) are supposed to provide evidence for a hidden variable somewhere in the syntactic structure of quantiWed noun-phrases.22 (Some authors go as far as to suggest that, without positing a hidden domain variable, it is not clear that sentences like (20) express ‘coherent propositions at all’ (Stanley and Szabo´ 2000a: 243).) In short, hidden indexicals, unbound, behave like free variables to which a semantic value must be contextually assigned. So construed, (1) is interpreted along the lines of (1 ), but when within the scope of another quantiWer, hidden indexicals can be bound by that quantiWer. In (21), the ‘i’ in ‘the red sign (i)’ can be bound by its initial quantiWer ‘every photo (k)’. That we can eVect the relevant reading of ‘the red sign (i)’

19 Lewis suggests that the best treatment for a (incomplete) deWnite description ‘The F ’ is always to assume they denote the most contextually salient F (Lewis 1983b). Sentences like (20) reveal a serious defect with this suggestion. In a context in which someone utters (20), there need not be any contextually salient F, and even if there were it could not help to secure correct truth conditions for (20), where ‘the red sign’ takes wide scope. 20 Recommended by Stanley and Szabo´ (2000a). 21 I think it might be best accounted for pragmatically. See the last section; see also Farkas (1997), who denies that the data require a syntactic treatment but are instead best accounted for semantically. 22 For the full development of this theory see Stanley and Szabo´ (2000a).

54

I. Incomplete Descriptions

by positing a variable that restricts its range looks to provide independent support for positing indexed (quantiWer) nominals in unembedded cases like (1) and (2).23 Obviously, in order to have any real conWdence in hidden indexicals, based on binding considerations, more detail is required. But even this brief sketch suYces to show that such accounts fail to secure diVerent readings for (20). For, suppose (21) accurately represents the truth conditions of (20), when ‘the red sign’ takes small scope. Should we represent its wide scope reading along the lines of (22)? (22)

[The y: red sign (i) y][every i: photo (k) i]( y stands out in i)

(22) does not parallel (21), since in swapping quantiWers to change relative scope, the relevant domain restriction, provided by ‘in’ in (21), has been conveniently dropped. If we replenish it, (22) becomes (23). (23)

[The y: red sign in (i) y][every i: photo (k) i]( y stands out in i)

How are we to interpret the Wrst token of ‘i’ in (23)? It cannot be bound by its lower quantiWer ‘every photo’, since that would merely resuscitate the small scope reading of ‘the red sign’. Even if it is an indexical, is it not perverse to presume that ‘red sign’ harbors a singular term whose semantic value is restricted to red signs in photos, or at least in something? Granted, every red sign is in something or other, but as a matter of physical fact, and not as a matter of meaning. Moreover, any such maneuver would require us to say that the semantic contribution of ‘red sign’ diVers across sentences. In ‘On every corner, the red sign stands out’, the denotation of ‘the red sign’, on its wide scope reading, would have to be restricted to red signs on corners or red signs on something? In ‘Above every entrance, the red sign sits’, on the wide scope reading for ‘the red sign’, it must denote, as a matter of meaning alone, red signs above entrances or red signs above something? And so on and so on. If a unique red sign is being denoted in each of these sentences, its meaning should not be so restricted. The problem is this: once we settle on (21) as specifying the meaning for (20 ), then dropping ‘in’, or ignoring its functional character, in interpreting (20), when ‘the red sign’ takes wide scope, is ad hoc.24 Also, there is the related problem of how to interpret sentences in which ‘the red sign’ is its sole quantiWer. Should we infer that ‘the red sign’ contributes a diVerent meaning, contingent on whether it is embedded and also contingent on how it is embedded? That is a peculiar way to go about assigning meaning to expressions. In short, for hidden indexicalists structures like ‘the red sign’ turn out to be multiply ambiguous, with their meaning (extension) depending upon the meaning of expres23 The criticisms that follow I intend to apply to any account that tries to explain how the range of one quantiWer can be (partly) determined by the range of another by exploiting the alleged context sensitivity of the dependent quantiWer. 24 If my concern is legitimate, then (some) hidden indexicalists about domain selection, namely, Stanley and Szabo´ (2000a), who urge treating the implicit indexical either as a variable bound by a higher quantiWer or, when free, as a singular term picking out a contextually salient set, would seem to be hoisted with their own petard.

Lepore, Abuse of Context in Semantics

55

sions seemingly independent of it (syntactically or semantically). This meaning dependency is just the sort of semantic non-innocence semantic theories must reject. The expression ‘red sign’ means red sign regardless of where it occurs; it never means red sign on or red sign above or red sign in. I want now to argue that hidden indexicalists permit sentences with incomplete deWnite descriptions to express too much, or much more than they are capable of expressing.25

Nonexistent Interpretations? 26 (1) can be used to express a truth, because its incomplete description is alleged to harbor a hidden indexical that can fasten on to a contextually salient item or domain.27 Suppose in using (1) a speaker indexes a location, and thereby, expresses what she would have had she instead uttered (10 ). (10 )

The table there is covered with books.

What shall we conclude about (24)? (24)

The table is there.

(24), with its incomplete deWnite description, is as likely to be used to express a truth as (1). But then, what is to stop, on a hidden indexicalist’s account, (24) from being used to express what (240 ) can? (240 )

The table there is there.

25 Another strategy I will not discuss invokes a notion of an indeterminate statement. In addition to the intuition that incomplete deWnite descriptions can be used to denote, another widely shared intuition is that they can be used to make ‘a determinate assertion’ (Salmon 1921: 39; see also Recanati 1986: 67; Neale 1990: 94). Suppose we reject this latter intuition, and assume that in many cases no unique completion is achievable. A speaker who utters (2) is not saying what she would say with (2 ) or any other sentence with a completed codenoting description. Rather, she indeterminately (or sort-of or vaguely) says them all, for every ‘one of each deWnite description that could be used to sharpen what [the speaker] vaguely meant’ (SchiVer 1995: 377; W. Blackburn 1988: 271). What is expressed with an utterance of (2) is true only if everything indeterminately expressed is true. In uttering (2), what is indeterminately expressed depends on what counts as an admissible sharpening. But what are these? Is the class of admissible sharpenings of, say, ‘the murderer’determined by what a speaker has in mind at the time of utterance? Can a speaker deem both ‘the murderer of that victim’ and ‘the murderer of John Smith’ as equally good sharpenings? Must these completions be present in a speaker’s or auditor’s mind? Or; instead, must a speaker be prepared to ‘fall back on’ them if asked to identify which supplementation he meant (W. Blackburn 1988: 271)? This suggestion faces a familiar worry: suppose an admissible sharpening is determined by false belief. Need that render what a speaker says with (2) without truthvalue, or even false? It does not seem so; suppose she based her utterance of (2) on the condition of an unknown victim. How, then, can false beliefs be excluded in determining admissible sharpenings? If they cannot, how do we explain how something true can be said with an incomplete description? The problem of identifying acceptable sharpenings, I expect, is irresoluble, but I won’t exploit my expectation in what follows. See Fodor and Lepore (1996) for further arguments against the coherence of an admissible sharpening. 26 These next few sub-sections have greatly beneWted from invaluable input by Lou Goble, Tim Williamson, Herman Cappelen, and especially, JeV King. 27 I will run my argument against strategies of the Wrst denomination, but it will be evident how to extend it to proposals of the second denomination as well.

56

I. Incomplete Descriptions

How bad is this? (240 ) looks a lot less informative than (24) does, but since hidden indexical strategies are semantic accounts about what can be said (or expressed) by sentences with (incomplete) deWnite descriptions, complaining that (240 ) looks less informative than (24) need not be a problem. After all, my present utterances of ‘I am here now’ and ‘Ernie Lepore is at Rutgers University on June 3, 2000’ are not equally informative, but according to one respectable semantic story, these two sentences can be used to say (or express) the same proposition. If the same proposition can be expressed by nonsynonymous sentences, where one is, in some sense, analytic and the other is not, then perhaps that explanation extends to (24) and (240 ). Any genuine challenge to a hidden indexicalist must establish that, appropriately contextually relativized, (24) and (240 ) express distinct propositions. I believe they do, and here are some considerations in support of my charge. First, neither (24) nor (240 ) has a true necessary reading yet, restricted to appropriate worlds, only the latter, and not the former, seems to have one. The argument can be run either by appropriately conditionalizing (24) and (240 ), or by restricting their evaluations to denoting worlds. I prefer the latter route (but the results are the same). To run this criticism, consider a restricted sort of necessity, one restricted to worlds where the deWnite descriptions denote. In all such denoting worlds, (240 ) is true. This is why (240 ) (and not (24)) seems trivial, because we often conWne our attention to such worlds. We (and Kripke encouraged us to) do a similar thing in asking whether it is necessary that Bill Clinton is human or whether it is necessary that the teacher of Alexander taught Alexander. Kripke uses this sort of argument against a version of the description theory of names when he writes: ‘If ‘‘Aristotle’’ meant ‘‘the man who taught Alexander the Great’’, then ‘‘Aristotle was a teacher of Alexander the Great’’ would be a mere tautology. But it isn’t’ (Kripke 1980: 30). Kripke is clearly expecting that we limit ourselves to considering worlds in which Alexander had a unique teacher (so that ‘the teacher of Alexander’ denotes). His view is that ‘The teacher of Alexander taught Alexander’ has this kind of limited necessity, and ‘Aristotle taught Alexander’ does not. The same applies to (24) and (240 ), namely, in any world in which the deWnite description in (240 ) denotes, it is true (thus, our feeling that it is trivial). But this is not so in any world in which the deWnite description in (24) denotes (thus, our not feeling that tokens of this sentence express something trivial). Minimally, it is worth pointing out that embracing a hidden indexical strategy requires accommodating such necessary interpretations. Second, along the same lines, consider ‘The table isn’t there’. It does not have a reading under which it expresses a necessarily false proposition, but hidden indexicalists must predict that it should, since, assigning its negation narrow scope, assuming both tokens of ‘there’ co-index, any utterance of ‘The table there isn’t there’ expresses a necessarily false proposition (again, with evaluations restricted to denoting worlds). Represented as ‘[The x: table at l (x)]NOT(x is at l)’, to be true, there must be a unique table at l of which it is false that it is at l. (On a standard Russellian account, the narrow

Lepore, Abuse of Context in Semantics

57

scope reading could be true (e.g. if there is not a unique table at l), since it could be represented as ‘NOT[The table at l: x](x is at l)’.) This strengthens the critical point, because one would expect the description to take wide scope over negation (compare other examples), and so the most natural reading of ‘The table there isn’t there’ could not be true; and its other reading is true only if there is no unique table at l. Though not knockout punches, the accretion of these counter-intuitions succeeds in clarifying commitments inherited from adopting a hidden indexical strategy.28 Each of these various cases appeals to speaker intuitions, but these intuitions seem no less Wrm than whatever other intuitions were supposed to have motivated supplementation in the Wrst place.

Screening OV A hidden indexicalists proponent might protest that these alleged troublesome interpretations are never available. Perhaps a policy of screening oV is in place, thereby, excluding any indexing of something explicitly referenced in a predicate (or elsewhere). So, for example, take a location l—if explicitly referenced by an utterance of (24), it is screened oV as a candidate completer. (24)

The table is there.

This constraint is supposed to pre-empt indexing that would render seemingly contingent sentences from expressing necessarily truths or falsehoods. Any item referenced, or expressed, explicitly is thereby rendered contextually irrelevant, at least for the purposes of supplementing a token of an incomplete description. If this ploy can be made to work, then what is expressed with (24) cannot be what would be expressed with (240 ), but, perhaps, instead what would be expressed with (25). (240 ) The table there is there. (25) Your table is there. Contextually supplementing a token of ‘the table’ by indexing an addressee secures uniqueness without rendering an utterance of (24), relative to the same assumptions, as expressing a necessity. Whatever can be said in favor of screening oV surely reXects nothing more than handy wisdom about the pragmatics of sound interpretation, and nothing about semantics. For how can semantics prohibit an explicitly referenced object from also 28 Of course, our intuition is also that we cannot express with an utterance of ‘The table is there’ a necessarily false proposition. But what prevents, on semantic grounds alone, a hidden indexical from picking out the location of utterance, issuing in a proposition that expresses what ‘The table here is there’ would, with distinct locations indexed. Or, an utterance of ‘The murderer didn’t murder him’ from expressing what ‘The murderer of him didn’t murder him’ would, with the same individual indexed? That is to say, nothing in the semantics prevents the location of the utterance and the demonstrated victim from being the most contextually salient items.

58

I. Incomplete Descriptions

being indexed?29 With uses of potentially complete deWnite descriptions, as in ‘your only table’ or ‘the table there’, an explicitly indexed person or location is typically also contextually salient. Indeed, why can a speaker not stipulate beforehand that she wants a certain location (and that location alone) to be the most contextually salient aspect of our conversation, and then proceed to utter (24)? Even in such circumstances, I presume, what she expresses is not, in any sense, rendered necessary. Furthermore, even if some sort of screening oV strategy could be made to work in the simple cases we have been discussing—though I do not see how—it would not avoid counter-intuitive consequences for more complicated cases. Contexts may or may not be accurately representable as sequences of items which context-sensitive expressions can take as semantic values (<speaker, addressee, time, place, topic of discourse, perceptually salient objects, etc.>),30 but it is not controversial that in any given context at most Wnitely many items are salient. Let C order < a1 , a2 , . . . an > every salient item, and then try to denote a table with a use of a sentence of form (26), (26)

The table w(i),

where ‘i’ is an explicit indexical inside a predicate that picks out a member of C, and ‘w’ speciWes the predicate of which ‘i’ is a constituent. ‘w(i)’, for example, might be ‘is there’, or ‘is here’ or ‘is that’, or ‘is yours’, etc. None of these envisaged instances of (26) seems to express a necessary truth (or falsehood) in C , yet counterparts of form (27) can (27)

The table (i1 )w1 (i2 ).

(as would ‘The table there is there’, ‘The table here is there’, and so on). Suppose, because ‘i2 ’ occurs in the predicate in (26), distinct contextually salient items must be indexed by ‘i1 ’ and ‘i2 ’ in (27). But consider a new sentence with enough disjoined predicates such that each item in C can be referenced by a distinct index, as in (28). (28)

The table w1 (i2 ) or w2 (i3 ) or . . . wm (in ).

According to hidden indexicalists, an utterance of (28) expresses in C what (280 ) would. (280 )

The table (i1 ) is w1 (i2 ) or w2 (i3 ) or . . . wm (in ).

But then (28), which expresses a seemingly contingent claim about the contextual salient setting of what, if anything, ‘the table’ denotes, has been transformed into a sentence which in that same contextual setting has a (nearly) necessary reading, without a possibility of further screening oV. That we can devise such sentences might convince you something is fundamentally wrong with hidden indexical strategies. 29 Or worse, harking back to the last footnote, the screening oV strategy would somehow have to be made to work so as to prohibit internal incoherence, as with ‘The table here is there’. 30 Montague (1974a); Kaplan (1989a); Lewis (1970).

Lepore, Abuse of Context in Semantics

59

By promiscuously forcing the semantics to acknowledge contextual aspects not explicitly represented as determinants of what is expressed on an occasion of use, hidden indexicalists compromise a pragmatic, but semantically independent, alliance between context and linguistic meaning. No one can (or should) deny that contextual salience is exploited by the speaker and auditors in their eVort to converge on some point or other. But most of us thought we were being subtle and creative in so doing and not that we were obliged to as a matter of meaning alone.31

4. Pragmatic Proposals Confronted with what seem like insuperable (or, at least, a lot of troublesome) diYculties for hidden indexicalists, one response is to resign oneself to every token of ‘the F ’ as nondenoting if incomplete. As things are (every token of ) (1) and (2) are not true. This does not mean one cannot communicate or convey something true using one, but in this case the proposition communicated or conveyed is not identical to whatever the uttered sentence expressed. It might be enough to explain (away) the intuitions driving the search for a semantic solution, namely, that something true was said, by establishing that (1) and (2) can be used to convey unsaid truths. Mutual knowledge between speaker and auditor that there are many F s triggers a search by auditors for what a speaker meant to convey but did not express. In uttering (1) or (2), a speaker gets across to auditors, perhaps, that the table there is covered with books, or his murderer is insane (Grice 1969: 142–3; Bach 1987: 103–4; Sainsbury 1979: 115–16; S. Blackburn 1984: 308–10; Reimer 1998a: 100–1). A speaker unrestrictedly expresses, or literally says, with an utterance of (1) that exactly one table exists and it is covered with books. Because every such utterance is obviously false, or lacks relevant speciWcity, given mutual knowledge that many tables exist, auditors will interpret a speaker as aiming to convey something else, perhaps, that the table over there is covered with books. There may be no unique proposition a speaker conveys but fails to express, if that requires something unique she has in mind to convey or a particular object she means to draw an auditor’s attention to. But indeterminacy or falsity need not hinder a pragmatic explanation, since speakers often convey more than what they express; and some of it may even be under-speciWed or false. In short, on a pragmatic explanation, an utterance of (1) does not prohibit something true from being conveyed.32 How speakers succeed in conveying unexpressed 31 Reimer (1992: 360–2) also expresses skepticism (for diVerent reasons) about supplementation strategies applied to incomplete deWnite descriptions, but nods with approval when they are extended to under-speciWed quantiWer expressions. 32 The same strategy might be tried on quantiWer under-speciWcation. When a speaker utters (8), what she says is manifestly false, but in that conversational setting, speaker and auditor have mutual knowledge that this is

60

I. Incomplete Descriptions

truths involves a non-linguistic story about speaker intentions, mutual beliefs, and so on—a familiar story about speaker meaning (Kripke 1977) and so about commonsense psychology.

Problems with Pragmatic Explanations Some skeptics Wnd pragmatic explanations of the phenomena of incompleteness intolerable. According to Neale, ‘in view of the fact that context-dependence is such a ubiquitous feature of the use of natural language, it seems likely that the [various semantic strategies for yielding denotation] yield predictions more in accordance with our intuitive ascriptions of truth and falsity’ (Neale 1990: 114–15; see, also, Soames 1986: 300; Reimer 1992: 361; 1998a: 102–3; Stanley and Szabo´ 2000a). Apparently, these skeptics’ intuitions about which utterances are true are stronger than those of the pragmatists. Larson and Segal Wnd pragmatic explanations unappealing because ‘one uses a clear falsehood to convey what one believes to be a truth only for some deWnite reason: out of politeness, coyness, a sense of drama, or, at the very least, for want of a better option. But in these cases there is no such motive: speakers are not being ironic or arch or coy’ (1995: 329).33 According to Larson and Segal, in order for it to be true that some proposition is speaker-meant but not expressed, certain conditions must obtain— which seem unfulWlled with most acceptable uses of incomplete deWnite descriptions. Pragmatists would be unmoved by these attacks, for, I suppose, they doubt we have unimpeachable intuitions about what is true—either the propositions expressed or those conveyed (or speaker-meant). In response to Larson and Segal, they would say we should not be too narrow-minded about the circumstances under which we can convey truths that are not expressed. Still, there are problems with pragmatic accounts, ones they share with semantic accounts, namely, both appeal to a notion of what is (literally) said and both ignore that what is said often exceeds what is either expressed or conveyed. I want to challenge the coherence of their shared notion of what is literally said and also its utility.

5. What is Said According to Soames, ‘the fundamental task of a semantic theory is to tell us what sentences say in various contexts of utterance. On this view, the meaning of a sentence can be thought of as a function from contexts to what is said by the sentences in so, and so the auditor concludes that the speaker is trying to convey a truth, e.g. that every student enrolled in class must hand in homework. 33 According to a pragmatic explanation, sentence (a) can be used to express a true proposition just in case every bottle in the universe is on the only table in the universe: (a) Every bottle is on the table. An audience may

Lepore, Abuse of Context in Semantics

61

those contexts’ (1989: 394; my emphasis). Soames’s picture is commonplace among participants in this debate about incompleteness, both pragmatists and semanticists.34 In what follows I will argue that this picture is wrong. BrieXy, here is how I see the relationship between sentences and their utterances. Sentences have truth conditions, as determined by their meaning. Utterances are used to make statements (i.e. to say things), ask questions, and perform other kinds of speech acts. Utterances typically (but not always) are of sentences, but what is said, what is asked, etc., by an utterance can depend not only on the truth conditions of the sentence uttered, but also upon a number of other ‘non-semantic’ features of the context of utterance. In short: (a) The truth conditions of a sentence S need not correspond to what is said or stated by an utterance of S. (b) What is said by an utterance of S can be true, even though the truth conditions for S are not satisWed (and vice versa). (c) What is said by an utterance of S can be ‘about’ something (e.g. NYC), even though the truth conditions for S make no reference (to that thing). (d ) Because of (a)–(c), intuitions about utterances of sentences can in no simple and direct way be used as guides to the truth conditions for those sentences. Here I will brieXy show how to support this general framework. In Cappelen and Lepore (1997), we pointed out that reports of the form ‘A said/ asserted/stated that p’ can be true even if A never uttered a sentence, appropriately contextualized, that is true just in case p. Since intuitions about what is said by an utterance are Wxed by acceptable indirect reports, it follows that (a)–(d ) are correct. I shall rehearse some data and arguments starting with rather obvious and trivial cases where there is a distinction between the truth conditions of the sentence uttered and what is said by that utterance. Al, convinced that Stanley is Smith’s murderer, says, looking at Stanley, ‘Smith’s murderer didn’t comb his hair today’. Bill wants to report Al’s utterance to Harriet, whom Bill knows is convinced of Stanley’s innocence. Since Bill knows Harriet is unaware both of Al’s contrary belief and of the context of Al’s utterance (staring at Stanley), he might report Al to Harriet as having said that Stanley didn’t comb his hair today. His so doing is appropriate, but is it correct? Even a philosopher whose deepest use general principles to infer from information available in context that a speaker means a diVerent proposition, maybe one with a more restricted domain. Stanley and Szabo´ (2000a) allege that the phenomenon of quantiWed contexts, which involves sentences with multiple quantiWed expressions whose intuitive readings can be captured only by assuming a variable representing the quantiWer domain of a second quantiWer expression bound by a Wrst, refute pragmatic accounts. Since pragmatic explanations do not postulate syntactically represented quantiWer domains, they cannot capture such readings. 34 The connection with pragmatic accounts is this: for anything to be pragmatically (or conversationally) implicated some speciWc p must be said, where the fact that p was said must be ascertainable and speciWable independent of and antecedently to Wxing what the conversational implicatures were given that it is p which was said.

62

I. Incomplete Descriptions

conviction is that deWnite descriptions are not ambiguous between referential and attributive uses can accept Bill’s report of Al’s utterance as true.35 Consider another sort of case, where Richard, after a particularly awful philosophy talk, remarks with heavy sarcasm, ‘That was really good.’ In most contexts it would be a mistake to report him as having said that the talk was really good—which is exactly what his words mean—but he can be reported correctly as having said that he did not like it much. Here we are obliged to appeal to the negation of what his words express to correctly articulate what he said.36 In both cases, pragmatic features of a reported utterance (a speaker meaning or an indirect speech act) get encoded into what was said. If correct reporting is a measure of what has been said—how could it not be?—how shall we account for the intuition that speakers can say something true with an incomplete description? Such talk, proponents of a more austere conception of what is said might argue, reXects nothing more than our practice of including into what is said more than whatever proposition was expressed. So, on this picture, what is said also, sometimes at least, includes what is meant but not expressed. This straightforwardly rejects the identiWcation of what is said with what is expressed, and so is incompatible with both pragmatist and semanticist strategies. Should we conXate what is said with what is meant? Do the data not recommend our doing so? Or, at least, do they not acknowledge that there is an ordinary notion of what is said that permits such conXation? But even accepting this conXation will not suYce to explain all of our intuitions about our uses of incomplete deWnite descriptions.

How the Context of a Report Can Determine What Is Said That what is said with an utterance can be partly a function of context is old hat. For any utterance of ‘That’s a toy’, what is thereby said is partly determined by a context of utterance, in particular, partly by which object gets demonstrated. So, what is said with utterances of this sentence can change even though conventional meaning does not. As we have seen, some semanticists hold that a speaker can utter (1), gesturing towards a table in front of him, and thereby say that the table there is covered with books. Appropriately contextually placed, a speaker can be correctly reported as having said truly that the table there is covered with books. On the basis of such considerations, some semanticists conclude that (1) must be contextually sensitive to the extent that its utterances can factor in contextual salience in rendering its incomplete description denoting. Some pragmatists conclude, instead, that, through speaker meaning, an utterance of (1) can convey a true proposition. But how can either semanticists or pragmatists explain the following sort of case? 35 Though I will not argue for it here, it is easy to see that we can use a description attributively to report someone who used it referentially, and vice versa. 36 Note that one cannot mimic the Wrst speaker’s sarcasm to get across that sarcasm. As best this indicates sarcasm of the reporter—or a direct report.

Lepore, Abuse of Context in Semantics

63

Suppose that whatever table is under discussion currently sits comfortably in the speaker’s father’s oYce. Has a speaker thereby said or even speaker meant with his earlier utterance of (1) that the table in his father’s oYce is covered with books? Can the means by which semanticists and pragmatists try to explain how a speaker with an utterance of (1) says or speaker means that the table there is covered with books extend to this sort of case? But how could either account factor being in his father’s oYce into what was said or meant, since a speaker himself might be ignorant or misinformed about what is or is not in his father’s oYce—maybe that table got moved there after he spoke? One reaction to this sort of case is that, though someone might report an utterance of (1) as saying that the table there is covered with books or that the table in his father’s oYce is covered with books, since these reports disagree about what was said, at least one of them must be wrong. The soundness of this reaction requires holding that an utterance of a (unambiguous) sentence can be used to say no more than one thing— which is exactly what Soames’s notion of what is said requires. So anyone who reacts this way to our alleged data must hold that: The relationship between an utterance of a sentence and what is said by that utterance prohibits a speaker from saying both that p and that q with a single utterance, if p and q are distinct propositions. But, then, how are we to explain that an utterance of ‘Rudy loves New York and New Jersey’ can say that Rudy loves New York, as well as that he loves New Jersey, as well as that he loves New York and New Jersey?37 Amending this restriction on what can be said with a single utterance, as Soames suggests, with ‘unless the proposition (‘‘immediately’’) follows from whatever is said’ fails to explain how Bush, in detailing a new economic program, says that he will not cut taxes—even though none of his words express or imply this proposition? Is that what Bush really said? Why yes. But did he literally say it, or strictly speaking, is this what he said? Are you asking me for a direct quote? If not, then that is what he literally or strictly speaking said. Can you imagine his denying it; or, the press recanting, ‘Well, yes, he sort of said it, but he didn’t literally say it?’ Or, take Franc¸ois who, in uttering ‘Amethyst is Maria’s favorite color’, not only says that amethyst is Maria’s favorite color, but also that the color of that stone is Maria’s favorite color? The latter clearly does not follow from what his words expressed. But if he has not said both, why, then, is it acceptable to report him as such? Indeed, it is inappropriate to report him as saying that amethyst is Maria’s favorite color, if you know your audience is unfamiliar with the color word ‘amethyst’. To continue, consider a professor who, when asked whether Alice failed her exam, replies ‘I failed no one’. Has he said that Alice passed? If not, why is it correct to report 37 Notice that if Rudy loves New Jersey but not New York, then, though the original utterance is false, at least one report of it attributes a truth, namely, Rudy said that he loves New Jersey.

64

I. Incomplete Descriptions

him as such, in which case what is said makes reference to Alice, though his words do not? Imagine telling poor Alice that her professor did not say whether she failed or not. Or, that, strictly speaking, or literally, or actually he did not say. That would be a lie. A skeptic might conclude that actual reporting practices are irrelevant in determining what is said. But how could anyone reach this conclusion without letting theory override practice? Still, one might protest, there are limits to what can be said with an utterance. Who would deny that? These various examples establish no more than that delimiting a priori what those limits are is not only diYcult but inadvisable. For example, suppose Frank utters (1) in a context with only one table present. Later, after another table is brought in, a question arises about what Frank said with his earlier utterance. On the sole table present when Frank spoke now sits a vase, and on the added one sits nothing. Anyone who reports Frank in this context as having said that the table is covered with books might grossly misrepresent what he said, but anyone who reports him as having said that the table with a vase on it is covered with books has gotten him just right. If you disagree, how would you in the context described usefully and correctly answer the question what did Frank say? Would you conclude that you cannot?

Diagnosis According to Frege (or at least Fregeans), what is said by a single utterance is a single proposition. On this picture, a speaker says that p just in case he assertively utters a sentence that expresses the proposition that p. Accordingly, a speaker’s words make reference to, or denote, an object just in case what he says with those words does as well. Therefore, any indirect report of his utterance according to which what he says makes reference to, or denotes, an object that neither he nor his words does, misrepresents what was said. However, what my various examples are supposed to have established is that correctly determining what is said by an utterance often requires attending to noninterpretive, non-semantic considerations. I can imagine a pragmatist applauding my eVorts but he shouldn’t, since analogous critical points apply to him. He, unlike his semanticist opponent, is more generous about what can be gotten across (but not said) with a single utterance. But he is still too restrictive, since he too wants to ascribe a special status to what is literally said, and, even more seriously, he wants to limit what can be conveyed or said to whatever propositions are determined by a speaker’s communicative intentions. The data adduced suggest that both the pragmatist and the semanticist are myopic. In addition to those cases already surveyed, consider these other particularly relevant sorts. John utters, ‘The new seven-foot center for the New York Knickerbockers played last night’. In fact, no seven-footer plays for the Knicks, but a new center has just joined the team (he is just under seven feet tall). Asked by someone else whether this new center played last night, I can correctly report John as having said that the new center for the New York Knickerbockers did indeed play last night. If my report is right, I can

Lepore, Abuse of Context in Semantics

65

correctly represent John as having said something true, even though his original utterance is false. I also can correctly represent him as denoting someone, even though his words do not. Or, suppose, you tell me, ‘I saw John wearing his blue jacket this morning’. Suppose further that John owns exactly one jacket and it is deWnitely not blue, though it is easy to see how someone not in the right lighting might think otherwise. If asked by someone else whether John was wearing his jacket this morning, I can correctly report you as having said that he was wearing his jacket. Again, context positions me to correctly report you as saying something true and also as denoting his jacket, though your exact words were false and none of them denoted John’s jacket. A distinct proposal for explaining speaker intuitions about acceptable uses of obviously incomplete deWnite descriptions and under-speciWed quantiWers emerges from this data. Incomplete deWnite descriptions are nondenoting, just as Russell taught us, and nothing semantically restricts the range of so-called under-speciWed quantiWers. The long critical discussion in this chapter is supposed to have established all of this. Of course, nothing close to a proof was (or can be) on oVer that no successful semantic account could ever be devised. But as evidence mounts, it does seem that any such eVort will Xagrantly violate other semantically innocent and theoretical intuitions we have. And, to boot, intuitions that underlie eVorts to construct semantic accounts to render incomplete descriptions denoting, and under-speciWed quantiWers suYciently restricted, can be explained away by appealing to overwhelming data that what is said with utterances of sentences involving such expressions (though neither expressed nor speaker-meant) denote and are appropriately restricted. To see how this can be possible, we had to remind ourselves that when we try to represent or articulate what is said by an utterance we aim to characterize a speaker’s act (that utterance). In so doing, our interests often are not in systematicity or generality, but rather our aim is to determine something about a particular act in a particular context C in order to pass it along onto to a particular audience situated in a (perhaps a very) diVerent (sort of ) context C . In determining what is said we obviously draw upon information about speciWc intentions, knowledge, and history of the speaker in C and, not so obviously, we can also draw upon like features of C , the context in which we report what is said. That a context of a report can inXuence a correct assessment of what is said by an utterance makes it quite diYcult to imagine how a pragmatic account, no matter how liberally construed, can explain what is said by an utterance. When our aim is to explain how what is said, with an incomplete (attributive) description, can be true or denoting, even though, according to the best semantic account available, nothing is denoted, it may be that both what his words mean and the pragmatics of their use fail us. Our reporting practices clarify that semantics should not a priori constrain what can and cannot be said by an utterance. Competent speakers make such judgements all the time, often relying on information that exceeds anything expressed or meant. This competence consists, in part, in a capacity to judge whether a report about what is said is accurate or misleading. Theorists who try to systematically incorporate contextual

66

I. Incomplete Descriptions

cues into semantic accounts of what is said seek to theorize about a practice that does not admit of it. There is no reason to believe that determining what is said will be simpler or more systematic than determining whether two items are similar.38

6. Conclusion One intended consequence of my presentation has been to reclaim determinations of what is said from philosophers who pirated that notion as constraining the semantics for natural language. Philosophers as diverse as Kaplan and Grice and Davidson have been telling us for so long that semantics aims to articulate what words say in contexts of use that it has become banal to agree with them. In their very diVerent frameworks, what is said and the proposition expressed are interchangeable notions. The upshot of our discussion is that, though there may be a technical notion of what is said for which this identiWcation is correct, if our actual reporting practices are to be any sort of clue to a determination of what is said by another’s uttered sentence, then what is reported as said and the uttered sentence reported often disagree in meaning and extension, right down to denotation and reference, and up to truth. Readers might fear that should semantics not be supposed to be occupied with determining what is said by words used, then what is left for it to do? Fret not. • Most semanticists were weaned on model-theoretic notions, like validity, tautology, inconsistency, and nothing in this discussion challenges the connection between truth and meaning, and so, the connection between these logical notions and meaning. • Many semanticists Wrst turned from formal languages (and their models) to natural language semantics, because of concerns about productivity and systematicity. It is (virtually) non-negotiable in the philosophy of language and linguistics today that no semantic theory can account for these aspects of natural language unless it is compositional, and so, one venerable project has been, and continues to be, to show how various intriguing fragments of natural language admit of a compositional semantics. Nothing in this discussion challenges the legitimacy of this project. Indeed, as far as I can tell, no legitimate endeavor for semantics is impugned by denying that determining what is said requires a semantic explanation. 38 Salmon (1991: 88–9) seems to endorse a similar view, though his use of ‘literally saying’ vs. ‘the loose or popular sense [of ‘‘say’’]’ is incompatible with the facts. There is nothing loose about reporting what was said by an utterance of (1) as described above; indeed, each may be a literal report. Furthermore, in correctly reporting an utterance with complement that does not express the proposition expressed, Salmon arbitrarily constrains what is acceptable, suggesting, wrongly I believe, that such departures are disguised de re reports (ibid. 88). See also Cappelen and Lepore (1997).

Lepore, Abuse of Context in Semantics

67

It has taken all these pages to try to establish that ‘the table’ means the table and that ‘every student’ means every student everywhere they occur. Do not believe anyone who tells you otherwise. I have also been trying to remind you that what happens to our words with their Wxed meanings once they leave our mouths is often beyond our control.39 39 This chapter was completed in 2000. In the delay between completing it and its publication, I have developed and in places changed my views. In other words, this chapter does not fully or accurately reﬂect my current views. My current views are laid out in my and Herman Cappelen’s forthcoming book Insensitive Semantics (Blackwell). Rather than rewrite this chapter completely, I will leave it here as is. However, I recommend consulting my book for the complete truth about these issues.

3 This, That, and the Other Stephen Neale

1. Introduction Debates over the correct way to accommodate so-called referential uses of deWnite (and to a lesser extent indeWnite) descriptions have aroused passions for half a century or more. The central controversy concerns the following question: does the existence of referential uses, which no one seems to doubt, have any bearing on the meaning of the deWnite article and on what a speaker says by uttering a sentence of the form ‘the f is c’ on a particular occasion (or on what he asks by uttering a sentence of the form ‘is the f c?’)? According to the Unitarian School, referential usage is a phenomenon readily explained without appeal to a referential semantics: what someone says by uttering a Longer and shorter versions of this chapter have been knocking around for some years, gathering strength, or at least steam, from the comments of readers or people subjected to this or that section in lectures, seminars, or conversation. Of necessity, the discussions of linguistic pragmatism and contexts in sections 2 and 3 overlap considerably with the corresponding sections of Neale (forthcoming a). Particular thanks go to Kent Bach, Anne Bezuidenhout, Emma Borg, Ingar Brinck, Ray Buchanan, Herman Cappelen, Mark Crimmins, Michael Devitt, Stan Dubinsky, Kevan Edwards, Paul Elbourne, Jerry Fodor, Haidy Geismar, Owen Greenhall, Richard Hanley, Gilbert Harman, James Higginbotham, Jennifer Hornsby, Paul Horwich, Jerry Katz, Saul Kripke, Richard Larson, Barry Lee, Ernie Lepore, Paisley Livingstone, Colin McGinn, Anna-SoWa Maurin, Gary Ostertag, Angel Pinillos, Franc¸ois Recanati, Marga Reimer, Mark Sainsbury, Stephen SchiVer, Barry Smith, Jason Stanley, Kenneth Taylor, Dag Westersta˚hl, Matthew Whelpton, Deirdre Wilson, and Zso´Wa Zvolenszky for comments, questions, and advice. Talks at the following institutions led to numerous improvements, largely as a result of penetrating and persistent questions: Rutgers University, New York University, University College London, the School of Advanced Studies, University of London, Oxford University, the University of Maryland, College Park, the University of Arizona, Tucson, the University of Delaware, Tulane University, the University of Genoa, the University of San Marino, the University of California, Los Angeles, Lund University, the University of Iceland, the Georg Brandes School, Institute for Nordic Philology, the University of Aalborg, and the University of Copenhagen. Special thanks go to Brian Loar, Colin McGinn, and Stephen SchiVer with whom I have taught seminars on this material; to Joseph Almog, Tyler Burge, David Kaplan, Tony Martin, and Terry Parsons for comments during and after seminars at UCLA in which much of the material was beaten into a better shape; to Michael Devitt for reading through drafts, spotting all sorts of nonsense and untidiness, and making sure something actually got to press; and to Anne Bezuidenhout, Marga Reimer, Peter MomtchiloV, and Rebecca Bryant for their patience and forgiveness. My indebtedness to the work of Dan Sperber and Deirdre Wilson, John Perry, Robyn Carston, Franc¸ois Recanati, and John Searle should be clear, and I make little attempt here to document this or that idea or piece of terminology. Finally, I gratefully acknowledge the generous support of Rutgers University, the University of Iceland, the Georg Brandes School, Institute for Nordic Philology, and the John Simon Guggenheim Foundation. The chapter is dedicated to the memory of Harry Snitcher QC.

Neale, This, That, and the Other

69

sentence of the form ‘the f is c’ always has Russellian, quantiWcational truth conditions: the speaker is saying that (roughly) there is exactly one f and every f is c. As it is often put, the speaker is expressing a general proposition that might be represented in the notations of structured propositions as (1)

hhTHE, fi, ci:

The truth conditions of (1) might be speciWed by the following sentence of a formal language containing restricted quantiWers, (2)

[the x: f(x)]c(x)

assuming something like the following Tarski-style axiom for the quantiWer the: (3)

[thek : f]c is true of a sequence s iV c is true of every sequence c is true of diVering from s at most in the k th position, and there is exactly one such sequence.

The Ambiguity School, by contrast, maintains that the deWnite article ‘the’ is ambiguous according as it is used with its Russellian or its referential semantics: on the former, what is said has quantiWcational truth conditions as speciWed above. But if it is used with its referential semantics the speaker expresses a singular (particular) proposition that might be represented as (4)

ha, ci

the truth conditions of which might be speciWed by the following sentence of the formal language: (5)

c(a).1

Several questions need answering. (i) How does the Unitarian propose to explain the facts of referential usage? (ii) How does the Unitarian propose to explain the perfectly felicitous use of so-called incomplete descriptions like ‘the table’ or ‘the man’, the matrices of which are not uniquely satisWed? (iii) Why has the viability of the Russellian analysis engendered so much debate? (iv) Why does one group of distinguished philosophers seem so convinced that the unitary Russellian analysis of descriptions is correct (Simon Blackburn, William Blackburn, Hector-Neri Castan˜eda, Donald Davidson, Martin Davies, Gareth Evans, Peter Geach, Paul Grice, Stuart Hampshire, Saul Kripke, Stephen Neale, Mark Sainsbury, Nathan Salmon, John Searle, Scott 1 The ambiguity in question is meant to be explicable or derivable in some way, an instance of polysemy rather than homonymy, more like the ambiguity in ‘horn’ (which is applied not only to the horns found on the heads of certain animals but also to certain musical instruments that are blown, early forms of which were made from animal horns) than the one in ‘bank’. Recanati (1986/87, 1989b, 1993) and Bezuidenhout (1997) explicitly deny they are postulating an ambiguity in the deWnite article, claiming that on their accounts ‘the’ has a single lexical meaning that permits some utterances of ‘the f is c’ to express general propositions, and others to express singular propositions. I took issue with this sort of position in ch. 3 of Descriptions (n. 36), and I am still inclined to think that, as stated, it is unacceptable; however, the desires that motivate it are highly instructive and I am inclined to think there is something valuable in it. Perhaps it just needs to be stated diVerently; see below.

70

I. Incomplete Descriptions

Soames, David Wiggins)? (v) Why does another seem so convinced that a semantically distinct referential reading is also needed (Joseph Almog, Jon Barwise, Anne Bezuidenhout, Robyn Carston, Michael Devitt, Keith Donnellan, Jennifer Hornsby, David Kaplan, David Lewis, Chris Peacocke, John Perry, Franc¸ois Recanati, Marga Reimer, Bede Rundle, Stephen SchiVer, Robert Stalnaker, Howard Wettstein)?2 To these questions may be added another, voiced clearly by Devitt (1997a, b; this volume), Devitt and Sterelny (1999), and Reimer (1998a), which the Unitarian needs to answer. (vi) Given that the referential use of descriptions is systematic, regular, conventional, and cross-linguistic, is it really plausible to maintain that no systematic facts about lexical meaning beyond those given by Russell’s Theory of Descriptions— and seemingly accurate for attributive uses—are invoked where we have referential uses? The question posed by Devitt and Reimer prompts three others. (vii) What is involved in polysemy or systematic ambiguity, and how does such a notion shape up with respect to Donnellan’s talk of pragmatic ambiguity or ambiguity of use? (viii) Are there related phenomena in natural language that can be explained in terms of such notions? For example, the behaviour of the third person pronoun ‘his’, which appears to admit of a use in (6) in which it is bound by ‘John’, and another in which it is merely co-referential with ‘John’: (6)

John loves his wife, and so does Paul.

Finally, (ix) what can we learn about the semantics and use of English deWnite, indeWnite, and demonstrative descriptions, and English pronouns, by looking at the use of their counterparts in other languages, and what are the concepts driving the existence and use of such devices? I shall sketch answers to most of these questions here, but it is impossible for me to talk about descriptions today without Wrst discussing the philosophical background I assumed, without much comment, in Descriptions in 1990. It has become clearer to me with each passing year that some philosophers and linguists writing about descriptions, pronouns, demonstratives, quantiWers, and names are involved in hybrid, formal 2 Although I am going to focus on the conceptual pressures implicated in the descriptions debate, I am not entirely convinced that sociological factors have not helped shape it. Revolution and reaction are as commonplace in philosophy as elsewhere, and so are the personality types drawn to both. Some who have argued against a unitary Russellian account might have been drawn to the idea of giant-slaying, while some of those who have reacted to such assaults might have been drawn to preserving the logico-semantic status quo, taking comfort in the simplicity, transparency, and range of a theory proposed or endorsed by a giant. There appears to have been some geographical and genealogical clustering. Philosophers who teach or once taught in or near UCLA (e.g. Almog, Donnellan, Kaplan, Lewis, Perry, SchiVer, Wettstein) tend to belong to the Ambiguity School; those who teach or once taught at Berkeley or Princeton (Grice, Davidson, Searle, Neale, Kripke, Soames, Salmon) tend to belong to the Unitarian School (the obvious exception being Lewis). The State of Maryland seems also to have once attracted Ambiguity Theorists (Devitt, Stich, and Wilson). The legacy of Carnap and formal language philosophy, and the inXuence of Kaplan may have played a part in the appeal of a referential interpretation around UCLA; and the inXuence of Grice and Kripke and the legacy of ordinary language philosophy may have contributed to the appeal of pragmatic explanations and the viability of a unitary Russellian interpretation at Berkeley and Princeton. No more speculation about personality and geography.

Neale, This, That, and the Other

71

exercises that have little to do with the project to which Descriptions was meant to be a small contribution. The nature of the overarching project is discussed in detail in forthcoming work, but I can and shall say just enough here to locate the ensuing discussion and render fully explicit certain syntactic, semantic, and pragmatic assumptions, particularly in connection with indexicals, syntax, and ellipsis, assumptions made by other like-minded philosophers and linguists, but which are not always announced or appreciated in the literature. With the background out of the way I want to shed some pragmatist light on utterances of ‘this’, ‘that’, ‘the’, ‘a’, and pronouns (particularly in the third person). I shall suggest (i) that reXection on elliptical utterances leads us to a place from where the ambiguity debate between the Unitarian and Ambiguity Schools seems to lack real substance, and (ii) that there is much to be learned in reaching this position, in isolating the assumptions that originally engendered debate, and in understanding the various relationships that hold between descriptive devices and pronouns. In the course of all this, I want to concede an important point to those who have presented what I shall call the Argument from Convention for ambiguity, whilst rebutting new versions of the Argument from Incompleteness and the Argument from Anaphora, as well as something I shall call the Argument from Binding.3 The pragmatist synthesis on oVer should, I think, be acceptable to Russellians and ambiguity theorists alike, especially those moved by the Argument from Convention. Finally, I want to look brieXy at binding and lay bare an important pragmatist point that has repercussions for the interpretation of utterances containing pronouns and descriptive devices, thereby holding out the prospect of a uniWed theory of pronouns. I am afraid much of the chapter is highly compressed and many things of relevance have been omitted to produce it, particularly in connection with abstraction, indexicality, the saying/implying distinction, speaker’s reference, and pronouns.

2. Linguistic Pragmatism A central goal of the philosophy of language and linguistics is to explain how language works its magic or, more accurately, how we work our magic using language: how are we able to accomplish so much by making various noises or marks, for example, how we are able to express our thoughts, or how we are able to convey or request information about the world, and about our beliefs, wishes, feelings, and so on so 3 I discussed four arguments in 1990, the Argument from Incompleteness (ch. 3), Argument from Misdescription (ch. 3), Argument from Opacity (ch. 4), and Argument from Anaphora (ch. 5). My discussion of the Argument from Opacity inherits an anachronism upon which much of ch. 4 was based (see Neale (forthcoming c).) The Argument from Binding was presented by Wilson (1991). I did not address the Argument from Convention, as I did not take it very seriously until I read Devitt’s and Reimer’s clear statements of its force. I imagine chapters in the present volume contain arguments to add to the present list of six. For an overview of the issues, see Ludlow and Neale (forthcoming).

72

I. Incomplete Descriptions

systematically and consistently?4 The noises and marks of particular interest are, of course, those belonging to, or at least governed by, systems called languages—which is not to say that examining non-linguistic acts of communication might not shed valuable light on the phenomenon of principal interest. Several pre-emptive strikes against oversight were made by the phrasing of my question. The verbs ‘convey’ and ‘request’ were used as a gentle reminder that we use noises and marks not only to make statements but also to ask questions and do a good many other things besides—threaten, warn, baptize, marry, and so on. The word ‘feelings’ was bunged in as a reminder that we sometimes use language to convey things we are nonetheless inclined to say we ‘cannot express in words’.5 And the nouns ‘noises’ and ‘marks’ were used as a reminder that, even if spoken language is viewed as primary or prior or dominant, the question must be answered in such a way that the Wnal theory extends to uses of language that do not involve speech. Let us call a theory that aims to explain how hearers manage to identify what speakers are seeking to communicate a theory of utterance interpretation, or a theory of interpretation for short. To say that we are interested in providing a theory of interpretation is not to say we are prejudicing the issue against communication that does not involve speech or writing. It might simply turn out—it surely will—that a theory of interpretation will make reference to cognitive capacities involved in interpreting nonlinguistic acts of communication, indeed non-linguistic acts more generally.6 Interpreting an utterance or inscription of a sentence involves substantially more than identifying and interpreting individual words and seeing how they have been put together to form that sentence. Probably there isn’t much it doesn’t involve, and it is hardly surprising that we have not yet succeeded in producing a theory of interpretation with much empirical clout. There have been successes in some of the subtheories—phonology and syntax, for example. But there is widespread suspicion that producing an overarching theory of interpretation will require nothing short of a complete theory of mind.7 The project of explaining interpretation has many components and involves people from several Welds.8 Philosophers have two roles, one in the boardroom, the other on 4 It has been the hope of many philosophers that the attention paid to this question will pay dividends elsewhere, e.g. by clarifying statements of independent philosophical importance, but I shall not be concerned with any of that here. Any account, however abstract, of how we are able to communicate will have to presuppose some sort of picture of what is involved in having a thought, a belief, or an intention to communicate. In principle, one’s position on the nature of psychological states may inXuence one’s Wnal position on what is involved in communication. This is particularly true if one is attracted to something like Fodor’s (1975, 1992) idea of a ‘language of thought’. 5 Throughout, I use ‘word’ and ‘sentence’ where some people might be inclined to use ‘word-type’ and ‘sentence-type’ (similarly for ‘phrase’ and ‘expression’), as I am talking about abstract linguistic entities that may be ‘tokened’ through speech and writing at least. Thus I talk about the word ‘man’ and about particular utterances and inscriptions thereof. 6 See Grice (1989), Sperber and Wilson (1986/1995), Carston (2002). 7 See e.g. Chomsky (2000), Davidson (1986), and Fodor (1983, 2001). For assessment, see Carston (2002). 8 Descriptions was (and still is) meant to form one small part of the project: to set out the relevant properties of deWnite descriptions and their components in a way that squares with and, in some sense, explains the various

Neale, This, That, and the Other

73

the shop-Xoor, as it were. First, they will attempt to articulate clearly the nature of the project (distinguishing it carefully from various other projects with which it might be confused), distinguish clearly the various sub-projects, and distinguish and analyse the central concepts or at least the relations between them (for example, meaning, saying, implying, referring, and intending). At the same time, they will attempt to work alongside linguists whose expertise involves explaining how individual words are assembled into sentences and the extent to which communicatively relevant features of the sentences we use to say things depend upon features of the words out of which they are assembled and the mode of assembly itself. And alongside psychologists who can tell philosophers and linguists about cognition, in particular about the way we integrate information from diVerent sources and channels in the process of identifying what someone is trying to communicate. With distinct nods to the American Pragmatists, to Wittgenstein, Sellars, and Quine, and to Sperber and Wilson, I call the general outlook I have on the matter of interpretation, linguistic pragmatism (or pragmatism for short). It is an outlook that can be held by philosophers, linguists, psychologists, and no doubt others—no diplomas are checked at the door—who take themselves to be involved in the project of constructing a general theory of (utterance) interpretation, construed as an empirical theory and, as such, a contribution to cognitive psychology. It might be seen as a collection of theses that can emerge only in the context of attempting to articulate the outlines of such a theory, theses whose truth may well have repercussions elsewhere but which are not themselves motivated by the desire to bolster this or that philosophical or political doctrine. (The pragmatist outlook may well be implicated in various works of a ‘contextualist’ nature, but I am anxious to distance my own views from extant contextualist proposals in epistemology, metaphysics, ethics, and political philosophy, many of which seem to me rather suspect.) Some of the central tenets of linguistic pragmatism were accepted by a number of British philosophers in the 1950s, particularly J. L. Austin, P. F. Strawson, and (contrary to the claims of some pragmatists) H. P. Grice.9 But it was not until the late 1970s, by which time the Language of Thought hypothesis articulated by Fodor, the Chomskyan idea of LF as a level of linguistic representation, and important distinctions made by Grice and Searle had truly sunk in, that the conceptual resources were forms of empirical data whilst respecting vital conceptual distinctions, which may themselves be sharpened or reWned in the process. 9 A somewhat simplistic picture of the relationship between the focus on ‘ordinary’ language and the use of ‘ideal’ or ‘formal’ languages appears to be accepted by many linguists and even some philosophers. The received view in linguistics appears to be that for some years there was a major philosophical conXict (between ‘formalists’ and ‘informalists’), which Grice somehow dissipated by distinguishing what a speaker said from what he ‘conversationally implicated’. (The picture is perhaps fostered by a naı¨ve reading of the opening paragraphs of Grice’s ‘Logic and Conversation’ and by Strawson’s (1969) bizarre claims about a ‘Homeric struggle’ in his inaugural lecture ‘Meaning and Truth’.) Some of the people in the grip of this picture have been led to conclude that Grice was not actually a pragmatist. I know from conversations with him (a) that he saw the problem of providing an accurate account of what the speaker says when using an incomplete description as providing powerful evidence for pragmatism, and (b) that he never intended to be seen as denying pragmatism.

74

I. Incomplete Descriptions

generally available to articulate the outlook clearly and in a form that made it relevant to more formal studies of language that were by that time blossoming in linguistics and philosophy departments in the United States. To the best of my knowledge, it was not until the work of Dan Sperber and Deirdre Wilson began to appear in print in the early 1980s that pragmatists made sustained eVorts to render explicit the basic tenets of their work. Indeed, without Sperber and Wilson’s work, and the work of Chomsky, Fodor, Grice, and Searle upon which it drew, philosophy and linguistics might still lack the distinctions and resources needed to say anything more substantive than the ramblings about ‘contextual meanings’ and ‘relative meanings’ that issue periodically from the darker areas of philosophy and linguistics departments (not to mention departments or ‘programs’ housing people unaccountably known as ‘theorists’ or ‘philosophers’). It would be a mistake, I think, to attempt a deWnition of linguistic pragmatism as it is essentially an outlook that engenders a very practical approach to interpretation. I cannot go into the sort of detail I go into in Neale (forthcoming b) here, so I have produced twenty-four numbered and labelled paragraphs to give the general Xavour of linguistic pragmatism as I see it. Some of the points are quite general or intuitive, others are very speciWc or theory-laden. Some are less central than others and could be withdrawn without upsetting the whole too much, but I am strongly inclined to go along with the whole lot, and there is no doubt that many gain strength through association with others. A few are held by some philosophers and linguists I would call anti-pragmatists; a few are rejected by people with a pragmatist outlook; and a few are conspicuous here by their absence elsewhere in the literature.10

1. Cooperation For the most part speakers (writers) want to be understood, and hearers (readers) seek to understand. To this extent they are involved in a cooperative exercise. Ceteris 10 Putting aside diVerences of terminology and philosophical temperament, as well as apparent disagreements about particular analyses, I am inclined to view all of the following as operating in a broadly pragmatist spirit—although my own brand of linguistic pragmatism is, I suspect, rather too ascetic, inferential, beholden to ordinary language strictures, and driven by underlying concerns about practicality and the concepts of society and regulation for many of them: Austin (1962), Bezuidenhout (1997), S. Blackburn (1984), W. Blackburn (1988), Blakemore (1987, 2002), Barwise and Perry (1983), Carston (1988, 1993, 2002), Chomsky (1976, 1986, 1995, 2000), Crimmins (1992), Crimmins and Perry (1989), Evans (1982, 1985), Fodor (1987, 2001), Grayling (1995), Grice (1989), Neale (1990, 1993), Quine (1940, 1960), Papafragou (1998), Perry (1986, 1993, 1998, 2001), Recanati (1986/87, 1989b, 1993), Rouchota (1992, 1994), Searle (1969, 1979), Sellars (1954), Sperber and Wilson (1986/1995), and Strawson (1950, 1952). I suspect pragmatism is also taken for granted by many others who take truth to be property of what a speaker says or expresses by uttering a sentence X on a speciWc occasion, and not of what X itself says or expresses relative to a context. I shall point to or draw liberally from the work of other pragmatists as I go. (In the case of Sperber and Wilson I shall not bother to cite: every page of the present chapter can be viewed as containing an aphonic footnote that could have been rendered, as ‘See the work of Sperber and Wilson, esp. their book Relevance’.) By saying I am a pragmatist, I do not mean to be saying that I endorse the details of Recanati’s (1989b, 1993) or Rouchota’s (1992) or Sperber and Wilson’s (1986/1995) or Bezuidenhout’s (1997) or Carston’s (2002) pragmatist analyses of deWnite descriptions and the attributive–referential distinction, all of which I Wnd problematic. I shall propose an account that should be attractive not only to them but to Russellians and ambiguity theorists alike.

Neale, This, That, and the Other

75

paribus, both parties tacitly assume they are using words with shared meanings, combining these words in accordance with a shared syntax, and operating in accordance with shared and very general, rational principles of interpretation.

2. Meaning A theory of interpretation should explain how hearers (and readers) manage to integrate linguistic and non-linguistic information to identify what a speaker (or writer) meant on a given occasion by uttering (or inscribing) a linguistic expression X.11 Valuable information can be gleaned from examining situations in which we report on speech acts using sentences of the form: By uttering (or writing) X, A meant that p where the reporter is using the expression replacing ‘A’ to pick out an agent and the expression replacing ‘X’ to pick out a linguistic expression, and where the expression replacing ‘p’ is a declarative sentence. Examples: By uttering, ‘I’m tired’, John meant that he was tired. By uttering, ‘I’m tired’, John meant that he wanted us to leave. We should be suspicious of locutions of the form ‘X means that p’, where the expression replacing ‘X’ is being used to pick out a sentence (e.g. ‘the sentence ‘‘snow is white’’ means that snow is white’).

3. Explanation To interpret is to provide an explanation, and the concept of interpretation makes no sense in the absence of a problem to be solved. We reXexively generate hypotheses about the things we perceive. Nowhere is this more in evidence than when we perceive one another’s actions. We act out of reasons. To interpret an action is to form a hypothesis about the intentions behind it, the intentions that explain it. Interpreting a speech act is a special case of this. The use of language is one form of rational activity, and the principles at work in the interpretation of linguistic behaviour are intimately related to those at work in interpreting intentional non-linguistic behaviour. What makes interpreting a speech act special is that a proprietary body of information, knowledge of language, is accessed immediately in the interpretation process. The hearer’s or reader’s 11 Linguistic pragmatism does not necessarily assume there is much chance of ever producing an empirically interesting theory of interpretation. At least two pragmatists, Chomsky (2000) and Fodor (1983, 1987, 2001), have argued that asking for a theory of interpretation is tantamount to asking for a ‘theory of everything’, a complete cognitive psychology, because virtually anything can impinge upon the holistic process of interpretation. For more optimistic pragmatist outlooks, see Sperber and Wilson (1995, 1996, 2002) and Carston (2002). The present chapter assumes neither outlook, but it brings into sharp relief the need for a clear picture of what the more tractable subtheories of a theory of interpretation are supposed to do and how they must come together.

76

I. Incomplete Descriptions

goal is to identify what the speaker or writer meant. When this has been done, the interpretive problem has been solved.12

4. Asymmetry The epistemic situations of the speaker and hearer are fundamentally asymmetric: the speaker knows what he means whereas the hearer has to work it out. If you want to Wnd out whether I’m hungry (or in pain) you will have to watch me, see what I do, or ask me. I don’t have to do that. I have ‘privileged access’ to that information.13 Similarly if you want to know whether I am worried about missing my Xight, where on an aeroplane I prefer to sit, or whether I think Norway is a member of the European Union. And similarly where we have speech. Unlike you, I have privileged access to what I mean when I utter X on a given occasion. We can characterize a typical speech situation as follows. Person A intends to communicate something to some other person B. He selects a form of words X that he thinks will, in the circumstances, get across his point (and, perhaps, also get it across in some particular way or other. A knows what he means by uttering, ‘That’s his bank’, for example. He knows which thing he meant by ‘that’, who and what relation he meant by ‘his’, and what he meant by ‘bank’. B’s situation is quite diVerent: B is trying to work out what A meant and he must use anything he can get his hands on to get the job done since he has no direct access to A’s communicative intentions. The words A uses constitute partial evidence for what A meant. Other evidence may come from the physical environment, from B ’s take on the conversation up to that point (if any), from B ’s beliefs about A, and a whole lot more besides. The epistemic asymmetry of speaker and hearer underscores (i) the need to separate the metaphysical question concerning what determines (or Wxes) what A means and the epistemological question concerning what is used to identify what A means, and (ii) the need to scrutinize simplistic appeals to contexts, maxims of conversation, salience, and pragmatic factors, which are frequently and mistakenly introduced together with intentions in contemporary discussions, as if these things conspire to bridge certain interpretive gaps. Scanning the context of utterance for salient objects and bringing to bear pragmatic principles (e.g. Grice’s conversational maxims) is not going to provide A with any information that will help him identify what he meant. From A’s perspective, context and pragmatic principles have already played their roles: 12 I mean this to apply equally to interpretations in literary theory. The idea of textual interpretation makes no sense if there is no problem (e.g. about a word or phrase, a character, a plot, a work, or even a whole genre) to which an interpretation constitutes a possible solution. It is embarrassing that some ‘theorists’ who also call themselves ‘philosophers’ are unable to see this. 13 There are philosophers concerned to deny this idea today, but no coherent case has been made (indeed could be made) for a total failure of asymmetry. The idea of ‘privileged access’ to a state is often introduced with the idea of ‘incorrigibility’ (the idea that I cannot be mistaken about whether I am in the state in question). Whilst the case for denying this might be more promising, the fact that incorrigibility is at least arguable (and has been argued) in the Wrst-person case, and is not in the least arguable (and has not been argued except, perhaps, by the deluded) in the third-person case, is enough to distinguish the cases here.

Neale, This, That, and the Other

77

A’s perception of the context—whatever a context turns out to be—his perception of B’s perception of the context, the assumption that B is operating in accordance with the same pragmatic principles as A, and A’s estimation of B’s ability to work things out (and probably a whole lot more besides) have already impinged upon whatever processes led A to use the particular form of words he used with the intentions with which he used them.

5. Reciprocity Despite the epistemic asymmetry, the perspectives of A and B are not independent. The asymmetry is reciprocal or complementary as in adjoining pieces of a jig-saw puzzle. In producing his utterance, A relies on what he takes to be B’s capacity to identify what he intends to convey; B assumes that A is so relying. And, possibly, so on. The ways in which A and B operate form a dovetail joint and are mutually sustaining. And to this extent, there is simply no possibility of making sense of B’s capacity to interpret A without making sense of A’s capacity to exploit that capacity, and vice versa. So the project of constructing a theory of interpretation may be approached from either of two complementary perspectives, and an adequate answer must make sense of both.

6. Intention What A meant by uttering X on a particular occasion is determined by, and only by, certain very speciWc interpreter-directed intentions A had in uttering X. The precise content of a psychological state such as a belief or intention may be determined, in part, by something external to A and beyond A’s control (‘externalism’). Furthermore, the formation of genuine intentions is severely constrained by beliefs. I cannot intend to become a prime number, intend to digest my food through my lungs on alternate Tuesdays, or swim from New York to Sydney because (roughly) I cannot intend what I believe to be impossible. (There is no need to get into the exact force of the modal or the exact formulation of the constraint here. It is enough to recognize, as Grice (1971) does, that it is severe.) If, as Grice suggests, what A meant by uttering X on a given occasion is determined by certain interpreter-directed intentions, then assuming he is being co-operative A cannot mean that p by uttering some sentence X if he believes it is impossible for his audience B (or at least any rational, reasonably well-informed interpreter in B’s shoes) to construe him as meaning that p. Among the things constraining A’s communicative intentions are A’s beliefs about the world, his (tacit) beliefs about the sorts of interpretive principles B will be employing, and his (tacit) estimation of B’s capacity to work certain things out (the list is not meant to be anywhere near exhaustive). So without some stage-setting A cannot mean that Jones is no good at philosophy by producing the sentence ‘Jones has excellent handwriting and is always punctual’, for example, or by reproducing the mating call of some exotic bird.

78

I. Incomplete Descriptions

7. Factorization What A meant by uttering X may be factored into what A said (or asked) by uttering X and what A only implied.14 Thus, again following Grice, what A said and what A implied are determined by, and only by, certain very speciWc interpreter-directed intentions A had in uttering X.15 Nonetheless, although it would be perverse to insist upon a distinction between what A meant and what A intended to mean (and for good reason if Grice is right), a distinction between what A said and what A intended to say is not one obviously lacking a point. So, in the Wrst instance, we should separate (i) what A intended to say by uttering X on a given occasion, and (ii) what a rational, reasonably wellinformed interpreter in B’s shoes would think A intended to say by uttering X on that occasion (which is not to say there are not problems with the idea of a rational, reasonably well-informed interpreter in B’s shoes). In cases where (i) ¼ (ii), we can talk freely about what the speaker said. (In cases where (i) 6¼ (ii), certainly we could argue about which of (i) or (ii) or some third thing has the ‘right’ to be called what is said, but what would be the point? First, what third thing distinct from (i) and (ii) could be of any signiWcance to a theory of interpretation? There is simply no role for a transcendent notion of what is said upon which (i) and (ii) converge when all goes well. Second, why is a choice between (i) and (ii) even needed in cases where (i) 6¼ (ii)? Conceptually they are distinct, and they are both needed in a theory of interpretation. When all goes well, they coincide, and it’s just too bad they don’t always do so. Surely there is no philosophical payoV in bestowing the honoriWc ‘what was said’ on one rather than the other when they diverge.) Saying-intentions are constrained by belief and knowledge. A cannot (intend to) say that p by uttering some sentence X if he believes it is impossible for his audience B (or at least any rational, reasonably well-informed interpreter in B’s shoes) to construe him as saying that p. Among the things that constrain the formation of A’s saying-intentions are A’s knowledge of the meanings of the words he is using and his (tacit) knowledge of the syntax of the language he is using. Thus A cannot (intend to) say that snow is white by uttering the sentence ‘grass is green’. And he cannot (intend to) say that John asked his brother to shave him by uttering ‘John asked his brother to shave himself ’. More generally, he cannot (intend to) say that p by uttering X if he believes it is impossible for his audience B (or at least any rational, reasonably well-informed interpreter in B’s shoes) to construe him as intending to say that p.

14 In my view it is vital to distinguish between (i) what A implied by uttering X, and (ii) what A implied by saying what he said, as this distinction gets to the heart of Grice’s distinction between conventional and conversational implicature, and to provide the framework within which to solve problems concerning the former (cf. Frege’s notion of tone or colouring), non-detachability, dictiveness and formality, and central vs non-central speech acts. 15 If what A said is determined by, and only by, certain very speciWc interpreter-directed intentions A had in uttering X, then at least some contemporary talk of ‘contexts’ ‘Wxing’ or ‘determining’ aspects of what A said— e.g. the references of indexical expressions—must involve some form of confusion. See below.

Neale, This, That, and the Other

79

8. Speakers Saying and implying are things people do. (Similarly, communicating.) Following ordinary usage, the speaker is taken to be the understood subject, so to speak, of the verbs ‘say’ and ‘imply’, the verbs in talk about ‘what is said’ and ‘what is implied’. (Similarly, with verbs such as ‘communicate’, ‘convey’, and ‘get across’.) We should be initially suspicious of talk about what uses of sentences say (imply, communicate, etc.) and talk about what sentences-relative-to-contexts say (imply, communicate, etc.), unless such talk is taken to be straightforwardly translatable into talk about things that speakers are doing. And we should deplore the unannounced slipping and sliding back and forth between diVerent subjects of ‘say’ (‘imply’, ‘communicate’, etc.). At the same time, we should be open to the idea that new, technical uses of the verbs ‘say’ (‘imply’, ‘communicate’, etc.) may need to be deWned, or at least developed, in the course of our inquiries, such uses earning their keep because of the ineliminable theoretical work they do.

9. Truth What A says and implies are the sorts of things that are true or false. (Perhaps A may say things that are neither true nor false, but it might prove useful to start out sceptical about this.) It does not follow that when A utters a sentence he says only one thing or that he implies only one thing. Nor does this talk of truth mean that in order to produce a theory of interpretation we shall have to construct a theory that recursively assigns truth conditions to sentences relativized to contexts of utterance (a semantic theory, in one sense of ‘semantic’), or construct a theory that assigns things in the world to linguistic expressions relativized to contexts of utterance (a theory of reference, in one sense of ‘reference’).

10. Judgment Our intuitive judgments about what A meant, said, and implied, and judgments about whether what A said was true or false in speciWed situations constitute the primary data for a theory of interpretation, the data it is the theory’s business to explain. (Since no one has intuitive judgments about what is said by a sentence X relative to a context C or about the semantic content of X relative to C (these being philosophers’ notions), several distinct mistakes would be involved in the claim that linguistic pragmatism aims to show that our intuitive judgments about what a speaker said may be ‘unreliable guides to semantic content’. If talk of the ‘semantic content’ of a sentence X relative to a context C is just a snazzy way of talking about what the speaker said by uttering X on a particular occasion—the occasion that C is being used to partially model—then of course we can accept its empirical signiWcance. If it is not, then its empirical signiWcance must be justiWed in some other way, from within the theory of interpretation by reference to some empirical role it is required to play in an explanation of what a

80

I. Incomplete Descriptions

speaker says and implies by uttering X on a given occasion, in much the same way that notions such as binding are motivated from within.

11. Reference Saying typically involves referring and saying of. That is, saying something typically involves referring to something and saying something of it. Saying that London is pretty, for example, involves referring to London and saying of it that it is pretty. One way of doing this is to use ‘London’to refer to London and ‘is pretty’to say of it that it is pretty. Following ordinary usage, the speaker is taken as the understood subject, so to speak, of ‘refer to’ and ‘say of ’. Initially, we should deplore the unannounced slipping and sliding, back and forth, between diVerent subjects of ‘refer to’ and ‘say of ’, and we should be suspicious of talk about what uses of words refer to and say of things, and of talk about what words-relative-to-contexts refer to and say of things—unless such talk is taken to be straightforwardly translatable into talk about things that speakers are doing. But we should be open to the idea that new, technical uses of ‘refer to’ and ‘say of ’ may emerge in the course of our inquiries. Who or what A is referring to by uttering some expression X is determined by A’s referential intentions in uttering X. (Talk of ‘contexts’ ‘Wxing’ or ‘determining’ the references of expressions—for example, the references of indexical expressions—must involve some form of confusion.) Nonetheless, a distinction between what A referred to and what A intended to refer to is not one obviously lacking a point. So, in the Wrst instance we should separate (i) who or what A intended to refer to by an expression X on a given occasion, and (ii) who or what a rational, reasonably well-informed interpreter in B’s shoes thinks A intended to refer to by X on that occasion. In cases where (i) ¼ (ii), we can talk freely about what the speaker referred to. (In cases where (i) 6¼ (ii), we could argue about which of (i) or (ii) or some third thing has the ‘right’ to be called the person or thing referred to, but what would be the point? First, what third thing distinct from (i) and (ii) could be of any signiWcance to a theory of interpretation? There is simply no role for a transcendent notion of what was referred to upon which (i) and (ii) converge when all goes well. Second, why is a choice between (i) and (ii) even needed in cases where (i) 6¼ (ii)? Conceptually they are distinct, and they are both needed in a theory of interpretation. When all goes well, they coincide, and it’s just too bad they don’t always do so. Surely there is no philosophical payoV in bestowing the honoriWc ‘what was referred to’ on one rather than the other when they diverge.) Referential intentions are constrained by belief and knowledge. Assuming he is being co-operative, A cannot (intend to) refer to some particular person a by uttering some expression X on a given occasion if he believes it is impossible for his audience B (or at least any rational, reasonably well-informed interpreter in B’s shoes) to construe him as referring to a.16 Among the things that constrain the 16 The matter is complicated by the fact that in certain circumstances one may seek to disguise one’s intended referent, e.g. in cryptic poetry or diary entries. The issues here are intimately connected to Grice’s (1989) discussions of communicative intentions in the absence of an audience.

Neale, This, That, and the Other

81

formation of A’s referential intentions are A’s knowledge of the meanings of the referring expressions he is using and his (tacit) knowledge of the syntax of the language he is using (which may bear on the matter of co-reference or binding where pronouns are concerned). Thus A cannot (intend to) refer to some particular individual a by X if he believes it is impossible for his audience B (or at least any rational, reasonably wellinformed interpreter in B’s shoes) to construe him as (intending to) refer to a by X. Similar points can be made in connection with A’s predicative intentions.

12. Aphonicity It is now time to get more theoretical. A distinction between PF (‘Phonetic Form’) and LF (‘Logical Form’) in something like Chomsky’s sense is almost certain to play a key role in a theory of interpretation, where a sentence’s PF is (roughly) a representation that expresses its phonology, and its LF a representation that expresses all syntactic properties relevant to interpretation.17 This distinction brings with it the possibility of revealing in the LF of a sentence X syntactic objects that have no counterparts in X ’s PF. Such ‘aphonic’ (‘phonologically null’, ‘phonologically empty’) expressions are as much in need of interpretation when X is uttered as any other elements in X ’s LF. (If a sentence’s LF expresses only syntactic properties relevant to interpretation, as current theory dictates, this becomes a matter of deWnition.)

13. Indexicality Identifying the LF of a sentence X does not constitute identifying what A says on a given occasion by uttering X. For one thing, X ’s LF may contain an indexical expression like ‘I’ or ‘he’ or ‘that’. So identifying X ’s LF still leaves B some interpretive work to do, work that will involve accessing and integrating all sorts of information not carried or revealed by the LF itself.18

14. Anchoring Idealization and abstraction from the details of particular speech situations or contexts are unavoidable if work is to proceed. To this extent, we may temporarily avail ourselves of the formal ‘indices’ or ‘contexts’ of indexical logics in order to anchor or coanchor the interpretations of indexical or anaphoric expressions that are not of primary 17 See Chomsky (1986, 1995, 2000). 18 This does not mean that the pragmatist cannot, for certain expository or investigative purposes, operate as if a description of a sentence’s LF gives us a description of what A said relative to certain heuristic stipulations. Formal ‘contexts’ or ‘indices’ are used in logic to anchor or co-anchor indexical elements in order to cancel or pair their eVects across similar structures. Without commitment to the view that formal contexts play any sort of role in a theory of interpretation, the linguistic pragmatist may sometimes borrow this technique in order that a particular investigation may focus on particular non-indexical properties of LFs that are relevant to a theory of utterance interpretation.

82

I. Incomplete Descriptions

concern at a certain point of investigation. We should not take formal indices themselves particularly seriously, however. They are useful transitory tools, methodological or heuristic devices, not serious posits in a theory of utterance interpretation.

15. Mongrels Since LFs may contain aphonics and may contain indexicals, we should be open to the possibility that they may contain aphonic indexicals. (At the same time, we should no more take seriously the idea that pairing a formal ‘context’ with an aphonic indexical in a sentence X eo ipso solves a genuine problem about the interpretation of an utterance of X than we should take seriously the idea that pairing a formal ‘context’ with a phonic indexical does so.) Aphonic indexicals are not the only possible mongrels. Since LFs may contain aphonics and may contain bindable variables, we should be open to the possibility that they may contain aphonic bindable variables. (Compare ‘everyone wants John to leave’ and ‘everyone wants to leave’. Perhaps the subject of ‘to leave’ in the latter is an aphonic variable bound by ‘everyone’. Certainly it is not an aphonic copy of ‘everyone’.) And why not aphonic, indexical, bindable variables?

16. Isomorphism It is at least methodologically useful to say that identifying what a speaker said by uttering a sentence X on a given occasion involves entertaining a ‘sentence’ of Mentalese; or that it involves entertaining a structured proposition. (Perhaps entertaining a sentence of Mentalese ultimately amounts to entertaining a structured proposition because it involves entertaining a representation whose role in our mental life can be explained only in terms of it having a certain ‘content’ that a structured proposition supplies. Who knows?) With Sperber and Wilson, let us work for the moment with Mentalese. Since LFs are not full-blown representations of Mentalese (and so do not express propositions), but only ‘blueprints’, ‘schemas’, ‘skeletons’, or ‘templates’ for such, in advance of serious empirical investigation we cannot rule out the possibility of a failure of isomorphism in the mapping between the LF of a sentence X and a Mentalese representation the entertaining of which constitutes understanding what A said by uttering X on a particular occasion (or in the mapping between X ’s LF and a structured proposition, the entertaining of which constitutes understanding what A said by uttering X on a particular occasion). That is, we cannot rule out atoms of the Mentalese representation (or atoms of the structured proposition) to which no element of X ’s LF corresponds. (A might utter the sentence ‘The embassy is closed’ on a particular occasion and B may be required to entertain the Mentalese sentence all too conveniently rendered as the us embassy in london is closed in order to grasp what A said.19 A may utter, ‘the ham sandwich wants extra pickles’, and B may be required to 19 If you know English and your shift-key works, Mentalese is a cinch (Mentalease?). Structured-Propositionese is a little harder: you need good angled brackets.

Neale, This, That, and the Other

83

entertain the Mentalese sentence the man who just ordered a ham sandwich wants extra pickles. A may utter, ‘the hostages landed back on American soil today’ and B may be required to entertain the Mentalese sentence the former american hostages at the us embassy in tehran landed back on american soil today. It would seem that A may even utter less than a whole sentence—for example, ‘no thank you’ or ‘a cappuccino, please’—and thereby say something.) We cannot rule out the possibility, however, that future work in syntax will indicate that we are closer to isomorphism than superWcial appearances suggest, for all sorts of aphonics in LF may be revealed. Presumably, Mentalese representations will have to contain elements that function as (or at least do the work done by) bound variables, so we may well have to consider the possibility that interpreting a particular utterance of a sentence X may involve entertaining a Mentalese representation that contains a mental variable with no counterpart in X ’s LF. On the other hand, syntactic evidence might be found for the existence of an aphonic variable in X ’s LF. We cannot dogmatically assume that there must be isomorphism, and we should recoil from the unargued goal of attaining isomorphism by freely adding aphonics to LFs as if adorning some garish Christmas tree with a new light wherever it seems too dark. (We shouldn’t get hooked on aphonics.)

17. Ellipsis Corresponding to the sentence–utterance distinction impressed upon us so forcefully by Austin, Grice, and Strawson, we must take seriously two important and distinct uses of the words ‘ellipsis’ and ‘elliptical’. The Wrst is a strict linguistic (or grammatical) notion found in talk of elliptical sentences in generative linguistics, a notion sometimes called deletion and which involves erasing elements in the generation of PF representations.20 Linguistic ellipsis concerns the superWcial incompleteness of structures, and as such is subject to a stringent condition on the constancy of form and interpretation that has been investigated by linguists under the rubric of recoverability. (A can use the sentence ‘I can tango but Mary can’t’ to say that he can tango but Mary can’t tango, but not to say that he can tango but Mary can’t sing. This is because it is elliptical for the complete sentence ‘I can tango but Mary can’t tango’.) The second notion of ellipsis is a pragmatic (or speech act) notion, found in talk of elliptical utterances of (elliptical or non-elliptical) sentences.21 Pragmatic ellipsis concerns the incompleteness of interpretations, and as such is governed only by general pragmatic principles governing interpretation. (A can use the sentence ‘I’m going to a party at the embassy’ to say that he’s going to a party at the British embassy in Athens, for example, or to say that he’s going to a party at the US embassy in London, because there is no particular complete 20 See e.g. Williams (1977), Sag (1976), Heim and Kratzer (1998), and May (2002). The last of these provides a particularly clear and user-friendly discussion of linguistic ellipsis. The grammatical notion clearly has its roots in talk of ellipsis and elision in some traditional grammars. 21 See e.g. Quine (1940), Sellars (1954), Brinton (1977), Bach (1981), Salmon (1982), and Neale (1990).

84

I. Incomplete Descriptions

sentence that the sentence A uttered is an incomplete or elliptical version of.) We must accept that people often speak elliptically without much (if any) conscious eVort and that hearers interpret elliptical utterances without much (if any) conscious eVort. In such situations typically the speaker and hearer can both readily expand upon the sentence uttered in such a way that explains the ellipsis.

18. Competence Three major components of a theory of interpretation are a syntactic theory, a semantic theory and a pragmatic theory. Certain preconceptions about the labels ‘syntactic’, ‘semantic’, and ‘pragmatic’ need to be put aside if the pragmatist position is to be understood, for these words are used in very precise ways. (Self-serving edicts from those who claim to have isolated the ‘correct’ way of making the semantics–pragmatics distinction or the ‘correct’ uses of the terms ‘semantics’ and ‘pragmatics’ should be ignored.) A syntactic theory for a person A who speaks a language L is an abstract description of A’s syntactic competence (in Chomsky’s sense), A’s tacit knowledge of the syntax of L. This only becomes interesting when we are clear about what counts as a syntactic fact or phenomenon. Are binding and scope semantic phenomena? A semantic theory is an abstract description of A’s semantic competence, his knowledge, tacit or otherwise, of the semantics of L. This only becomes interesting when we are clear about what counts as a semantic fact or phenomenon. Are binding and scope semantic phenomena? Binding shows that a sharp division between syntax and semantics is illusory (which is why Chomsky is prepared to use the label ‘syntactic’ in connection with much of what many philosophers and linguists label ‘semantic’). Drawing a sharp line between semantics and pragmatics is straightforward. A pragmatic theory transcends individual speakers and particular languages. It is an abstract description of the mechanisms that make it possible for interpreters to identify what a speaker means by uttering a sentence (or sentence fragment) X on a given occasion given (at most) what a semantic theory has to say about X. As such, a pragmatic theory is a description of an intentional and richly inferential system, our common pragmatic competence. There is no assumption here nor is there any antecedent reason to suspect that this semantics–pragmatics distinction just drawn will be coordinate with the saying– implying distinction.

19. Semantics Words and the ways in which they can be combined have properties that enter into an explanation of why speakers use the particular combinations they do, and why hearers interpret speakers using these combinations in the ways they do. The two most obvious properties are the meanings of words and syntax. Qua description of semantic competence, a semantic theory for a language will explain how the syntactic structure of a sentence (or sentence fragment) X and the meanings of the individual words in X

Neale, This, That, and the Other

85

conspire to constrain what speakers can say using X. Flushing out the modal: a semantic theory for a language L will provide, for each sentence X of L, a blueprint for (a template, a schematic or skeletal representation of ) what someone will be taken to be saying when using X to say something. The blueprint associated with X is its semantics, and the set of such blueprints, one for every sentence of a language L, is the semantics for L. (The study of the these blueprints is also called semantics. The study of the role of word meanings is called lexical semantics; the study of the role of syntax is called compositional semantics.) Semantic competence comprises at least (i) knowledge of the meanings of individual words and (ii) knowledge of syntax (syntactic competence). It is a matter of debate whether it involves more. On the one hand, if A claims not to understand a sentence X, then it would seem that either the meaning of some word in X eludes him or else some aspect of X ’s structure (ultimately X ’s LF) does. On the other, having a model aeroplane kit, a foolproof set of instructions, excellent glue, plenty of space, good lighting, and the Wngers of a heart surgeon is not the same thing as having the model aeroplane (that’s why they write ‘kit’ on the box). Settling this debate involves settling (among other things) what syntactic competence amounts to and how bad the model aeroplane analogy is. (Certainly knowledge (in the requisite sense) of syntax does not amount to the propositional representation of a set of syntactic rules.)

20. Pragmatics Whereas each language (perhaps even each idiolect) has its own syntax and its own semantics—which is not to say that vital syntactic and (hence) semantic properties are not shared across languages as a result of our common biological endowment—there is, so to speak, only one pragmatics. Qua description of our shared pragmatic competence, a pragmatic theory will explain how interpreters identify what a speaker means by uttering a sentence (or sentence fragment) X on a given occasion given (at most) what a semantic theory has to say about X. The semantics–pragmatics distinction, thus construed, is not co-ordinate with the saying–implying distinction. What A means by uttering X on a given occasion comprises what A said and what A implied. So a pragmatic theory will explain how interpreters identify what A said and implied by uttering X on that occasion given (at most) what a semantic theory has to say about X. If a pragmatic theory explained only how interpreters identify what A implied given (at most) what the speaker said as ‘input’, a gaping hole in our taxonomy of theories would appear. A semantic theory speciWes the constraints that word meanings and syntax place on what A can say by uttering X, a blueprint for X. What would we call a theory that explains how interpreters identify what A said on that occasion? Not a semantic theory, for that speciWes only a blueprint for what A said, i.e. the sort of thing he said. Clearly, a pragmatic theory has two roles in a theory of interpretation. Even if an utterance of a sentence X always wore on its sleeve an unambiguous representation of its syntactic structure with no ambiguous elements,

86

I. Incomplete Descriptions

a semantic theory could still fail to identify fully what A said by uttering X on a particular occasion. For one thing, X may contain /he/, /this/, /here/ or /John/, in which case the interpreter needs to identify who or what A is referring to.22 Since these words are not ambiguous in the way /pen/ or /bank/ are said to be, and since each is not merely the unambiguous surface form of a context-insensitive deWnite description, something other than a semantic theory must be invoked.23 The slack is taken up by a pragmatic theory: identifying what A said involves the exercise of cognitive capacities that integrate the semantic information carried by the sentence uttered and all sorts of ‘pragmatic’ or ‘contextual’ information including, but 22 My wording should make it clear that I am putting aside, for now, the fact, stressed by many pragmatists, that a pragmatic theory may have to be invoked in order to identify what sentence A uttered because of ambiguities at PF. Among the things a hearer or reader has to do in order to identify what A is saying on a given occasion, is identify which words A is using. /Bank/ is the superWcial form of either a single, ambiguous word of English or else of two distinct unambiguous words, and I do not want one’s position on this matter to impinge upon one’s understanding of IT. (If /bank/ is the superWcial form of a single, ambiguous word, then identifying what A is saying when he utters /I’m going to the bank/ involves identifying which meaning A has in mind for /bank/; if ‘bank is the superWcial form of two distinct, unambiguous words then identifying what A is saying when he utters /I’m going to the bank/ involves identifying which of the two words A is using. The latter view seems more useful in theorizing about language. ‘Word’ and ‘sentence’ are quasi-technical terms, there are no ambiguous words or sentences, and (following Chomsky) every sentence comprises a superWcial form PF and an underlying form LF, the former being what is relevant to speech perception, the latter what is relevant to speech comprehension. When I wish to talk explicitly about an expression’s PF or about the sound common to two expressions, and when I wish to avoid commitment one way or the other as to whether I am talking about one expression or two, I shall borrow the old slash notation of phonology (but with standard orthography rather than a phonological representation enclosed, as in /bank/) to individuate coarsely in terms of phonological properties. (On one use, then, /bank/ is what Perry (1998) calls a vocable.) Thus I sometimes use /he/, /him/, and /his/ in my discussions of pronouns because it is arguable that each corresponds to two distinct words in English, one that is ‘bound and another that is not, an idea that appears to gain some plausibility when we consider languages like Icelandic where a related distinction is lexicalized. (On this matter, see Neale (forthcoming a).) Similarly, I sometimes use /the/ so as not prejudge the issue on the matter of a purported ambiguity in the deWnite article(s). Of course if /he/ and /the/ really are ambiguous, the ambiguity in question will have to be more systematic than the sort found with /pen/ or /bank/. It is easy enough to cause trouble for my use of the slash notation. Almost certainly we want to distinguish the phonologically identical but orthographically distinct ‘so’, ‘sew’, and ‘sow a ’ (as in seeds), distinguish the orthographically identical but phonologically distinct ‘sow a ’ (as in seeds) and ‘sow b ’ (as in pig), and distinguish the phonologically and orthographically identical ‘pena ’ (as in writing instrument) and ‘penb ’ (as in enclosure); but probably we need not bother distinguishing the (merely) orthographically distinct ‘judgment’ and ‘judgement’ or the (merely) phonologically distinct ‘co´ntroversy’ and ‘contro´versy.’ (Actually, I’m not so sure I should have said that: perhaps there are worries here not entirely unconnected to those Kripke (1979) brings up in connection with ‘Paderewski’.) So when I want to individuate coarsely in terms of (roughly) phonology, I use the slash notation. Thus /pen/ and /sew/ (and, unfortunately, /sow a / and /sow b /). If a single word can have two distinct orthographies (‘judgment’ and ‘judgement’) and a single word can have two distinct phonologies (‘co´ntroversy’ and ‘contro´versy’), should we explore the idea that a single word can have two distinct orthographies and two distinct phonologies? Or should we say that something is a single word only if it is grounded in a single phonology or a single orthography? Or should some intermediate position be explored that invokes etymology or the relative similarity of distinct orthographies and phonologies, a position according to which /doctor/ and /physician/ would be too far apart to qualify, orthographically, phonologically, and etymologically? (Notice how the notation just exploded.) Many Greek villages or islands still have two names, and the reason we talk this way is because the names seem too far apart to count as a single name (e.g. ‘Thı´ra’ and ‘Santorı´ni’). But what about ‘Apera´thou’ and ‘Apeı´ranthos’ (a village on the island of Na´xos)? 23 Three points must be separated here: (i) failure of lexical ambiguity, (ii) context sensitivity, and (iii) rigidity.

Neale, This, That, and the Other

87

not limited to, information obtained by perception from the physical environment, information about the interpretation of prior utterances in the conversation (if any), information in memory, and information about how people typically behave, particularly in communicative exchanges. That is, identifying what A said involves processing not only the semantic information encoded in a sentence’s form, but accessing and processing information that must be picked up by listening, watching, remembering, hypothesizing, and inferring, essentially the capacities exercised in identifying what A implied. To this extent, then, identifying what is said is a pragmatic as well as a semantic matter. It involves pragmatic inference as well as linguistic decoding. Identifying what a speaker implied is something explained by a pragmatic theory, typically taking into account what A said; but identifying what the speaker said is also something explained by a pragmatic theory, taking into account (in a big way, to be sure) a sentence’s blueprint, which is explained by a semantic theory. Underpinning the diVerence between identifying what A said and what he implied is a distinction in the type of typical input: to identify what A said on a given occasion by uttering X the pragmatic system typically takes as its primary input the output of the semantic system (the semantic information encoded in X ); to identify what a speaker implied takes as its primary input what the speaker said. This leaves many questions open. (1) To what extent is pragmatic processing deductive? (2) To what extent does it take place unconsciously? (3) What sorts of things aVect its speed? (4) To what extent is it task-speciWc or modular?

21. Underdetermination It is now possible to bring together several points. The role of a pragmatic theory in identifying what A said by uttering X on a given occasion is not restricted to identifying who or what A is referring to by any referential expressions in X. Saying involves referring and predicating; and just as identifying what A is up to with any referential devices in X involves more than consulting a mental lexicon, so does identifying what he is up to with any predicative devices in X. It may, for example, require the ‘saturation’ of an ‘implicit argument’ as in ‘It’s raining’ or ‘I’ve Wnished’. (Some implicit arguments may be mandated by syntax as well as semantics.) Or it may require ‘enriching’ a predicate in some way that is reasonably obvious and presumably acceptable to A. The sentence ‘every woman has a job’ might be used to say that every woman in Flint has a job, or that every woman in Woodside has a job, etc. It will pay to separate and rejoin two points here, one epistemological, the other metaphysical, both intimately connected to the points made earlier about Asymmetry and Reciprocity. The epistemological point concerns insuYciency, the metaphysical point underdetermination. From the hearer’s perspective, we can talk Wrst about the fact that knowledge of the syntax of X and knowledge of the meanings of all the words in X do not suYce for identifying what A is saying by uttering X (even where the superWcial form evinces no lexical or structural ambiguity). At most they yield a

88

I. Incomplete Descriptions

blueprint. Now we can bring in A himself. What A says is wholly determined by certain speciWc intentions he had in speaking, intentions massively constrained by his knowledge of syntax and word meaning (and a whole lot more). A tacitly knows that B’s knowledge of word meaning and syntax will not suYce to furnish B with a complete account of what he has said. We can now introduce some theoretical shorthand to obviate the need to keep talking about speakers’ and hearers’ knowledge, tacit or otherwise. Let us say that syntax and word meaning together underdetermine what is said (all the time remembering this is shorthand). But we are not yet where we need to be. The Wrst thesis we need is this: (IT) The InsuYciency Thesis: Identifying what a speaker or writer, A, is saying by uttering an unambiguous, declarative sentence X on a given occasion goes well beyond recovering X ’s underlying syntax, knowing the meanings of all of the words in X, and identifying who or what A is referring to by any referential expressions in X. This goes beyond the insuYciency just mentioned because it entails that even when B has identiWed who A is referring to by any referential expressions in X (/John/, /he/, /here/, /that/, /I/, /you/, and so on), B still doesn’t have everything he needs to identify what A said. Since A tacitly knows that B’s knowledge of word meaning, knowledge of syntax, and knowledge of who or what A is referring to by any referring expressions in X will not suYce to furnish B with a complete account of what he has said, we can formulate the shorthand we really want: (UT) The Underdetermination Thesis: What A says by uttering an unambiguous, declarative sentence X on a given occasion is underdetermined by X ’s syntax, the meanings of the words (and any other morphemes) in X (and the meanings, if any, of prosodic features of X), and the assignment of references to any referring expressions in X.24 24 The Underdetermination Thesis is regularly stressed by linguistic pragmatists, and some view it as a cornerstone of the general outlook. This use of ‘underdetermination’ is found in the work of Sperber and Wilson (1986/1995) and borrowed by many of those they have inXuenced, including Bezuidenhout, Blakemore, Carston, Papafragou, Recanati, Rouchota, and me (Descriptions, 1990: 114 n. 46). In the language of Perry (1986, 1993, 1998, 2001), talk of underdetermination is roughly equivalent to talk of constituents of propositions expressed that are ‘unarticulated’, i.e. constituents corresponding to no constituents of the sentence uttered. (As Perry sometimes puts it, we don’t always articulate things when it’s clear from context what they are.) I don’t know why underdetermination is rejected so vehemently by some philosophers of language, but I have a suspicion two related factors may be implicated, both involving fear and philosophical temperament. The Wrst is a simple unwillingness to concede apparently hard-earned territory, a reluctance to accept that some of the traditional problems involved in so-called ‘compositional semantics’ are actually the products of specious questions in the philosophy of language with genuine and important counterparts in the philosophy of mind, mostly about inference and the composition of thought (a reluctance to accept that, as far as natural language is concerned, trying to build pure content is as futile as trying to build pure character). The reluctance, it seems to me, amounts to little more than obstinacy or fear of a philosophical pink slip. The second fear might be viewed as an extension of the Wrst: the fear that, if there is not at least one component (what is said) of what a speaker means that can be nailed down precisely and completely without taking into account too many ‘pragmatic considerations’, then systematic semantics as typically understood is doomed, and with it any chance of produ-

Neale, This, That, and the Other

89

As far as constructing and evaluating a theory of interpretation are concerned, we must make sure we separate talk of (a) the interpretive target (stipulated in advance by the theorist in particular cases on the basis of intuitive judgment, modulo reXective equilibrium with the best theory up to that point), and talk of (b) the knowledge and mechanisms in play (under investigation and hypothesized by the theorist). And in talk of knowledge and mechanisms we must be careful to separate (i) the role of syntax; (ii) the role of word meaning; and (iii) the actual pragmatic mechanisms. An interpretive target is a characterization of what our intuitive judgments reveal the speaker to have said and which an adequate theory of interpretation should deliver. It is common to specify interpretive targets using more language: By uttering ‘she has a job’ A is saying that Margaret Thatcher has a job; by uttering ‘every woman has a job’, A is saying that every woman living in Woodside has a job. There is, of course, something a bit funny about this, for surely we can now ask for a characterization of what the theorist said when he uttered the sentence ‘By uttering ‘‘every woman has a job’’, A is saying that every woman living in Woodside has a job’. Nonetheless, this is what we do, and when pressed often we wheel out some set theory: by uttering ‘every woman has a job’, A is saying something that is true iV at the time of utterance t, the intersection of the set of things that are women at t and the set of things living in Woodside at t is a subset of the set of things that have jobs at t. While this may provide us with the conditions under which what A said is true, it falls short of specifying what A said for familiar reasons. (First, A is sure to deny it. Second, surely A would have said something diVerent had he uttered, ‘every woman has a job and 192 ¼ 3610 .) So when pressed again we wheel out something like a situation or a structured proposition. Where knowledge of syntax, word meaning, and the theory of blueprints are concerned, there is much work to be done by philosophers of language and linguists together. As far as mechanisms are concerned, we are squarely in the realm of psychology, and some philosophers of language and linguists may well opt out. The psychological part of the overall project may certainly be informed by philosophical reXections, such as Grice’s, on the nature of rational, purposive behaviour, but ultimately it is a wholly empirical enterprise the aim of which is to identify the cognitive mechanisms whereby the hearer eVects the relevant identiWcations on a given occasion.

22. Indeterminacy What a speaker says and implies may be indeterminate in at least the following sense: in any vocabulary in which what someone says or implies can be usefully speciWed, there will be alternative and strictly distinct speciWcations between which no principled choice can be made. We should not worry that indeterminacy of this sort presents problems for particular semantic proposals. (For example, we should not regard cing a serious theory of language. This fear seems to me entirely unwarranted. Natural language semantics may not be quite as straightforward or far-reaching as many have thought, but there is plenty of systematic semantics for all of us (and more) to do for longer than we will ever have to do it.

90

I. Incomplete Descriptions

traditional accounts of descriptions as damaged in any way by the indeterminacy attaching to intuitive ‘completions’ of those that are said to be ‘incomplete’ relative to a particular occasion of utterance.)

23. Convergence A univocal saying–implying distinction is empirical, ordinary, and beneWcial (practical). The distinction is empirical insofar as it assumes that, typically, representations corresponding to what A said and implied are the outputs of cognitive mechanisms involved in the interpretation process. It corresponds to something entrenched in ordinary talk (despite the fact that we may disagree in particular cases). And it underpins the very idea of codifying principles meant to regulate societies and the behaviour of their members (e.g. laws, contracts, and commitments) by virtue of being a distinction one side of which (saying) is about as objective as anything can be, a fact itself guaranteed by the empirical and ordinary nature of the distinction. (To say this is not to say there cannot be disputes about what was said, changes of opinion after discussion of problematic cases, or specialists (or at least professionals) in societies to whom tough cases are referred when the issue needs forcing. Rather it is to say that there is enough overlap in judgment to render regulation, commitment, and so on, meaningful notions.)

24. Formalism Advances in our thinking about language have come out of developments in logic and formal philosophy, particularly by way of the construction and use of various types of broadly mathematical theories, systems, or analyses—the predicate calculus, model theory, modal logics, set theory, recursion theory, and generalized quantiWer theory to name the most obvious. But it does not follow that associating utterances with models, possible worlds, structured propositions, indices, functions, or even favoured formalisms ipso facto constitutes part of a theory of utterance interpretation. Rigorous formalism almost certainly has its place; but a favoured mathematical idea and an associated formalism must not so dominate our inquiry that the questions motivating it in the Wrst place become obscured or transmogriWed to the point of demanding purely technical answers. We should strive to use our formalisms judiciously, sparingly, only where they are needed to eVect a useful idealization or abstraction, forestall a potential ambiguity, capture a generalization, facilitate a transition, or usefully abbreviate something. Appeals to, say, higher-order functions or set-theoretic entities (and the use of corresponding notations) are ultimately dispensable in the theory of interpretation, sets and functions being no more than occasional, transitory tools of no intrinsic interest outside mathematics proper and the philosophy thereof. Such entities are not the objects of semantic investigation themselves, and it should not be a goal of any branch of philosophy to drag them into investigations whenever the opportunity presents itself.

Neale, This, That, and the Other

91

3. The Irrelevance of Contexts Following Grice, let us talk about what a speaker A means on a given occasion by uttering some sentence X, factoring this into what A says and what he (merely) implies. As Grice notes, identifying what the speaker is saying is not simply a matter of identifying X and recovering its linguistic meaning (blueprint), if only because of the existence of pronouns. Unlike some of his critics, Grice is careful not to run together epistemological and metaphysical points here, despite their evident interconnectedness. The important metaphysical question is: what determines what a speaker said on a given occasion? And the Gricean answer is: certain speciWc intentions he had in producing his utterance, intentions that are severely constrained by his beliefs about the meanings of the words he uses, about his audience, about the context, about the topic of conversation, and probably a whole lot more.25 The important epistemological question is: what knowledge or information does a hearer use in identifying what the speaker said? And the Gricean answer is: knowledge of the linguistic meaning of the sentence uttered, pragmatic knowledge about the way rational, cooperative beings operate, knowledge about the speaker, knowledge of context, and just about anything else he can get his hands on. Let us take a concrete example. If I say something by uttering a sentence X that contains the personal pronoun ‘he’ and the demonstrative pronoun ‘this’, then (in the simplest case, at any rate) my referential intentions determine who I mean by ‘he’ and what I meant by ‘this’. (Similarly, my lexical intentions determine what I meant by /bank/ if X contains one of the words we write that way.) Your job as hearer is to identify what I meant by uttering X, and very likely you will not succeed unless you identify who I meant by ‘he’ and what I meant by ‘this’. (Similarly, what I meant by /bank/.) Sperber and Wilson (1986/1995) point out that pronouns are just the tip of a pragmatic iceberg for traditional, titanic accounts of what is said: quite generally, what a speaker says is underdetermined by the meaning of the sentence uttered, even relative to reference assignment. Now one can perfectly consistently accept Sperber and Wilson’s underdetermination thesis without rushing to embrace the details of their Relevance Theory. For that theory is meant to provide an account of the mechanics of utterance interpretation, of the richly inferential processes providing the basis of an empirically satisfying account of how interpreters (i) identify which sentence a speaker has produced on a given occasion in cases where identiWcation of phonological form fails to yield a unique result; (ii) identify what the speaker said by uttering X on a given occasion in cases where identiWcation of the meaning of X falls short; and (iii) identify what the speaker implied by uttering X on that occasion. Relevance Theory goes well beyond accepting that the meaning of a sentence X may underdetermine what a speaker says by uttering it on a given occasion and well beyond the vague Gricean idea 25 See Grice (1957, 1969, 1989), Neale (1992, forthcoming b). Very roughly, one cannot intend what one believes to be impossible.

92

I. Incomplete Descriptions

that quite general principles governing the way we reason about the behaviour of others lie at the heart of an explanation of how we communicate. In principle just about any information could be relevant or brought to bear on interpretation, and one of the main problems involved in constructing a pragmatic theory is explaining how the information that actually is brought to bear is delimited.26 The second problem concerns how it is brought to bear. Linguistic pragmatism Wnds little sense to the idea that two quite distinct sets of information-gathering and inferential mechanisms are at work when a hearer tries to identify what a speaker means, one set that works on sentence meanings and yields what the speaker said, and another set that works on what the speaker said and yields what he meant but did not say (i.e. what he implies). It is odd that some philosophers write as if (or even claim that) two quite distinct sets of cognitive mechanisms must be at work. I detect two related ideas lurking behind this assumption: (i) something to do with ‘simplicity’ (or ‘degree of diYculty’) or ‘systematicity’ (or ‘range of possibilities’); (ii) the inXuence of ‘indexical logics’. (i) Whilst it is true that identifying who or what a speaker intends to be referring to on a given occasion by some particular referring expression X is constrained by the linguistic conventions governing the use of X, this does not necessarily make matters particularly straightforward or reduce the number of hypotheses that could, in principle, be investigated and assessed. Consider the interpretation of an utterance of the pronoun ‘it’. As Sperber and Wilson (1986: 187) note, all that the linguistic conventions governing the pronoun ‘it’ insist upon, in any context, is that the object should be non-human, giving every hearer in every context an indeWnitely large choice of possible referents. And surely the same general considerations about, say, relevance, truthfulness, informativeness, or whatever, that are invoked in identifying what a speaker is implying on a given occasion will be invoked in identifying who or what a speaker is referring to by ‘it’ on a given occasion. The point can also be made in connection with an incomplete description like ‘the table’ or some other incomplete quantiWer expression like ‘every man’ or ‘no one’. It is sometimes said that identifying what A says by uttering a sentence containing such an expression involves either (a) coming up with an appropriate domain of quantiWcation implicit in the utterance (the ‘implicit’ approach), or (b) coming up with an appropriately ‘richer’ nominal A could have used to make his meaning more explicit (the ‘explicit’ approach). Whatever the Wnal merits of such suggestions, one thing is quite clear: the same general considerations about, say, relevance, truthfulness, informativeness, or whatever, that are invoked in identifying what a speaker is implying on a given

26 There is a division here between the optimists and the pessimists. Unlike Sperber and Wilson, Blakemore (1987), Carston (2002), and other Relevance theorists, Fodor (1983, 2001), Chomsky (2000), and Davidson (1986) suspect that producing an overarching theory of interpretation will require nothing short of a complete theory of mind.

Neale, This, That, and the Other

93

occasion will have to be invoked in identifying an appropriate completing domain or an appropriate completing expression.27 (ii) Many philosophers write as if (or even argue that) understanding what a speaker A said on a given occasion by uttering a sentence X with its conventional meaning is a matter determined by the meaning of that sentence and a ‘context’, in a sense of this frequently invoked word that is meant to make it more than simply a label for whatever it is that ‘bridges the gap’ between the meaning of X and what A said by uttering X on that occasion. For example, it is frequently claimed that all one needs to bridge the gap is some sort of formal object, an ‘index’ or ‘context’ in the form of an ordered n-tuple that secures the references of a few annoying ‘indexical’ pronouns (‘I’, ‘you’, and ‘he’, for example) and one or two other ‘indexical’ words that have a somewhat pronominal nature (‘here’ and ‘now’, for example).28 This idea is rightly spurned in Sperber and Wilson’s Relevance, spurned in Evans’s The Varieties of Reference and Collected Papers, and spurned in Chomsky’s ReXections on Language—indeed spurned in every book or article that Chomsky has ever written in which the interpretation of pronouns is discussed. For there is an implicit recognition in these works, and in many others that bear their inXuence, that whilst formal contexts may have a useful methodological role from time to time, they are strictly irrelevant to a proper theory of utterance interpretation. For various semantic and syntactic purposes, it is often desirable—if not mandatory—to abstract or idealize away from facts to do with particular speech situations— ‘pragmatic’ or ‘contextual’ factors, as they are sometimes called—in order to get on with a particular piece of work. And as long as caution is exercised there is no harm in this. For example, with certain restricted purposes in mind—and without any sort of absurd commitment to the idea that such entities play a role in utterance interpretation—formal ‘indices’ can be introduced to serve as ‘contexts’ with which sentences can be paired in order to ‘anchor’ or ‘co-anchor’ the interpretations of certain indexical expressions. The usual idea is to treat such expressions as free variables and treat indices as sequences or functions that assign them values. Famously, this idea has been used to capture model-theoretically the validity of inferences whose premises and conclusions are stated using indexical sentences:29

27 Postulating an aphonic indexical, domain variable in underlying syntax makes no more of a contribution to explaining how hearers interpret utterances than does postulating an aphonic indexical assertion variable in underlying syntax (or an aphonic indexical irony variable). That is, the interpretive task facing the hearer is made no easier by the existence of an aphonic contextual variable, even when, as in the case of the supposed assertion or irony variables, there are just two possible values to choose from. And of course, giving phonetic form to such operators—‘asserting-or-not, it’s Tuesday’ or ‘being-ironical-or-not, it’s a lovely day’—doesn’t help the hearer either. 28 The word ‘indexical’ is itself part of the problem, suggesting as it does that interpreting such devices involves merely looking something up in an ‘index’. People can be more inXuenced by labels than they sometimes realize. 29 See (e.g.) Kaplan (1989a), Lewis (1975), Montague (1974a). A lot in this area turns on one’s conception of logic, and my wording evinces a particular stance, though not one I want to insist on: logical relations

94

I. Incomplete Descriptions

A: If the next left is not Bank Street, that man gave you the wrong directions. B: It’s not Bank Street; so he gave me the wrong directions.30

It is paramount in such work to keep things tightly under control in the following sense: the logician wants a mechanism that can (a) scan a set of sentences for occurrences of symbols on some pre-existing list of devices that do not carry their values with them, then (b) use an index to assign a value to each occurrence of such a symbol. If this goes well, logical deductions can proceed (assuming a semantics for items of a preselected ‘logical’ vocabulary of course). If there is still slippage after the index has made its assignments, on standard assumptions there is only one solution: posit further indexical symbols in the sentences involved, symbols which are invisible in surface syntax yet revealed by an analysis of their ‘logical forms’, then try again.31 In the philosophy of language, indices have a methodological role for they can be used to anchor or co-anchor indexical and anaphoric expressions and so allow work to proceed more easily on other expressions and on what people say (and imply for that matter) by uttering them on given occasions.32 However, there is an idea that has emerged from work on indexical logics for which we can have little sympathy. This is the idea that sentence meanings and contexts can be paired to provide something of empirical signiWcance: what a sentence X says relative to a context C.33 We must not lose hold among what is expressed by sentences not among sentences themselves. (Various issues about the notion of formal validity and inference rule must be faced (but usually are not) by people who hold this view of logic.) The point I am making in the text is not dependent upon this stance. Cf. discussions of the diVerence between the logical form of a proposition and the logical form of a particular sentence used to express that proposition. 30 A related point might be made in connection with anaphora: (i) every man loves his mother; (ii) John is a man; therefore (iii) John loves his mother. For discussion, see Neale (forthcoming a). 31 It is, perhaps, tacit recognition of this fact that has led some philosophers to conclude that there is no hope of producing a theory of utterance interpretation without positing all sorts of phonetically null, indexical elements in the underlying syntax of natural language sentences. We may use anything we like to throw light on the syntax of natural language, but we must never lose sight of the fact that discerning the syntactic structures of our sentences is an empirical exercise. Certainly the idea of aphonic elements in syntax is not objectionable in itself. On the assumption that syntax relates sound and meaning, we must certainly allow for the possibility of elements that have sound but no meaning (‘it’ in ‘it’s raining’?), or meaning but no sound (the understood subject of ‘leave’ in ‘Tom wants to leave’?). And there can be little doubt today that great advances in our understanding of syntax have been made by those such as Chomsky who have not shied away from the idea of aphonic items in syntax and argued for their existence and explanatory value. But we cannot simply assume that whenever we encounter some feature of what is said that does not appear to correspond to any element or feature of the sentence uttered it means there is some element in underlying syntax waiting to be exposed. 32 e.g. see Descriptions, ch 3. 33 I am putting aside here some very real concerns about talk of sentences saying things relative to contexts. I am sceptical about the value or relevance of the use of the verb ‘say’ assumed in this way of talking to the project of constructing a theory of utterance interpretation, unless it is understood as a stylistic variant of talk of speakers saying things by uttering sentences on given occasions. Judgments about what a speaker said, and about whether what he said was true or false in speciWed situations, constitute the primary data for a theory of interpretation, the data it is the business of such a theory to explain. What a speaker says and what he implies (e.g. conversationally implicates) on a given occasion are the things that together constitute what the speaker means, and a theory of interpretation is meant to explain the role of linguistic meaning and inference in the hearer’s identiWcation of what the speaker meant. No one has intuitions about what is said by a sentence X relative to a context C or about

Neale, This, That, and the Other

95

sight of certain facts. First, as far as utterance interpretation is concerned, such ‘contexts’ are strictly irrelevant. Utterances do not come with such devices attached that anchor or co-anchor indexical, demonstrative, or anaphoric pronouns. The hearer has plenty of pragmatic work to do, much of it rightly called inferential, albeit inferential in a way that is steered by the meanings of individual words. Evans (1982, 1985) summarizes the situation well: All that the conventions governing the referring expression ‘he’ insist upon, in any given context, is that the object referred to should be male. (1982: 312) There is no linguistic rule which determines that a ‘he’ or a ‘that man’ refers to x rather than y in the vicinity, or that it refers to someone who has just left rather than someone who has been recently mentioned. (1985: 230–1) ‘This’ and ‘that’ are even less speciWc, contributing merely the vaguest suggestion of a contrast between nearer and further (in some generalised sense) . . . [Footnote: Often the predicate does more to narrow down the range of possible interpretations of the referring expression than does the referring expression itself . . . ] (1982: 312) Let me take another example: the expression ‘you’: If a speaker addresses a remark to someone, saying, ‘You are a crook’, it is surely clear that an identiWcation is called for on the part of the audience: in order to understand the remark, it is not enough to know that there is one, and only one, person whom the speaker is addressing, and that the speaker is saying of that person that he is a crook . . . a quite speciWc kind of identiWcation is called for; the person addressed has not understood the remark unless he realizes that the speaker is saying that he is a crook . . . understanding the remark requires the hearer to know of an individual that he is being addressed. (1982: 314)

Nothing about the meaning of the word ‘you’ tells you that you are being addressed.34 We need, then, to distinguish two ideas, one sensible, the other silly. The silly idea is that utterances come with pre-packaged ‘contexts’ that provide values for indexical expressions. The sensible idea is what I call methodological anchoring (anchoring for short). For various pragmatic, semantic, and syntactic purposes, it is often helpful, perhaps even mandatory, for a theorist to abstract from certain ‘contextual eVects’ or the truth or falsity of X relative to C unless this is just a formal way of talking about what the speaker said by uttering X on a particular occasion—the occasion that C is being used to partially model. If such talk is straightforwardly transposable into talk about what the speaker said then we can accept its empirical signiWcance. If it is not so transposable, then its empirical signiWcance must be justiWed in some other way, from within the theory of interpretation, by reference to some empirical role it is required to play in an explanation of what a speaker says and implies by uttering X on a given occasion, in much the same way that notions such as LF (‘Logical Form’), scope, and binding are motivated from within. If some such motivation is forthcoming, we should be only too happy to listen. I suspect it will not be forthcoming because the notion of what a sentence says relative to a context is going to be too thin and overly detached from speakers’ communicative intentions to carry any empirical weight. Nonetheless, I adopt a wait-and-see approach. We are involved in an empirical enterprise after all. 34 As soon as we introduce anaphoric pronouns—those that are linked in some interpretative fashion to other expressions (their ‘antecedents’)—matters become more complicated. The reXexive ‘himself ’ must be interpreted via an antecedent; the non-reXexives ‘he’, ‘him’, and ‘his’ can be so interpreted (under certain conditions). Very roughly, reXexives cannot be ‘too far away’ from their antecedents, and non-reXexives cannot be ‘too close’ to them, putting the two in virtual complementary distribution as far as interpretive dependence is concerned, as suggested by Chomsky’s (1981, 1986) Binding Theory. For discussion within the present framework, see Neale (forthcoming a).

96

I. Incomplete Descriptions

‘pragmatic factors’ in order to get on with a piece of work, and so it is sometimes useful to use an ‘index’ as a way of anchoring the interpretations of indexical expressions that are not, at that moment, the objects of primary concern, even though the theorist knows the interpretation of these indexicals is not as straightforward as invoking an index might suggest. If one is working on deWnite descriptions, for example, one might want to prescind, as much as possible, from the eVects of, say, indexical pronouns occurring inside nominals; and if one is working on ‘and’, for example, one might want to prescind, as much as possible, from the eVects of, say, indexical pronouns occurring inside conjuncts:35 (1) He drove home and he drank those six beers you bought him (2) He drank those six beers you bought him and he drove home. To this end, we might use an index to anchor or co-anchor these expressions, to keep their special features and the complexities they introduce out of the picture as it were.36 A certain amount of care is needed in the use of the word ‘semantic’ when indices are used to anchor (or co-anchor) indexical expressions. To the extent that we are investigating the conventions governing a word whose role cannot be set out clearly without taking into account the conventions governing other expression(s) with which it combines to form larger expressions, we may Wnd it convenient to talk about the (derived) conventions governing the larger phrases with respect to a particular index. For example, if the semantics of ‘the’, is being investigated, it may be useful, even mandatory, to anchor indexicals so that other contextual eVects may be monitored. And although we may want to talk about the ‘linguistic meaning’ of, the ‘semantics’ of, the ‘conventions governing’ an indexical or any other expression, we may also wish to talk about its ‘semantic value’ relative to a particular index, the object conveniently assigned to it by an index in order that work on pressing matters is not held up needlessly.37 There is no harm in such talk as long as everyone is clear about what is going on. ‘Semantic values’, in this sense, are just stipulated interpretations, and the anchoring it involves is quite consistent with the idea that the interpretation of indexical expressions is basically a pragmatic matter only steered by semantic constraints. Although I have not seen the point discussed explicitly in the literature, I get the impression some ‘anti-pragmatist’ sentiment may have as its underlying source the worry that the sort of pragmatism inspired by Sperber and Wilson involves selfsuspension, a willingness to abstract from contextual eVects in ways that are 35 The following examples are due to Deirdre Wilson. 36 Carston (1988, 1993, 2002) implicitly anchors in her examinations of ‘pragmatic enrichments’ in connection with utterances of conjunctions (indeed, it is what she implicitly does throughout). Similarly, Evans implicitly anchors in The Varieties of Reference (and elsewhere), Sperber and Wilson do it throughout Relevance (and elsewhere), and I do it explicitly in ch. 3 of Descriptions in connection with the eVects of indexicals appearing in deWnite descriptions such as ‘the Wrst person I saw this morning’, ‘my mother’, ‘the present king of France’, and ‘the girl who made this’. Chomsky also does something analogous to anchoring in every work in which he discusses pronouns. (I say ‘analogous’ because of Chomsky’s concerns about reference.) 37 This convenience is employed time and again in ch. 3 of Descriptions.

Neale, This, That, and the Other

97

self-defeating or paradoxical. In reality, the situation is not that diVerent from Neurath’s. The pragmatist certainly has to tread carefully, all the while monitoring for and then abstracting from aspects of what is said that are not Wxed by syntax and word meaning, and as a matter of working practice, subtle and silent measures are usually taken to prevent things becoming unmanageable, measures that certainly narrow the pragmatist’s options on the vexed matter of the relation between linguistic structure and the structure of thought.38 Pragmatist abstractions from context are always going to be juggling acts, the artistry of which is rather like that involved in solving for several variables at once whilst looking for an unknown number of others that are not yet in the equation and cannot be located without extremely good approximate values for those that are. On the basis of perceived use, intuition, discussions with friends, books we have read, and who knows what else, we isolate what we take to be the ‘linguistic meaning’ or ‘semantics’ of an expression a—its invariant role in determining what someone says by uttering sentences containing it—not unreasonably conWdent that certain things speakers mean when they use a in their speech and writing are explicable in very general terms as things they only imply, and are not of a gravity suYcient to make us question the meaning we think we have isolated. Holding the meaning of a constant, we go on to investigate and isolate the meaning of b and Wnd we can make some headway by appealing to ‘facts’ about the meaning of a. And so on, until we get further up the alphabet and Wnd our best attempts at meaning isolation force us seriously to question whether we were really right about all of a, b, g, etc. and to reexamine our methods of abstraction and idealization, which may have led us to oversimplify in ways we now worry about. This is basically the position Grice and others have found themselves in with the natural language counterparts of the logical particles, and certainly it takes some skill to keep all of the balls in the air in a stable conWguration.39

4. Elliptical Speech Among the numerous overlapping and interconnected uses of the words ‘elliptical’ and ‘ellipsis’ we can, I think, isolate and make precise three of central concern to the philosophy of language, one corresponding to each of (a) what A implied on a given occasion, (b) what A said, and (c) the sentence A used. We can call these (a) conversational ellipsis, (b) utterance ellipsis, and (c) sentence ellipsis, respectively.40

38 If there is a worry about pragmatism, it surely resides here and not in minutiae like relativization (implicit binding). 39 See, in particular, Carston’s (1988, 1993, 2002) successively reWned analyses of the meaning and use of ‘and’. 40 Cf. Gk. elleipsis, lit. a condition of falling short, from the verb elleipein: to leave in, leave out, or fall short.

98

I. Incomplete Descriptions

(a) Conversational Ellipsis (Elliptical Remarks) When asked by a colleague for his opinion of one of his students, a professor says, ‘Smith has wonderful handwriting and is always punctual.’ The example, due to Grice (1961), is meant to illustrate the diVerence between saying something and merely implying it (‘conversationally implicating’ it, as he puts it in later work). People unfamiliar with philosophical discussion in this area might say that Smith was ‘damned with faint praise’ or that the professor chose a rather ‘indirect’ or ‘elliptical’ or ‘circuitous’ or ‘roundabout’ way of letting his colleague know what he thought of Smith. The thought behind these forms of words is that the professor has avoided using certain words, and the beauty of the word ‘elliptical’ here is the connotation not only of something omitted or avoided but also of a path that curves, a path that is ‘circuitous’ or ‘roundabout’, and hence longer and less direct. (Sometimes speech or writing is also said to be elliptical when it so concise or compressed as to be diYcult to understand, an idea clearly related to omission again.) Conversational ellipsis is very common in everyday speech and writing, and only a full-Xedged pragmatic theory will throw any light on how we manage to get away with it. There is a modal dimension to conversational ellipsis: it is usually possible to Wnd some alternative form of words the speaker or writer could have used to make his point more directly or less elliptically. Indeed the relevant form of words is often used, after the fact, to specify what the speaker or writer was implying. By saying ‘he has wonderful handwriting and is always punctual’, the professor means that Smith is no good at philosophy.41 But, as Grice stressed, there may be a considerable degree of indeterminacy—did the professor really mean that Smith is no good at philosophy rather than, say, Smith is not particularly bright or Smith does not deserve a fellowship for next year?

(b) Utterance Ellipsis (Elliptical Utterances) When discussing deWnite descriptions in his 1940 book Mathematical Logic, W. V. Quine says, ‘Everyday use of descriptions is often elliptical, essential parts of the condition ‘‘ . . . x . . . ’’ being left understood; thus we may say simply ‘‘the yellow house’’ . . . when what is to be understood is rather ‘‘the yellow house in the third block of Lee Street, Tulsa’’ ’ (1940: 146). Obviously Quine is not here claiming that someone who uses ‘the yellow house’ is not using a whole noun-phrase; and he is not claiming that someone who uses (1)

the yellow house is on Wre

41 As I have argued elsewhere, Grice’s idea that a conversational implicature is ‘cancellable’ is best described modally (in the same situation, the speaker could have made an utterance that is a continuation of the utterance he did make, by which he would not have conversationally implicated what he did in fact implicate, e.g. ‘Smith has wonderful handwriting and is always punctual; what’s more, he’s erudite, one of the best philosophers of his generation’).

Neale, This, That, and the Other

99

is not using a whole sentence. It would be absurdly uncharitable to attribute such views to anyone, and here they would, in any event, Xy in the face of the text: Quine, ever picky about his words, is careful to say that everyday use of descriptions is elliptical, and not that everyday descriptions (the expressions themselves) are elliptical. What Quine seeks to get across can be captured, without remainder, by saying that someone may speak elliptically using a perfectly well-formed description.42 The point seems obvious, and I suspect Quine saw himself as expressing something of a platitude: the linguistic expression ‘the yellow house’ is not itself elliptical, but everyday uses are: since there is more than one yellow house, understanding an utterance of (1) will involve recovering more than the lexical meanings of ‘the’, ‘yellow’, ‘house’, ‘is’, ‘on’, and ‘Wre’, and projecting these in some way in accordance with the way they are here syntactically combined. The basic point seems undeniable. In a similar vein, Wilfrid Sellars in his 1954 article, ‘Presupposing’, says that, ‘a given utterance of [‘‘The table is large’’] is elliptical and states what would be nonelliptically stated, for example, by ‘‘The table over here is large’’ . . . the context functions to give the statement the force, for example, of ‘‘The table over here is large’’ ’ (1954: 199).43 Obviously Sellars is not here claiming that someone who uses ‘the table’ is not using a whole noun-phrase; and he is not claiming that someone who uses (2)

the table is large

is not using a whole sentence. It would be absurdly uncharitable to attribute such views to anyone, and here they would, in any event, Xy in the face of the text: Sellars, ever picky about his words, is careful to say that a given utterance of (2) may be elliptical, not that the sentence (2) itself is. Sellars is surely making the same point as Quine: although the linguistic expression ‘the table’ is not itself elliptical, one may use it elliptically in the sense that understanding an utterance of a sentence that contains it will typically involve recovering more than the lexical meanings of ‘the’, ‘table’, ‘is’, and ‘large’, and projecting these in some way in accordance with the way they are here syntactically combined. Again, the basic point seems incontrovertible. And in stating the point 42 The noun ‘use’ is deployed in several distinct but related ways in the philosophy of language. Sometimes it is used in the sense of ‘type of use’ or ‘way of using’, as e.g. when there is talk of the bound and free uses of pronouns, of the attributive and referential uses of deWnite descriptions, of Strawson’s uniquely referring use and of the generic (‘the whale is a mammal’) use of descriptions. At other times it is used in the sense of ‘instance of a way of using’, as e.g. when there is talk of a particular bound use of a pronoun, a particular referential use of a description, a particular uniquely referring use of a description. (Perhaps these are not diVerent uses in any very interesting sense, just examples of polysemy. Can we usefully think of particular elliptical uses of descriptions as instances of the elliptical use of descriptions? Surely there is nothing to gain by so-doing. Some have argued there is nothing to gain by analogously typing particular referential uses of descriptions either at least as far as semantic theory is concerned. Strawson himself takes advantage of another use of ‘use’: if two men utter the sentence ‘the king of France is wise’, one in the reign of Louis XV, the other in the reign of Louis XIV, ‘each made a diVerent use of the same sentence’; if two men utter the sentence simultaneously in the reign of, say, Louis XIV, ‘[they] made the same use of the same sentence’ (1950: 325–6). For discussion, see Neale (forthcoming c). 43 See also Reichenbach (1947). Talking of the description ‘the train’ as it occurs in ‘the train will arrive at 7 P.M.’ Reichenbach says, ‘The necessary addition then is understood. It usually consists in a reference to a preceding utterance; for instance, it may be assumed in the form ‘‘the train of which we spoke’’’ (1947: 258).

100

I. Incomplete Descriptions

Quine and Sellars very rightly steer away from talking about elliptical expressions per se and introduce their respective speech act words: ‘use’ and ‘utterance’.44 These guys were no slouches. There is no diVerence of any importance to any point I shall make here between Quine’s use of ‘use’ and Sellars’s use of ‘utterance’ in the passages quoted above; in what follows I shall use Sellars’s terminology (as I did in Descriptions). So when Quine says that ‘Everyday use of descriptions is often elliptical’, we can transpose this into Sellars’s terminology by construing Quine as saying, ‘Everyday utterances of descriptions are often elliptical.’45 Again, there is a modal dimension, which Sellars makes explicit in the passage just quoted: the utterance states ‘what would be nonelliptically stated’ by an utterance of some longer sentence. In Descriptions, I called this the explicit response to the problem of so-called incomplete descriptions (it is sometimes called the ellipsis response in the literature). The connotation I had in mind was the possibility of producing a lengthier utterance that was more explicit, i.e. one that more explicitly speciWed what the speaker was saying.46 According to the explicit response, utterances of, say, ‘the table’ (or ‘every table’) are elliptical for utterances of richer (‘more explicit’) descriptions, as suggested by Quine and Sellars. (I took Sellars’s discussion, with its explicitly modal characterization, as the locus classicus of the explicit response, although I did also mention Quine’s 44 See also Bach (1981, 1987, 1994). The distinction was particularly important to Sellars, who was responding, in part, to Strawson’s (1950) complaint that Russell failed to distinguish linguistic expressions and particular dated utterances (or uses) of those expressions. See below. 45 We may call a particular dated utterance of an expression a particular dated use of the expression or a particular dated tokening of it without too much harm—which is not to say we might not want to use ‘use’ and ‘utterance’diVerently elsewhere. See Strawson (1950) and n. 43. For convenience I will stick to talk of particular dated utterances. The word ‘utterance’ (also ‘assertion’, ‘statement’, ‘remark’, ‘question’, ‘inscription’, ‘expression’, ‘interjection’, and others) has a convenient act-object (or process-product) ambiguity that can be nicely exploited as long as great care is taken (e.g. as it is by Sellars (1954) and also by Grice (1989)). It would be quite an undertaking to sort out the logical grammar of such devices, and I have no intention of getting bogged down in it here (although I am strongly inclined to say that someone who utters ‘Tom’s utterance was loud but true’ or ‘his question was well-timed and very relevant’ or ‘his outburst was justiWed and broke a window’ is making a category mistake (perhaps in order to make a joke)). When ‘utterance’ is used for the object/product in, say, ‘my utterance of X ’, at least some of the time it is replaceable by ‘what I said by uttering X ’ or by ‘what I said by my utterance of X ’ in which ‘utterance’ is used in the sense of act/process. And it is for this reason, I think, that many of us are drawn to talk of true or false utterances. When I talk of utterances being true or false this is what I shall have in mind in what follows. There are many other adjectives that we put alongside ‘utterance’: e.g. ‘reassuring’, ‘frightening’, ‘intelligent’, ‘hurtful’, ‘courageous’, ‘bold’, ‘rude’, ‘sentimental’, ‘loud’, ‘ironic’, ‘metaphorical’. We must exercise caution here: in many cases we are predicating something of what the speaker said, of the fact that something was said, of the speaker’s act of saying what he said, or of the particular words used in the act. (If Bill, who has been knocked down and seems close to death, manages to utter the words ‘I am dying’, his utterance may be reassuring in one sense but not in another.) All sorts of dangerous ambiguities lurk here because of issues about the applicability of adjectives such as those listed above to the sorts of things that ‘statement’, ‘utterance’, ‘inscription’, ‘remark’, ‘question’, and so on are applied to. There is clearly an enormous amount of work to be done here, but I think I can get across the main points of the present chapter without getting bogged down in it. This use of ‘elliptical’ in connection with incomplete descriptions has been borrowed by many people. See below. 46 I do not recall whether I intended a nod in the direction of Sperber and Wilson’s (1986) talk of ‘explicit content’ in my choice of ‘explicit’ over ‘overt’, as Zso´Wa Zvolensky has suggested. It is quite possible.

Neale, This, That, and the Other

101

earlier discussion.) No thesis about syntax is implied by Quine and Sellars, by my giving a label to the general response their remarks typify, or by anything else in my discussion. It would be a mistake to think that Quine and Sellars are making a point they take to be speciWc to utterances of descriptions (or even utterances of quantiWcational expressions generally); and it would be a mistake to see them as proposing a theory in any interesting sense or positing some sort of psychological process that will form part of a theory. They are just making an elementary observation about a widespread phenomenon—context-sensitivity—in the use of language, and pointing out that the phenomenon itself constitutes no refutation of Russell’s Theory of Descriptions (which is not to deny that it may be possible to Wnd speciWc cases of incompleteness that are more problematic). As Bach (1981) observes in a discussion of elliptical uses of descriptions, if I say ‘I drink only Scotch’, I would be stating not that I drink nothing but Scotch but merely that the only liquor I drink is Scotch . . . The phenomenon of elliptical speech is commonplace, indeed, it often seems stilted not to suppress words that can easily be inferred . . . Using incomplete descriptions elliptically. . . is just another case of this familiar phenomenon. (1981: 238)

Like Quine and Sellars, I was not rash enough to oVer a theory about how we manage to pull oV the interpretation of incomplete utterances; I posited no psychological process to form part of a theory, and I held out no promissory note. I was defending a particular semantic proposal, a proposal about the linguistic meaning of the word ‘the’ as it occurs in expressions of the form ‘the f’, a proposal that seems to be threatened by incomplete utterances. It was no part of my task to provide a theory that explains how we manage to identify what a speaker is saying and implying on the basis of the meagre evidence we get from knowledge of syntax and word meaning (a theory of interpretation in the sense described earlier). That is a task for cognitive science. Sperber and Wilson (1986/1995) appear to have made some headway here, and the observation made by Quine and Sellars provides grist for their mill. As I said in Descriptions: As (e.g.) Sellars (1954) Sperber and Wilson (1986) have stressed, in many cases the linguistic meaning of the sentence—or sentence fragment—uttered radically underdetermines the proposition it is used to express on a given occasion. We have already considered the sort of contextual supplementation that is required where an utterance contains overtly indexical or demonstrative components; but context-sensitivity does not end there. (1990: 114 n. 46)

In short, I was quite explicit about the larger problem of the underdetermination of what is said, that Quine, Sellars, and Sperber and Wilson provide examples of. And I took the standard philosophy cop-out: that is, like practically every other philosopher who discusses this matter, I did the philosophical equivalent of taking the Wfth: when we wish to avoid discussing genuine psychological processes (being philosophers not psychologists) but also wish to emphasize that we think pragmatic, inferential

102

I. Incomplete Descriptions

processes must be at work, we gesture in the direction of Grice or at least in the direction of ‘Gricean considerations’. Useless to cognitive science, of course, but our way of acknowledging that we are out of our depth and that we realize there is serious work to be done by cognitive psychology. Our interpretive abilities are so good that we can reasonably expect our audiences to identify the thoughts we seek to express even when we use expressions whose linguistic meanings fall short of serving up the precise concepts involved in the thought. The point is easily seen with predication (indeed, it might be argued that the point is about predication). Often a speaker will use a simple predicate, even if a richer or more complex predicate might be used, one that could, in principle, leave the hearer with less inferential work to do. (To say this is not to say that use of the richer predicate will speed up communication—it could slow it down.) The broadly pragmatist literature brims with examples which might be borrowed to make this point (the parenthetical expressions making more explicit what was left only implicit in a particular conversational setting):47 I haven’t had breakfast (this morning) It’s snowing (in Reykjavik) Maria wants to get married (to Fred) Maria and Fred want to get married (to one another) Maria and Fred pushed the car to the garage (together) I hadn’t noticed (that Mike was limping) Maria wants to leave (this party) Maria is ready (to leave (this party)) I haven’t seen Maria (here tonight) You are not going to die (from that injury). What is true of the predication inherent in using verb phrases (as in the examples above) is true of the predication inherent in using noun phrases: everyone (at Ragga’s party last night) had a great time no one (from the US embassy) has arrived yet the car (Tom bought this morning) broke down on the way (to Tom’s) home the Russian (judge) voted for the Russian (skater) the (former) hostages were greeted at the White House (the man on) table six wants to change his order the (man who ordered a) ham sandwich on (table) twelve wants pickles every farmer (in my village) owns exactly one donkey and feeds the donkey (he owns) at night. Something similar appears to be going on with the following where an overall ‘point of evaluation’ may be suppressed: 47 For dozens more examples, see Bach (2001a), from which a few of those above are also borrowed.

Neale, This, That, and the Other

103

(In the Sherlock Holmes stories) Moriarty was Holmes’s arch-enemy (In Homer) Penelope is the wife of Odysseus (In Greek mythology) Ares is the god of war. There is nothing of great theoretical signiWcance in the groupings, and certainly no syntactic thesis is intended. The Wrst group merely illustrates cases in which a richer sentence of the sort required may be provided by beeWng up the predicate that constitutes the VP; those in the second group merely illustrate cases in which a richer sentence of the sort required may be provided by beeWng up the predicate inside the subject noun-phrase; and the third merely illustrates cases in which a richer sentence of the sort required may be provided by adding some sort of sentential operator that can be used to indicate an overall ‘point of evaluation’. Talking involves a trade oV. More detail can lead to greater precision and reduce misunderstanding, but it can also sound stilted, as Bach said, or bore or confuse an audience. We all know people who use too few or too many words for our own tastes, and we have all been in situations where we have used too few or too many ourselves— sometimes we have suVered accordingly. All of this is pretty obvious really, and it is somewhat surprising that philosophers have tied themselves in knots over examples which are so linguistically trivial. (As far as examples involving incomplete descriptions are concerned, Quine (1940) and Sellars (1954) realized just how trivial half a century ago.) What is not trivial, of course, is the articulation of a general pragmatic theory that answers the following question: how it is that we manage to identify what a speaker is saying—let alone implying—by uttering a sentence X on a given occasion given the precious little information we obtain by virtue of our knowledge of the meanings of the words in X and our knowledge of X ’s syntactic structure? It is an empirical theory of utterance interpretation in this sense that people like Sperber and Wilson (1986) are trying to construct. Obviously Quine and Sellars were not presenting any sort of theory of that sort: they were just making commonsense observations.

(c) Sentence Ellipsis (Elliptical Sentences) Bach (1987) and Ostertag (1998, 1999) explicitly warn against confusing the speech act notion of utterance ellipsis just discussed with a quite diVerent syntactic notion that emerged from talk in generative grammar of deletion transformations. Here is Ostertag: the notion of ellipsis appealed to is not the one familiar from syntactic theory. . . whatever the relationship is that holds between [a sentence containing an incomplete description] and [a sentence containing a completed one] it is not syntactic. (1998: 20) There seems no reason to suppose that the relation between, say, [‘the table is covered with books’] and its completion is anything like that between: ‘John left and Bill did too’ and ‘John left and Bill left’. (1999: 144 n. 3)

104

I. Incomplete Descriptions

In a recent encyclopaedia article on ellipsis, Robert May (2002) provides a succinct description of the notion of ellipsis familiar from syntactic theory: ‘A linguistic ellipsis, most generally expressed, is a truncated or partial linguistic form. This partiality is measured relative to a complete sentence; an elliptical sentence is one in which some of the constituent parts of a ‘‘full’’ sentence are missing’ (2002: 1094). The following forms of words are elliptical in this sense: (3) (4)

Peter went to Paris, Bob to Brussels Peter has learned to drive, but Bob hasn’t.

Do the expressions ‘Bob to Brussels’ and ‘Bob hasn’t’ constitute ‘whole sentences’? Ultimately we will want to defer to empirical linguistics here, and given the current state of play this will probably mean positing (a) various types of aphonic expressions, and (b) syntactic structures that resist the sort of description that can be produced by imposing constituent structure (using pairs of labelled brackets, for example) on the word sequences of our usual, non-technical orthography. For concreteness, then, let us follow May (and common practice in linguistics) in saying that we are dealing with sentences: the linguistic forms are ‘partial’ or ‘incomplete’, relative to complete sentences, because they are ‘linguistic form[s] in which constituents normally occurring in a sentence are superWcially absent, licensed by structurally prior antecedents’ (2002: 1094). Certainly I speak elliptically, in the sense of the second grade of ellipsis, when I utter (3) or (4): that is, my utterance is elliptical. But surely there is something more going on here, something that even traditional grammars recognize, let alone generative linguistics. The fact that my utterance is elliptical can be explained in linguistic terms. As we might put it, the elliptical nature of my utterance is attributable to the fact that the sentence I utter is elliptical for some longer sentence that can be straightforwardly ‘recovered’: (30 ) (40 )

Peter went to Paris, Bob went to Brussels Peter has learned to drive, but Bob hasn’t learned to drive.

The modal dimension in sentence ellipsis is clear: there is some longer surface form the speaker could have uttered for which the surface form actually uttered is elliptical, a sentence that can be restored from the syntactic context. Putting things rather simplistically for a moment, we might say that, as a matter of grammar, (i) we can here recover the ellipsed elements by copying them over from elsewhere (subject to certain speciWable constraints), and (ii) we can create an elliptical sentence by deleting elements that are duplicated elsewhere (again subject to certain speciWable constraints). This is very rough, but it’s good enough to set the scene.48 In this respect, sentences (3) and (4) are unlike sentences (1) and (2), and this is surely because (1) and (2) are not elliptical sentences.49 48 Issues to do with (e.g.) binding make a Wnal statement more complex. See below. 49 Of course there may be cases in which it is easy enough to create a longer sentence by drawing explicitly upon linguistic material elsewhere (as in ‘the yellow house in the third block of Lee Street, Tulsa is on Wre. The house was built in 1865’). But that is not the point.

Neale, This, That, and the Other

105

So back to our question: are (3) and (4) whole sentences? Let generative linguistics decide; we’ll call them ‘elliptical sentences’ and if generative grammar says they are sentences, no harm is done; equally, if generative grammar says they are not sentences then ‘elliptical’ in this context is rather like ‘plastic’ in ‘plastic Xowers’ or ‘counterfeit’ in ‘counterfeit banknote’, and still no harm is done. So, given, the current state of play, as presented by May (2002), let’s call them elliptical sentences.

5. The Incompleteness Question The label ‘incomplete description’ is misleading. But we need to begin somewhere, so let us have some preliminary deWnitions. Let us say for the moment that a description is proper if, and only if, its nominal—or its superWcial matrix in some standard system of representation—is true of exactly one thing, and improper otherwise. And let us say that an improper description is empty if it is true of nothing, and incomplete if it is true of more than one thing. (As the need arises, we can tolerate loose but intelligible talk of the matrix of an English description being incomplete, and this will allow us to move back and forth between talk of the matrix table x, for example, and (loose) talk of the matrix ‘table’. No confusion should arise.) Incomplete descriptions are supposed to be interesting because of a question they force the Russellian to answer, one simple variant of which might be put thus: how are we to explain the incontrovertible fact that A can use a description ‘the f’ in an utterance of the simple form ‘the f is c’ and thereby perform a perfectly felicitous speech act, indeed say something true, even though A and B (the hearer) both know that f(x) is true of more than one thing? The question generalizes: how are we to explain the fact that (roughly) for a range of determiners, D, A can use ‘D f’ in an utterance of the simple form ‘D f is c’ and thereby perform a perfectly felicitous speech act, indeed say something true, even though A and B both know that f(x) is true of some things that are not relevant to the truth or falsity of what A said?50 Many philosophers appear to think the answer to the question the Russellian must answer is obvious. ‘There’s always an implicit background restriction on the domain over which a quantiWer expression ranges’, is one old reply. Another is, ‘An utterance of ‘‘the f is c’’ is sometimes elliptical for an utterance of ‘‘the f that z is c’’, where z is something the speaker could have made explicit but didn’t.’ Call these the implicit reply and explicit replies, respectively, based only on the appearance of the words ‘implicit’ and ‘explicit’ in the quoted remarks. Quine (1940), Sellars (1954), Sainsbury (1979), Davies (1981), Evans (1982), and many others have replied in one or both of these ways, and when I was writing Descriptions I thought their remarks were sensible enough, the sort of general remarks that set a target a pragmatic theory should reach, 50 The words ‘roughly’, ‘range’, and ‘simple’ appear here so that I can say something straightforward without bringing up issues to do with negation, monotonicity, persistence, and binding.

106

I. Incomplete Descriptions

explaining as it should how we integrate information linguistic and non-linguistic information to work out what the speaker is saying. I spent very little time comparing or even discussing the implicit and explicit responses because (for better or worse) I saw incomplete descriptions as posing a pretty spurious threat to the account of deWnite descriptions I was defending. I argued that methodological considerations of the sort advanced by Grice and Kripke strongly favoured a unitary Russellian analysis of descriptions, and that the usual arguments for semantically distinct referential readings—including the Argument from Incompleteness—failed to demonstrate the desired conclusion, indications to the contrary being largely false impressions engendered by inattention to the distinction between what a speaker says and what he means. The discussion in Descriptions was, I think, rather lazy, and I now believe only half of it. To be precise, I still think the Russellian analysis is basically correct for both attributive and referential uses of descriptions; but I no longer think the diVerence between saying and meaning lies at the heart of a characterization of referential usage, and I want to provide a more satisfying (less lazy) account of the distinction. What prompted me to seek something more satisfying? Interesting problems for the unitary Russellian position I favoured in 1990 posed by Bezuidenhout (1997), Carston (2002), Devitt (1997a, b), Devitt and Sterelny (1999), Larson and Segal (1995), Recanati (1993, 1996), Ramchandran (1993, 1995), Reimer (1992, 1998a), Rouchota (1992), SchiVer (1995), G. Wilson (1991), and Zvolenszky (2000), problems that either take the form of counter-examples exploiting apparent weaknesses in one or other of the explicit and implicit approaches, or else build on foundational or methodological worries revolving around the fact that referential uses of descriptions are common, standard, systematic, and cross-linguistic.51 The works just mentioned have advanced my understanding of the issues considerably. Not only were they instrumental in leading me to change my mind on one or two of them, more importantly they led me to see certain things in new ways and to believe the central debate in this area is the product of a powerful illusion. Explaining that illusion is a central task in what follows.

51 Bezuidenhout, Devitt, Ramchandran, Recanati, are primarily interested in motivating a supplementary, semantically distinct, referential interpretation. Reimer is more concerned with explaining why the explicit/ ellipsis approach is inferior to the implict/domain restriction approach, a novel variation of which she motivates and seeks to distance from anything that could be construed as a notational variant of the explicit approach. Reimer’s discussion displays a keen appreciation of the virtues of the implicit approach, but ultimately I think the attempt to separate it from the explicit approach fails (see below). Wilson and Ramchandran aim to undermine the unitary Russellian account by producing examples apparently requiring a semantically distinct referential interpretation to supplement the Russellian interpretation. In this way, Wilson attempts to motivate his own ‘pronominal’ theory of referentially used descriptions. SchiVer’s examples are presented in a broader context, as part of an elaborate argument designed to produce a dilemma for the Russellian who also holds a direct reference theory of indexicals, itself part of a larger argument designed to produce a dilemma for the hidden-indexical theory of attitude reports. Larson and Segal are rightly concerned with producing a theory that is satisfactory from a syntactic as well as a semantic perspective, and see the way in which syntax and semantics come together as strongly suggesting the need for a supplementary referential interpretation for certain descriptions used referentially.

Neale, This, That, and the Other

107

Before getting to that illusion, however, certain others need to be dispelled. Our question, recall, is, ‘How are we to explain the incontrovertible fact that a speaker can use a description ‘‘the f’’ in an utterance of the simple form ‘‘the f is c’’ and thereby perform a perfectly felicitous speech act, indeed say something true, even though he and his hearer both know that f(x) is true of more than one thing?’ By deigning to provide even a vague answer to this particular question—rather than rejecting it or answering a question with which it might be confused—one has, in eVect, already accepted a central tenet of linguistic pragmatism, the underdetermination thesis, found in embryonic form in Quine (1940) and Sellars (1954) in their discussions of incomplete descriptions, lurking in the work of Austin (1962), and articulated clearly by Sperber and Wilson (1986). It is a point familiar from discussions of context-sensitive or indexical expressions such as ‘that’, ‘he’, ‘I’, and ‘you’ that knowing what sentence has been uttered on a given occasion and knowing its linguistic meaning may be insuYcient for identifying what the speaker said. The interpretation of utterances containing incomplete descriptions extends the point, for even when the references of any overt indexicals and other referring expressions have been identiWed, there may be further context-sensitivity to be resolved. This was something I tried to make clear in Descriptions, where I explicitly accepted the underdetermination thesis as already noted. Underdetermination that goes beyond that induced by ‘overtly indexical or demonstrative’ or other referring expressions in a sentence X may be of two types, distinguishable in respect of their syntactic commitments. (i) It may be due to Sellarsian utterance ellipsis, a notion that has no syntactic dimension. Or (ii) it may be due to the existence of covert, i.e. aphonic indexical or demonstrative elements in X’s syntax, elements just as much in need of interpretation as occurrences of the phonic elements ‘he’ or ‘that’. Elementary reXections on utterance interpretation and empirical work on syntactic structure suggest very strongly that we should accept the existence of both utterance ellipsis and aphonic items in syntax. Nonetheless, there are those who seek to deny the existence of one or the other, or to reduce all cases of one to cases of the other, motivated it would seem by either a prejudice against pragmatism or a prejudice against Chomskyan aphonics.52 On the assumption that we are involved in an empirical exercise that looks as though many of the tenets of pragmatism and Chomskyan syntax will loom large, I see no good reason to think either form of reduction will be successful, so I will assume the existence of both utterance ellipsis and aphonic elements in syntax until a good argument against one or the other comes along.

52 The former prejudice appears in the work of Stanley and Szabo´ (2000a, b), Stanley (2000, 2002a, b, c), the latter in the work of Barwise and Perry (1983).

108

I. Incomplete Descriptions

The postulation of a crucial aphonic in every sentence containing a description does not really involve rejecting the original question the Russellian has to face: for aphonics are just as much in need of interpretation as phonics. Rather positing an aphonic constitutes a tottering Wrst step towards providing an answer to the question, and does not change the basic point: knowing the blueprint for ‘the table is brown’ (by virtue of knowing its syntax and the linguistic meanings of the words ‘the’, ‘table’, ‘is’, and ‘brown’) does not suYce for grasping what A has said by uttering it on a given occasion.53 And the mere act of positing a context-sensitive aphonic lurking in the sentence’s LF that gets interpreted in the right way, thereby making everything work out just right, does not itself constitute an explanation of how it gets interpreted in the right way (if only producing a theory of interpretation were so easy!). One still needs to explain the semantics of this new element, explain the sorts of values the speaker can intend it to have on diVerent occasions, and then explain how hearers go about identifying these values. The last of these is something a pragmatic theory must explain, and the explanation is going to be very similar to the one given for words like ‘he’, ‘she’, ‘it’, ‘this’, ‘that’, and ‘here’. A semantic theory explains the sorts of values the speaker can intend, say, ‘he’ to have on diVerent occasions, and a pragmatic theory explains how hearers go about identifying these values on particular occasions. Only a full-Xedged pragmatic theory can explain how the speaker is able to decide how thin a description he can get away with, being reasonably certain the hearer will identify what he intends to be saying. And only a full-Xedged pragmatic theory can explain how hearers do in fact identify what speakers are saying when they are using incomplete descriptions. A cognitive theory of the sort being constructed by Sperber and Wilson is what is needed here—not the sort of Mickey Mouse Gricean theory I 53 To the best of my knowledge the only sustained attempt to deny this has come from Bach (1981, 1987, 1994, 2000), who argues that, if there is more than one f, the speaker actually said something false by uttering ‘the f is c’, but nonetheless meant something true. In Descriptions I was pretty short with this idea on the grounds that it is inconsistent with a basic tenet of linguistic pragmatism: it clashes with our intuitive judgments of truth and falsity by virtue of clashing with our intuitive judgments about what the speaker is saying, which I take to be the principal data a theory of interpretation is meant to explain. (Soames (1986), Reimer (1998a), and Stanley and Szabo´ (2000a, b) also reject this type of approach.) Following L. J. Cohen (1971), the standard way of making trouble for theories that propose a thin notion of what is said in a particular type of example is to bury it inside a conditional. Thus Cohen objected to Grice’s thin (truth-functional) account of ‘and’ by contrasting ‘if f and c then z’ and ‘if c and f then z’. (For detailed discussion of this tactic, see Carston (2002).) So one is naturally drawn to examine examples such as the following in connection with Bach’s claim about descriptions: (i) If the table is dirty, then the waitress will be dismissed. I take it the existence of a dirty table at Famous Ray’s Pizza could not be cited by a waitress at Tom’s Diner in her defence when faced with dismissal (from Tom’s Diner). In order to have a framework within which to discuss the relevant issues, I am assuming, as I did in Descriptions, that we can get by with just the two most intuitive components of what a speaker, A, meant, at least for the purposes of discussing deWnite descriptions, viz. what A said (the proposition(s) A expressed) and what A implied (the propositions A implied). Within that framework, giving a direct answer to our original question means accepting that incomplete descriptions introduce a measure of insuYciency, and rejecting that question means denying they do. This is not to say, of course, that one cannot construct a more sophisticated framework within which one distinguishes what A says and A states, for example, as Bach and Harnish (1979) and Bach (1987, 1994) do; or distinguish what A says and what A’s utterance says. It may well be the case that the use of incomplete descriptions is precisely the sort of thing that motivates making such distinctions, and perhaps I am remiss in not making all the distinctions I mentioned at the beginning of ch. 2 of Descriptions.

Neale, This, That, and the Other

109

was using in Descriptions to illustrate the intuitive and, one would imagine, theoretically signiWcant, distinction between what a speaker says and what he means but does not say. I make no excuse for not having addressed the big cognitive questions. What I could have said more about, however, were the labels ‘incomplete’, ‘implicit’, and ‘explicit’. The Wrst had been used for many years in connection with descriptions. For better or worse, the second and third were picked up from Descriptions by philosophers and linguists attempting to set out the virtues and vices of one or both of the general strategies. Labels are only tags, of course, but often we select this or that one because of what it connotes; and this can, on occasion, increase rather than decrease the risk of misunderstanding. The connotations I had in mind for ‘explicit’ and ‘implicit’—I contemplated ‘overt’ and ‘covert’—were these: the possibility of producing a windier utterance that explicitly (overtly) speciWes what an incomplete quantiWer, in context, is taken to express; the existence of an implicit (covert) limitation on the domain of quantiWcation. I shall say more about ‘explicit’ and ‘implicit’ in the next section. First I need to say something about ‘incomplete’. The label ‘incomplete’ is misleading in its application to a type of linguistic expression such as a deWnite description (similarly, ‘empty’ and ‘improper’).54 The label— which appears to have beaten out such rivals as ‘imperfect’, ‘indeWnite’, and ‘elliptical’—seems to have emerged from Sellars’s discussion, and it may seem strained at Wrst because of an unfortunate connotation. The word ‘incomplete’ carries a suggestion of something missing and of something falling short, something that needs completing. And so care is required, for it is easy to slip back and forth between diVerent objects that seem to be in need of completion. Of course no one seriously involved in the debate over incomplete descriptions over the past sixty years or so construes it as a debate about points of grammar. That is, no one takes it to be a debate about grammatically incomplete or grammatically deWcient expressions in need of additional linguistic material in order to be turned into grammatically well-formed descriptions; no one actually thinks that ‘the table’, ‘the house’, and so on fall short of being complete English noun-phrases, or that ‘the table is large’ or ‘the house is on Wre’ fall short of being complete English sentences; and no one thinks that the stock explicit and implicit responses to the original question involve methods for turning grammatically incomplete or defective expressions into ones that are not grammatically incomplete or defective. Grammar is simply not the issue. So what sorts of things have we really been attributing incompleteness to for the past sixty years? The remarks by Quine and Sellars quoted above suggest we have been talking all along about incomplete uses or utterances of descriptions. Recall that they brought the suggestive word ‘elliptical’ into the debate in the course of sketching their own answers to the question the Russellian must answer. They talk of elliptical ‘uses’ (Quine) or elliptical ‘utterances’ (Sellars) of descriptions, and not of descriptions per se 54 By talking of the deWnite description as a type of expression, I do not mean to be suggesting that the concept of a deWnite description is basic enough to appear in any Wnal syntactic or semantic theory, unlike say the concept of a determiner.

110

I. Incomplete Descriptions

being elliptical. According to Sellars, an utterance of ‘the table’ will typically be elliptical for an utterance the speaker could have made of a richer description such as ‘the table over here’ or ‘the table beside me’ (1954: 200). The connection between ellipsis and incompleteness in Sellars’s thinking manifests itself when he says (i) that ‘in ellipsis the context completes the utterance and enables it to say something which it otherwise would not, diVerent contexts enabling it to say diVerent things’, (ii) that some ‘utterances . . . are not complete and are only made complete by the context in which they are uttered’, and (iii) that ‘statements which are non-elliptical . . . do not depend on their contexts for their completion’ (1954: 200). Drawing upon these early discussions, we might talk of incomplete ‘utterances’ of descriptions. On one occasion, the speaker and hearer may both know there is exactly one f, in which case an utterance of ‘the f’ will not be incomplete. On a later occasion they may both know there is more than one f, in which case an utterance of ‘the f’ will be incomplete. Of course a time the speaker refers to (or describes) in the utterance (rather than the time of the utterance) might be the relevant one. If speaker and hearer know that in 1950 there was exactly one A and that in 1999 there was more than one w, an utterance of ‘the f’ occurring as part of an utterance of ‘In 1950 the w is c’ (in which ‘in 1950’ has larger scope) will not be incomplete; but an utterance of ‘the f’ occurring as part of a simultaneous utterance of ‘In 1999 the f is c’ (in which ‘in 1999’ has larger scope) will be incomplete. (Further issues are raised, of course, because of sequences of tense and mood operators used to signal connections with one another or with a time frame already being assumed in the discourse.) So when we use the expression ‘incomplete description’, we should probably construe this as shorthand for the rather ugly expression ‘incomplete utterance-occurrence of a description’ or as shorthand for the less ugly but very long-winded ‘description that is incomplete relative to a particular utterance-occurrence’.55 No wonder we use ‘incomplete description’! (Mutatis mutandis, for talk of an ‘empty description’ and an ‘improper description’.) Another, perhaps preferable, way of legitimizing ‘incomplete’ as it applies to descriptions emerges if we set the discussion against the sort of background against which it should be set: a theory of utterance interpretation in the sense discussed earlier. The leading idea is that a complete interpretation of an utterance of a sentence X(d) containing a description d cannot always be extracted from the linguistic form of X(d) alone; in particular, a complete interpretation of the sub-utterance of d cannot always be extracted from the linguistic form of d alone: contextual considerations have to be exploited by the hearer in order to identify the speaker’s intentions.56 55 I add ‘occurrence’ only to highlight the fact that a description d should not be regarded as (in)complete with respect to the whole utterance in which it occurs but with respect to a particular occurrence within the sentence uttered. In principle, one utterance-occurrence of d may be complete and another occurrence incomplete in the very same utterance. Convincing cases are, perhaps, not that easy to produce, but it would be rash not to guard against the possibility of such cases. 56 As I said in ch. 5, when summarizing part of ch. 3, ‘even the descriptive content of an overt description is not always Wxed by purely linguistic factors’ (p. 201). The word ‘even’ appears here and the word ‘overt’ was

Neale, This, That, and the Other

111

This way of stating things turns out to be rather illuminating and avoids prejudicing the issue in favour of either the Unitarian or the Ambiguity theorist. The central disagreement between these theorists will be about the form of complete interpretations in cases of referential usage, the Unitarian arguing that they are to be expressed in terms of general propositions, the Ambiguity theorist arguing that they are to be expressed in terms of singular propositions. Attacks on either position will take the form of attacks either on the truth conditions of the favoured proposition or on the favoured method of completion. Whether or not they are correct, the truth conditions supplied by Russell’s theory for utterances of sentences containing (sub-utterances of ) improper descriptions are at least clear, and they Xow directly from the theory without further ado (in this respect the theory diVers markedly from many other theories). Nonetheless, incomplete descriptions appear to present interesting challenges. The (optimistic) line I took in Descriptions was basically that incomplete descriptions were an argumentative deadend: no matter how clever or spare the matrix, how Wendish the context, it was always possible, I suggested, to generate essentially the same incompleteness problems using occurrences of descriptions that were uncontroversially non-referential, or occurrences of other noun-phrases that were uncontroversially assigned quantiWcational interpretations.57 Rather than committing resources to a futile ideological debate—for that is what I thought it had become—I thought we should try to understand the general problem of what I called quantiWer incompleteness: how is it that we can legitimately use, say, ‘every horse’, ‘no horse’, ‘the horses’, or ‘the horse’ without being understood as making claims about every horse in existence, or use ‘the horse’ without being understood as claiming (in part) that there exists only one horse? On the assumption that (much of ) the semantic power of (many) noun-phrases in natural language can be understood in terms of restricted quantiWers of the form [Dx: f(x)], where D is a quantiWcational determiner, the problem of matrix interpretation is to explain how it is that f(x) is often understood, in context, as true of fewer objects than it is true of when taken at face value.58 Let us now examine this matter in more detail. italicized because I was, in that passage, commenting Wrst on the fact that the descriptive content of descriptive (i.e. D-type) pronouns is not always determined by purely linguistic factors. 57 That incomplete descriptions may be used attributively was noted by Donnellan (1968). That any adequate account of incomplete descriptions must handle such uses as well as referential uses is stressed by Peacocke (1975), Evans (1982), Soames (1986), Bach (1987), and by me in Descriptions. 58 I said that quantiWcational noun-phrases can be understood in terms of restricted quantiWers of the form [Dx : f(x)], not that the noun-phrases are, or are represented as, such quantiWers. Strictly speaking, the English _ quantiWcational noun-phrases ‘every man’ and ‘the man’ are not even of the same syntactic category as the restricted quantiWers [every x: man x] and [the x: man x]: the former combine with a verb phrase to form a sentence, the latter combine with a formula c to form a formula :z, and c: must contain an overt variable x if z is to count as an adequate representation of a sentence of English. To avoid wordiness, I shall henceforth pass over this detail and allow a certain amount of intelligible and readily corrigible loose talk. It will be useful on many occasions to move back and forth between English descriptions such ‘the man’ and RQ ‘descriptions’ such as [the x: man x], and on many such occasions I will use ‘quantiWcational noun-phrase’ and ‘restricted quantiWer’ interchangeably, despite the strict inaccuracy of such talk. I shall also move back and forth between sentences of English and formulae of the formal language. No confusion should arise.

112

I. Incomplete Descriptions

6. LF and Aphonicity Following Chomsky, it is common in much work in syntax to assume some sort of distinction between a sentence’s superWcial and underlying form. The details may have changed over the decades, labels may have come and gone, and various theory-internal commitments may have changed or evolved, but the idea of some sort of distinction between superWcial and underlying form is still with us, except that today it is, in many ways, more natural than ever in so far as it is connected more transparently to sound and meaning. With Chomsky (1995), let us identify a sentence with a pair hp, l:i: where p is a PF (‘Phonetic Form’) to be read by the sound system, and l an LF (‘Logical Form’) to be read by the intentional system. As Chomsky has stressed for a quarter of a century, LFs are not full-Xedged intentional representations, as rich in content as the those involved in beliefs, intentions, or expectations, objects with truth conditions (or the analogues thereof ), inferential roles, and so on. LFs are simply the grammar’s contribution to the generation of such representations by the intentional system—which receives inputs from various cognitive faculties. The basic idea behind the concept of LF representations has remained robust since its inception: An LF incorporates, ‘whatever features of sentence structure (1) enter into the semantic interpretation of sentences and (2) are strictly determined by properties of sentence grammar’ (Chomsky, 1976: 305). The only diVerence today is what is meant by ‘strictly determined by properties of sentence grammar’. With the emergence of the minimalist outlook, this phrase may be usefully understood as ‘strictly determined by the exigencies of connecting sound and meaning’. Running through Chomsky’s (1995, 2000) recent work on syntax is an argument from ‘virtual conceptual necessity’: complexity and stipulation are to be avoided as, all else being equal, language will employ only those devices needed to link sound and meaning. On the assumption that there is a component of the mind/brain dedicated to language, the human language faculty, one consequence of Chomsky’s ‘minimalist’ outlook is that all properties of sentences relevant to sound and meaning should be derivable from quite general considerations about the way the language faculty must engage with two other cognitive systems, one dealing with the articulation of sounds and their perception (henceforth the sound system), the other trading in intentional/conceptual representations (henceforth the intentional system). A particular language can be seen as an instantiation of the language faculty (with certain options speciWed), something that can provide ‘instructions’ to be interpreted by the sound system, on the one hand, and the intentional system, on the other. More speciWcally, a language is a computational system that generates pairs hp, li of representations. If LFs are not full-Xedged intentional representations, what are they? They are, I said, the grammar’s contribution to the generation of such representations by the

Neale, This, That, and the Other

113

intentional system. But what does this amount to? To get the Xavour it is helpful to think of the intentional system as receiving inputs from various cognitive faculties and trading in representations something like the sentences of the ‘language of thought’ in Fodor’s (1975, 1983) sense, a modality-neutral symbolic system of representation in which thought takes place and into whose sentences utterances of natural language sentences must be mapped if understanding is to take place. LFs exhaust the grammar’s contributions to this system; an LF is not an interpreted object with an intentional content: it is simply a syntactic representation, determined by sentence grammar, of those features of grammatical structure that enter into the interpretation of utterances of that sentence (for example, relations of scope and binding). Not only do LFs fail, for example, to specify references for referentially independent occurrences of pronouns, they fall short of being full-blown intentional representations in all sorts of other ways.59 Reference to the theory of phrase structure is virtually eliminated in the minimalist framework, there being no phrase structure rules in the traditional sense, and strictly speaking there is no constituent structure to a PF representation. So let us borrow the

59 See Chomsky (1976, 1986, 1995, 2000), Sperber and Wilson (1986/1995), and Carston (2002). In a Davidsonian spirit, Higginbotham and May (1981), Higginbotham (1983a, b, 1985), Larson and Segal (1995), Ludlow (1989, 2002), Neale (1994), and others have treated LFs as objects which (relative to assignments of values to referential elements, some of which may be of an indexical nature) have recursively speciWable truth conditions, an idea Chomsky rejects. There is, I think, something of value in the alternative conception of LF that needs to be superimposed upon the oYcial Chomskyan conception to give it bite, but it is diYcult to make it precise. In trying to eVect the superimposition, my own earlier discussions of LF evince deeply worrying ambiguities. In ‘Events and LF’ (1988) and Descriptions (1990) the discussion is very Chomskyan: I treat LFs as no more than syntactic objects encoding those aspects of syntax relevant to interpretation, and I am careful to distinguish LFs themselves from the formulae in a system of restricted quantiWcation I use to represent the truth conditions of utterances of sentences, formulae that could depart signiWcantly in structure from the LFs of the sentences uttered (e.g. in the discussions of perceptual reports, incomplete descriptions, and descriptive pronouns). Various people tried to convince me I needed to embrace, or at least move closer to the truth-evaluable conception of LFs, but the interpretation of utterances containing incomplete descriptions and other underspeciWed DPs, as well as the interpretation of D-type pronouns, held me back as it seemed, and still seems, preposterous to treat LFs as containing all sorts of unrecoverable predicative devices not present in surface syntax. At the same time, it has always seemed to me that if the Chomskyan LF of a sentence X is meant to lay bare the contribution made to the process of interpretation by X ’s syntax, it must provide genuine constraints on what someone uttering X can be saying, something for which the alternative truth-conditional conception is tailormade. In ‘Logical Form and LF’ (1994), I attempted a reconciliation of sorts—originally with Larson—by abstracting as much as possible from pragmatically determined aspects of the truth conditions of utterances. The attempt was ultimately unsuccessful, I think, because I strayed so far from the Chomskyan conception of LF. The truth-evaluable notion of LF appears to have a simpler time with inference. A classic case of ‘logical form’ revealing inferentially vital elements is Davidson’s (1967) account of action sentences: ‘John left quickly’ is meant to entail ‘John left’ because each is an existential quantiWcation over events and Wrst-order logic guarantees the validity of the following: (i) (9x)(leave(John, x) • quick(x)); therefore (ii) (9x)leave(John, x). It is in this spirit that Higginbotham (1983b, 1985) posits variables corresponding to those Davidson sees in the ‘logical forms’ of English sentences in their LFs, again assuming that LFs are ripe with truth conditions (relative to assignments). On the assumption that the LF of ‘Mary saw John leave’ also contains an event variable corresponding to the one in the premise of the inference below, Higginbotham is able to capture further inferences by appealing to LFs: (i) (9x)(leave(John, x) • saw(Mary, x)); therefore (ii) (9x)leave(John, x).

114

I. Incomplete Descriptions

old slash notation from phonology when talking about PFs, but with standard orthography rather than phonological symbols inside the slashes: (1) (2)

/John said he was at the bank/ /every man loves his mother/

(This notational convention allows us to talk of /bank/, /he/, and /his/ without prejudging the issue of ambiguity.) We can still allow ourselves the convenience of using phrase structure trees or labelled bracketing to explicate superWcial structure, so to speak, even if what we write down is strictly an unholy hodgepodge of PF and LF. Thus we might use (3’) to explicate the PF (3): (3) /the king snores/ (30 ) S

DP

D

VP

NP

V

N

the

king

snores

[S [DP [D the] [NP [N king]]] [VP [V snores]]] Following Abney (1987), today it is common to call what used to be called an NP (noun-phrase) a DP (determiner phrase), to respect the idea that its head is the D (determiner) not the N (noun). On this usage, NP is the label for the nominal expression, simple or complex, with which a D merges to form a DP: (4) (5) (6)

[DP [D the][NP [N king]]] [D [D the][NP [A French][N king]]]: S

DP1

D

S

NP

DP

V

N

the

king

VP

x1

snores.

[S [DP [D the] [NP [N king]]]0 [S x1 [VP [V snores]]]]

Neale, This, That, and the Other

115

The LF corresponding to the PF (3) will look something like (6). In (6) the quantiWer expression ‘the king’ has been extracted—indeed forced by general principles of morphosyntax—from its original position (discernible in the hodgepodge (3’)) and has merged with the original S node to form another (it has been ‘Chomsky-adjoined’ to S). From this ‘new’ position it binds the variable x that now occupies its ‘original’ position, that position now being within its scope. ‘The king’ and x are co-indexed: the numerical subscript on x is an index indicating that it is to be read as bound by the quantiWer expression ‘the king’, which bears the same index as superscript.60 This talk of variables and binding amounts to a description of an important part of the interpretive information carried by (6), precisely the sort of thing Chomsky has always ascribed to LFs.61 One question that will have to be addressed is whether quantiWer expressions are the only DPs that are raised at LF or whether the phenomenon is fully general, involving names, pronouns, and possessives for example (DPs whose structures we will look at in a moment):62 (7) [S [DP Anne]1 [S [DP the king]2 [S x 1 [VP respects x2 ]]]] (8) [S [DP she]1 [S [DP the king]2 [S x 1 [VP respects x2 ]]]] (9) [S [DP Marga]1 [S [DP her1 mayor]2 [S x 1 [VP respects x2 ]]]]. My own suspicion is that raising will have to be fully general, and I shall give my reasons later. (To jump ahead: notice ‘her’ in (9) is co-indexed with ‘Marga’, and notice the order of the raised DPs.) Consider (10) and (11): (10) (11)

John promised Ann to sing John asked Ann to sing.

Traditional grammars talk about the ‘understood subject’ of the verb ‘sing’ in such examples: in (10) it is ‘John’, in (11) it is ‘Ann’. In syntactic theory this is captured by a diVerence in the status of an aphonic element occupying the subject position of the embedded inWnitival clause:63 (10’) (11’)

[S John1 promised Ann [S e1 to sing]] [S John asked Ann2 [S e2 to sing]].

60 Using only subscripts to co-index would obscure the fact that binding is an asymmetric relation, unlike, say, co-reference; this will be important later. I am deliberately simplifying here. For example, I have ignored the fact that the superscripted index on the DP ‘the king’ has been projected upwards from the index on the D ‘the’. 61 Variables in syntactic theory are syntactic objects, but as Heim and Kratzer (1998) stress they are not merely syntactic objects. The concept that variable binding attempts to formalize is an interpretive one. Variables are expressions interpreted in a certain way, expressions whose values vary with something or other, as the etymology suggests. For discussion see Neale (forthcoming a). 62 If, as seems plausible, possessives are deWnite descriptions, and deWnite descriptions are quantiWer expressions, then we already have our answer in one case. 63 For simplicity I have not raised the names here.

116

I. Incomplete Descriptions

In (10’), the interpretation of e is required (by the syntax and the meaning of the verb ‘promise’) to proceed by way of the interpretation of the subject expression ‘John’ (‘promise’ is a subject-control verb). In (11’), by contrast, the interpretation of e is required (by the syntax and semantics of the verb ‘ask’) to proceed via the interpretation of the object expression ‘Ann’ (‘ask’ is an object-control verb (in this construction)). It is worth pointing out a syntactic diVerence between the aphonic DPs in (10) and (11), on the one hand, and those in (6)–(9), on the other: in the terminology of older syntactic theory, those in (6)–(9) are movement-generated (assuming a ‘raising’ operation), whereas those in (10) and (11) are base-generated (given the lexical entries for ‘promise’ and ‘ask’). For present purposes the important syntactic diVerences between types of aphonics can be put aside. A word of caution. As in Descriptions, (i) my use in the present chapter of formulae of a semi-formal system of restricted quantiWcation to capture the truth conditions of what a speaker says by uttering a sentence X in a way that respects X ’s semantically relevant structure, and (ii) my appeal to a Chomskyan picture of syntax according to which X may be factored into a surface form (its PF in today’s lingo) and its LF, where the latter reveals X’s semantically relevant structure and may contain elements no counterparts of which appear in X ’s surface form, does not mean the formulae of the semi-formal language are LFs. They most certainly are not LFs, containing as they do all sorts of constituents no counterparts of which appear in LFs, most noticeably where incomplete descriptions are concerned. Like the notation of structured propositions, they are methodologically useful precisely because they enable us to characterize the failure of isomorphism between form and content and to do so using a notation that nonetheless respects the view that what is said bears some systematic relation to syntax and word meaning.

7. ‘Implicit’ and ‘Explicit’ In Descriptions I alluded to two common responses philosophers and logicians have made over the past Wfty or sixty years to the problem of incomplete descriptions (‘two of the more popular approaches’ as Reimer (1998a: 96) quite rightly puts it).64 And I suggested, perhaps rashly, that attempts to spell them out in satisfactory ways might render them notational variants (when ‘all is said and done’, as I put it). I also said that both responses might be necessary in any Wnal understanding of incompleteness. Assuming a Russellian account of descriptions, the problem of incompleteness is, in its most general terms, the problem of explaining how it is that an occurrence of a matrix f(x) occurring in a quantiWed expression [the x: f(x)] is understood, in context, 64 Reimer’s (1998a) paper is an illuminating discussion of the two approaches.

Neale, This, That, and the Other

117

as true of fewer objects than its superWcial form seems to require. Putting the problem this way allows us to see it, as I claimed we should see it, as an instance of a more general problem that aVects the use of quantiWed noun-phrases. When this problem arises, we are faced with a case in which we seem to have slippage between language and the world. And there are only two things we can do about that: tinker with language, or tinker with the world. When we tinker with language: we do something about the matrix f(x). When we tinker with the world, we do something about the objects that (potentially) satisfy the matrix. If we tinker with f(x) let us say that we are adopting an ‘explicit’ approach to the problem; if we tinker with the objects let us call it an ‘implicit’ approach. (This choice of terminology will become clear.) The distinction between the world-tinkering, implicit approach and the languagetinkering, explicit approach corresponds to a diVerence in focus and in attitude with respect to diVerent parts of a DP (this is perhaps easier to see under the DP hypothesis): (1) [DP [D the][NP [N table]]] (2) [D [D the][NP [A red][N table]]]. A personal pronoun can be viewed as a D that eVectively serves as a full DP by virtue of merging (combining) with an aphonic NP (one that has no phonological matrix and so does not appear, so to speak, at PF)65. For example, ‘he’ would have the following structure, where e is aphonic: (3)

[DP [D he][NP e]].66

We Wnd Ds occurring with aphonic NPs elsewhere:

65 For discussion, see Abney (1987), Cardinaletti (1994), Cardinaletti and Starke (1999), Chomsky (1995), Szabolcsi (1994). The idea that pronouns might be determiners appears to originate with Postal (1969). In the elementary exposition of DPs that follows, I simplify dramatically as on-going debates in linguistics about the details are not crucial to the philosophical issues of concern here. 66 The postulation of expressions that are aphonic despite having syntactic roˆles and semantic properties is surely no more or less problematic than the postulation of expressions that are semantically empty despite having syntactic roˆles and phonological properties (‘it’ in ‘it’s raining’, for example). The idea of an expression that is phonetically and semantically empty is harder to gets one’s mind around, and on the interpretation of Chomsky’s present framework I endorse—syntax is whatever it is that relates PF and LF—the possibility of such an expression is straightforwardly excluded. The discovery or postulation of any expression constitutes a contribution to syntax in the Wrst instance, but its existence is justiWed only if it is doing something at LF or PF. Consequently, the discovery or postulation of an aphonic expression must be justiWed by its roˆle at LF, and to this extent, it will contribute in one way to the project of producing a theory of utterance interpretation. (The general point should not be exaggerated, however. The discovery or postulation of an aphonic, indexical expression, one every bit as Xexible in its interpretation as the overt expressions ‘this’ or ‘that’ or ‘he’ (when used to make independent reference) does not Wx interpretation: the interpretation of an utterance of a sentence containing such an expression is always going to be a full-Xedged pragmatic, i.e. inferential matter, the semantics of the aphonic expression itself placing only non-deterministic constraints on interpretation.) So how does [NP e] contribute to LF? What is its semantic role? The answer I explore in Neale (forthcoming a) is that it is interpreted as a formula x k ¼ x j (k 6¼ j). the idea is that the DP [DP [D he][NP e] ] is interpreted as [he x k : x k ¼ x j ], assuming an axiom for he that is a trivial modiWcation of the Russellian axiom for the, as Postal’s hypothesis would anyway suggest. For a brief sketch, see the end of Section 24 of the present chapter.

118

I. Incomplete Descriptions (4) (5) (6) (7)

[DP [D one][NP e]] is broken [DP [D this][NP e]] is hot [DP [D his][NP e]] is broken67 [DP [D many][NP e]] applied but [DP [D few][NP e]] were chosen.

There are diVerences in (4)–(7) that need not detain us, but in each case the postulated aphonic [NP e] may be replaced by a phonic NP like ‘plate’. (With pronouns this can be done, but it feels epenthetic: ‘I, Stephen’, ‘you, dear reader’, ‘he, Chomsky’. I suppose any decent theory positing (3) as the structure of ‘he’ should explain this.) By contrast with pronouns, proper names can be viewed as Ns eVectively serving as full DPs by virtue of merging (in English) with aphonic Ds:68 (8)

[D [D e][NP John]].

Perhaps we Wnd NPs with aphonic Ds elsewhere: (9)

[DP [D e][NP whales]] are [DP [D e][NP mammals]].

This picture of DPs helps frame the two traditional ways of thinking about incomplete descriptions. If we tinker with language, we end up doing something about the matrix f(x). Let us call this the ‘explicit’ response to the incompleteness problem. If we tinker with the world, we end up doing something about the objects that satisfy the matrix. Let us call this the ‘implicit’ response. Consider (10). The D ‘the’ is the head of the DP ‘the table’. The implicit response purports to explain how we get away with 67 In English the distinction between genitive and possessive determiners is unmarked, at least phonologically. This is not universal. In Icelandic, e.g. third person genitive and possessive pronouns are quite distinct, lexically and also in respect of inXectional morphology. Thus (i) is translated as (ii) or (iii) according as ‘his’ is [bound by] by ‘John’ [‘every man’]: (i) John [every man] loves his wife; (ii) Jo´n [se´rhver maður] elskar konuna sı´na; (iii) Jo´n [se´rhver maður] elskar konuna hans. In (ii), sı´na is a reXexive possessive (or possessive reXexive) that has to be understood as bound by (and hence within the scope of, i.e. c-commanded by) the subject expression. (There are curious exceptions in subjunctive sentences involving logophoricity. For discussion, see Reuland (2001), Neale (forthcoming a) and references therein.) It occurs in the feminine, accusative, and singular to agree with konuna, the noun it qualiWes. In (iii) hans is the simple genitive, which (unlike sı´na) occurs in the masculine and enters into no agreement relations whatsoever with konuna. The Icelandic deWnite article typically takes the form of a suYx added to the noun (very likely this originated in a free-standing deWnite article which is rarely encountered in ordinary talk today). Without the suYx the noun is typically understood as indeWnite. So, e.g. in the nominative (used primarily for the subject of a verb) the feminine noun kona (‘woman’ or ‘wife’) becomes konan (‘the woman’ or ‘the wife’) when deWnite. In the accusative (used primarily for the direct object of a verb but also with some prepositions) konu becomes konuna when made deWnite, as in the examples above. Apart from certain standardized exceptions, the counterparts of English possessive descriptions are formed using the deWnite rather than the indeWnite form of the noun. There are two main types of exception. The Wrst is where the noun to which the possessive is attached expresses a close family relation, with the notable exception of kona (which can translate ‘woman’ as well as ‘wife’). The second is where the genitive is used with proper names to indicate possession. The accusative form of ‘John’s wife’ is konu Jo´ns: so unlike cases involving possessive pronouns (konuna sı´na) and cases involving genitive pronouns (konuna hans), in cases involving genitive names (konu Jo´ns), the suYx for the deWnite article is not added to the noun. 68 In many languages (e.g. ancient and modern Greek), names appear regularly with the deWnite article. In certain contexts names may occur with the deWnite article and other determiners in English: ‘That John Smith is not the John Smith I was talking about.’

Neale, This, That, and the Other

119

using incomplete DPs by focusing on how the head node, D, or its projection, DP, is to be interpreted, both of these nodes corresponding to a quantiWcational expression. On the assumption that the quantiWcational structure of a sentence ‘the table is c’ is represented reasonably using (10) DP

(11)

D

NP

the

table

[the x: table(x)] c(x)

there are two quantiWers to look at, the unrestricted quantiWer the x, corresponding to the D node, and the restricted quantiWer [the x: table x], corresponding to the DP node. The implicit, world-tinkering, approach explains incomplete usage by limiting the number of objects satisfying table(x), and the fact that there are two quantiWers to consider means there are (at least) two ways of eVecting the required delimitation. Let r be some subset of the objects in the domain of quantiWcation (r for ‘restricted’). In our formal language, RQ, we can represent the two ways of delimiting the domain as follows, subscripting r onto the quantiWer whose domain is to be delimited: (12) (13)

[the xr : table x]c(x) [the x: table x]r c(x)

‘the x (in r) such that x is a table’ ‘the x such that x is a table, (in r).’

I prefer a simple r rather that the variable-containing x 2 r to avoid the suggestion that in (12) and (13) the descriptive content itself is modiWed, as it is in, say, (14): (14)

[the x: table(x) • x 2 r]

‘the x such that x is a table and x 2 r.’

For as I said in Descriptions, one of the central tenets of the implicit approach is that it ‘leaves the descriptive content untouched’ (1990: 95). In (12), the unrestricted quantiWer the x ranges over the things in r; in (13) the restricted quantiWer [the x: table(x)] ranges over the things in r. If r is a proper subset of the original domain, then the felicitous use of an incomplete description can, in principle, be explained. If A utters the sentence ‘the table is c’, if r contains only objects within two metres of where A is sitting, and if there is exactly one thing in r that is a table, and exactly one thing that is a table in r, (12) and (13) will each present exactly one thing, and a table at that, to examine as a potential satisWer of c(x). A roaring success, apparently, for both accounts of the domain limitation. It is not immediately obvious that either of (12) or (13) has an advantage over the other; but we cannot rule out the possibility that we might come across phenomena that bring out important diVerences, so let us keep both on the table, at least for now.

120

I. Incomplete Descriptions

No syntactic thesis is implied by the implicit approach. It is compatible with the thesis that the English DP ‘the table’ contains an aphonic expression corresponding roughly to the subscript r in our formal language, but it is also compatible with the syntactic thesis that what you hear is what you get. That is, the postulation of an aphonic in ‘the table’ is no part of the implicit approach itself, it is, rather, a particular syntactic proposal for implementing it, one that might be motivated or rejected on (presumably) syntactic grounds. An interesting version of the implicit response was produced by Barwise and Perry (1983), in connection with incomplete descriptions used non-referentially. Drawing upon Barwise and Cooper’s work on generalized quantiWers, Barwise and Perry introduce the notion of persistence as a property certain statements have. Suppose statements are evaluated not with respect to the whole world, they suggest, but with respect to speciWed parts of the world, ‘situations’ as they call them. To simplify matters in a way that does not bear on present concerns in any threatening way, let us prescind from time: if we consider the world at a time we can think of situations as just spatial parts of the world (the maximal situation). So, for example, Britain is part of the world. Now consider the following sentence: (15)

the richest person lives in London.

The idea Barwise and Perry want us to entertain is that what is expressed by an utterance of (15) can be true with respect to Britain (because the richest person in Britain lives in London), and at the same time false with respect the world at large (because there is at least one person outside Britain who is richer than the richest person inside). Now any particular utterance of (15) is uttered in a situation S and intended to be evaluated at some particular situation S 0 , which need not be identical to S and which may be smaller than the entire world (the maximal situation). And it is this fact, or so it is suggested, that explains how utterances of (15) work (for now, let us put aside so-called referential uses of descriptions). The phenomenon can be inverted by considering utterances of (16): (16)

some princes are ministers.

What is expressed by an utterance of (16) can be false with respect to Britain, but at the same time true with respect to the world at large (because, say, some Belgian princes are ministers). On Barwise and Perry’s account, the idea is not that someone uttering (15) expresses diVerent propositions according as the utterance is to be evaluated with respect to Britain or the world at large—the proposition that the richest man in Britain lives in London and the proposition that the richest man in the world lives in London, for example. Rather, the idea is that the proposition expressed is the same, but is true with respect to Britain and false with respect to the larger world. Mutatis mutandis for utterances of (16). In Barwise and Perry’s terminology, what is expressed by an utterance of (16) is persistent in (roughly) the following sense: if it is true with respect to a

Neale, This, That, and the Other

121

situation S (for example, Belgium), then it is true with respect to every more encompassing situation S 0 (for example, Europe, or the northern hemisphere, or the whole world). By contrast, what is expressed by an utterance of (15) is non-persistent. It can be true at a situation S (for example, Britain) whilst being false at some more encompassing situation S 0 (for example, Europe, or the northern hemisphere, or the whole world). The persistence of (16) and the non-persistence of (15) are traceable to the determiners each contains: ‘some’ is persistent (in a sense to be deWned rigorously) whilst ‘the’ is not. Various problems with this and other versions of the implicit approach— indeed for the general idea that our utterances are evaluated with respect to less than the entire world—were raised almost immediately by Westersta˚hl (1985a) and Soames (1986) (think about ‘the dog bit another dog’). Before looking at them, let us turn to the explicit response. The explicit response leaves the world alone. It purports to explain the felicitous use of ‘the table’ by looking at the other node in (1), the one not occupied by a quantiWer but by the nominal, the NP ‘table’. The basic idea is explicitly modal: the nominal is often shorthand for, elliptical for, an abbreviation of at least one richer nominal the speaker could have used and could produce if asked to be more explicit. (Hence the name.) Consider the following dialogue: A: The table is scratched B: Which table? A: The table I bought this morning. (Or: The one I bought this morning.)

According to the explicit approach, this type of dialogue is suggestive of what is going on when we make felicitous uses of incomplete descriptions. B is intended to interpret A’s utterance of ‘the table’ as if it were an utterance of ‘the table I bought this morning’. There need not be a unique description that A can supply, but there had better be at least one—and one that B could reasonably have been expected to construct at that—if the speech act is to be felicitous.69 This is vague, of course. But that’s Wne: it’s just a general description of an approach to completion that many philosophers think should be explored, together with its target, one that cannot be made more precise without looking at the mechanisms, a general theory of utterance interpretation that explains how hearers integrate linguistic and non-linguistic information to arrive at interpretations, a theory of the pragmatic, inferential processes involved in utterance interpretation. So on the explicit response, someone producing an utterance of an incomplete DP (e.g. ‘every citizen’ or ‘the president’) is understood as expressing what he would have 69 W. Blackburn (1988) stresses an important point about the explicit approach that I did not appreciate until I had virtually Wnished writing Descriptions (it was Blackburn’s article that made me appreciate it). In the context of its utterance, an incomplete description is typically understood with the force of any of a batch of richer descriptions the speaker could have used to make his point, plenty of which the speaker could produce if asked to be ‘more explicit’.

122

I. Incomplete Descriptions

expressed more explicitly had he uttered a richer (‘complete’) DP (e.g. ‘every US citizen’, ‘the US president’), a DP he could have used in place of the incomplete one. The explicit strategy is sometimes called the ellipsis strategy in the literature, presumably in deference to the suggestions made by Quine (1940) and Sellars (1954), who talk, respectively, of elliptical uses and elliptical utterances of descriptions. In consequence, I shall use ‘explicit response’ and ‘ellipsis response’ interchangeably. A word of warning: I shall sometimes talk about ‘completing an incomplete description’, but this is just a bit of shorthand for ‘coming up with a richer description that does a good job in capturing the speaker’s intention’. Just as there is no syntactic thesis implied by the implicit response, so none is implied by the explicit response. There is certainly no implication, for example, that expressions are transformationally deleted between levels of grammatical representation in a Chomskyan grammar—indeed, on standard assumptions there could not be such a syntactic thesis because such deletions would violate the principle of recoverability, which requires deleted elements to be recoverable from linguistic context (see below).70 Like the implicit approach, the explicit approach is meant only to describe how speakers intend their utterances of incomplete descriptions to be interpreted on particular occasions and to describe the interpretations hearers do seem to get. Obviously it involves no cognitive claim about the mechanisms whereby hearers manage to come up with particular interpretations on particular occasions: that is something for a theory of the pragmatic, inferential processes involved in utterance interpretation to explain. Using the notation given above, the basic idea—which can be developed in a number of ways—is that sometimes the matrix f(x) of a quantiWed DP is understood, in context, as if it were a richer matrix f(x, a) containing an additional argument or a conjunction f(x) • z(x) which the speaker could have readily produced. The meanings of some nouns make it clear an additional argument is called for: ‘the murderer (of x)’, ‘the king (of x)’. As pragmatists sometimes put it, interpretation requires ‘saturation’ of an ‘implicit argument’. The meanings of others clearly don’t invite saturation: ‘the table’, ‘the man’. Here pragmatists sometimes talk of ‘enrichment’, which is constrained only by the exigencies of the interpretation process.71 Talk of saturation and enrichment is not itself meant to constitute a theory in any interesting sense: the only theory involved is a general pragmatic theory, a theory of the cognitive processes involved in utterance interpretation, a theory that explains how hearers integrate linguistic and non-linguistic information in interpreting one another, a theory that explains not only how we interpret utterances of incomplete descriptions but also how we assign reference to names and pronouns, establish binding (where syntax falls

70 Oddly, Stanley and Szabo´ (2000a) read just such a syntactic thesis into my summary of the explicit response in Descriptions, and by implication into the explicit responses of Quine (1940), Sellars (1954), Davies (1981), and Evans (1982), amongst others. For detailed discussion, see below. 71 For discussion of the literature on saturation and enrichment see Carston (2002).

Neale, This, That, and the Other

123

short), and resolve potential ambiguities, and how we identify and interpret utterances replete with irony, metaphor, elision, anacoluthon, aposiopesis, and on top of all of this how we identify what a speaker is implying as well as saying. And these cognitive processes must be appealed to by any account of incompleteness, whether it involves specifying some richer description, specifying some suitable background domain restriction, or specifying some suitable value for some hitherto unappreciated, aphonic domain variable cohabiting a syntactic node with a nominal in the manner of Stanley and Szabo´ (2000a) and Stanley (2000, 2002a, b). However you look at it, it’s magic, and it betrays a misunderstanding of the issues to complain that on the explicit approach the hearer performs an act of magic (the recovery of a magical ellipsis) no counterpart of which the hearer performs on an approach that requires the hearer to supply properties or sets or whatever as values for aphonic domain variables.72 Similarly, it will not do to claim, with Stanley (2002a: 158 n. 12) that the explicit approach ‘simply amounts to a re-description of the phenomenon to be explained, rather than an account of it’ if the implication is that this is less of an account than positing a domain variable that takes on whatever value is required to make things work out correctly. Neither the explicit nor the implicit approach, nor Stanley and Szabo´’s contextual variable approach, constitutes a theory in any sense relevant to interpretation. Whichever way we go here, all of the work is done by pragmatic inference. It is diYcult to imagine anyone sympathetic to the explicit response ever viewing it as subject to the following strange and quite ad hoc constraint: any descriptions that occur in a single sentence must be completed in precisely the same way. At the end of boxing match between a Russian and a Swede I might say to you, upon hearing that the panel of eleven international judges has declared the Swede the winner by ten votes to one, ‘I know why it wasn’t unanimous.’ ‘Why?’ you ask? I reply with (17): (17)

the Russian voted for the Russian.73

72 A hearer who had to assign the sorts of values to aphonic domain variables required by the theory Stanley (2000, 2002a, b) and Stanley and Szabo´ (2000a) propose would seem to be in a worse position than one who had to come up with richer descriptions. Linguistic pragmatism is quite happy to posit aphonic variables where syntactic theory requires them but does not see them as a universal panacea to problems of interpretation. A theory that posits the existence of aphonic domain variables in syntax is primarily a syntactic proposal concerning the LF of a sentence that may be uttered on diVerent occasions to say diVerent things, and it should not be confused with a theory that explains how hearers assign values to these variables. Interpretation of any postulated context-sensitive expression on a given occasion of utterance is itself a pragmatic, richly inferential matter, the product of integrating linguistic and non-linguistic information, something that is done by a pragmatic theory. As far as interpretation of incomplete matrices is concerned, the only substantive diVerence between the pragmatist and someone who postulates an aphonic element co-occurring with a nominal is that the latter insists that the search for and integration of contextual information in the interpretation process is triggered syntactically. I know of no good argument that an item in syntax is necessary for such a search and or for such integration to take place—such an argument would have to come from psychology. Merely pointing to the well-known phenomenon of ‘implicit binding’ certainly does not demonstrate the existence of aphonic variables, as has been recognized since at least Evans’s (1977) account of the interpretation of E-type pronouns, where implicit binding does not involve positing aphonic variables in underlying syntax. 73 This form of example is used by Soames (1986) and Westersta˚hl (1985a) in a related context. See below.

124

I. Incomplete Descriptions

Obviously I would be saying that the Russian judge (in this contest) voted for the Russian boxer (in this contest). The point is almost too obvious to be worth stating, but probably I should have mentioned it in Descriptions and then laid out the consequences as I saw them for the implicit approach and its relation to the explicit approach. Why did I suggest, in Descriptions, (i) that attempts to spell out the explicit and implicit responses in satisfactory ways may render them notational variants, and (ii) that a full understanding of incompleteness may require both responses? Isn’t there a tension between these suggestions? The two suggestions are, in fact, intimately connected. Here was my thinking. First, as Soames (1986) showed in his discussion of Barwise and Perry’s version of the implicit approach, examples like the boxing match present an awful problem, for there is no situation that can be carved out of the world (the maximal situation), and no restricted domain of individuals in the world that can serve as a background against which my utterance of ‘the Russian voted for the Russian’ in the envisioned context comes out true. You need a situation or a domain with two distinct Russians in it (the judge and the boxer), but once you have that you are doomed to declaring my utterance false. In short, the implicit theory fails here because no situation or domain contains exactly one Russian and two distinct Russians. Second, at the time I was writing Descriptions I could see only three things the implicit theorist could do at this point. (a) Accept that the implicit response needs help from the explicit response, and explore the idea that there is some sort of principled way to integrate the two responses as well as some sort of principled way to identify where one rather than the other is doing the lion’s share of the work. (A daunting task.) Hence the suggestion that both responses might be needed to fully understand incompleteness. (b) Attempt to replicate the explicit theorist’s success with the boxing match example by bringing in two distinct situations or restricted domains, one containing judges, the other boxers. The problem here, as I saw it, was explaining what it meant to say that my utterance could be true against two background situations or domains, one that came into view only in connection with the Wrst occurrence of ‘the Russian’ in my utterance, the second only in connection with the second occurrence. And it was always possible to bring prime ministers into the picture. ‘The prime ministers of Britain, France, and Russia have been trying to inXuence the judges’, I might say. And I might continue by uttering (18): (18)

the Russian told the Russian to vote for the Russian.

(The explicit approach has no problem here, of course. My utterance of (18) is understood as elliptical for an utterance of ‘the Russian prime minister told the Russian judge to vote for the Russian boxer’.) And then it was always possible to throw in a few nonextensional doodads and bound variables: (19)

the Dane didn’t know the Swede suspected that the Russian had been ordered by his president to vote for the Russian.

Neale, This, That, and the Other

125

I had no idea how to bring together two let alone three situations or restricted domains in the required way. I couldn’t even see what the required way was! (c) Give up one of the two fundamental tenets of the implicit approach. Not domain restrictions themselves, but the idea that domain restrictions do not get into the proposition expressed. Hence the suggestion that when all is said and done the implicit and explicit responses might end up notational variants. Not because the explicit approach needed modifying—it always seemed Wne to me—but because the only way of rescuing the implicit approach from obscurity might involve putting the (originally) background situations or restricted domains into the proposition expressed, as in (20), where p, j, and b represent the restricted domains or situations in question: (20)

[the x: Russian(x) • x 2 p] [they: Russian(y) • y 2 j] [the z: Russian(z) • z 2 b] x told y to vote for z.

Quite what this would amount to when incomplete descriptions occurred within the scopes of various non-extensional operators, I could never get clear about. (The explicit approach is at least clear here.) But as far as I could make out, only by working appropriate restrictions into the proposition expressed could distinct descriptions occurring in the same sentence take advantage of distinct domains of quantiWcation. And of course it involves abandoning one of the two characteristic features of the implicit approach: the idea of an implicit restriction that does not make itself into the proposition expressed. And the resulting proposal has the air of a pointlessly formal restatement of the explicit response. Replacing the idea that an occurrence of ‘the Russian’ is interpreted as if it were an occurrence of ‘the Russian judge’ by the idea that it is interpreted as ‘the Russian in the set of the things that are judges’—or, for that matter, ‘the Russian having the property of being a judge’ does not seem like great progress.

8. Pronouns and Incomplete Descriptions OYcially, I favoured neither the explicit response nor the implicit response over the other in Descriptions, but my preference for the explicit response is evident from its appearance throughout chapters 5 and 6 in the discussion of descriptive pronouns— those most incomplete of descriptions. After uttering (1), (1)

A man walked over to our table

I might continue with any of (2)–(6), perhaps in descending order of likelihood: (2) (3)

He said nothing The man said nothing

126

I. Incomplete Descriptions (4) (5) (6)

The man (in question) said nothing The man who had walked over said nothing The man who had walked over to our table said nothing.74

Sometimes, it is essential to use a description to avoid ambiguity. Compare the following: (7) (8)

Smith bought an apple. It cost 50¢ ? Smith bought an apple and a banana. It cost 50¢.

In (8) the pronoun ‘it’ must be replaced by ‘the apple’ (or ‘the banana’, ‘the former’ or ‘the latter’) to make one’s meaning clear. This suggests that an utterance of a pronoun anaphoric on but outside the scope of (and hence not bound by) a quantiWed expression is understood as if it were an utterance of a description constructible from linguistic and conversational context. Indeed, this is the basic idea behind the postulation of a natural class of pronouns, which might be seen, on such a view, as conventional devices of ellipsis, viz. D-type pronouns.75 The occurrences of ‘he’ in (2), ‘the man’ in (3), ‘the man in question’ in (4), and ‘the man who had walked over’ in (5) might all be seen, in the context in question, as elliptical for ‘the man who had walked over to our table’.76 Unsurprisingly, nothing precludes an utterance of D-type pronoun being understood as if it were an utterance of a description containing a name or a pronoun. Indeed, this fact seems to provide a plausible explanation of what is going on in cases of so-called donkey anaphora. For example, an utterance of (9) might be understood as if it were an utterance of (90 ): (9) (90 )

John bought only one donkey and Paul vaccinated it. John bought only one donkey and Paul vaccinated the donkey John bought.77

74 Similarly where we have plurals: ‘Two men walked over to our table’, ‘They said nothing’, ‘The men said nothing’, etc. The idea that certain ‘minimum’ deWnite descriptions—e.g. ‘the man’, ‘the woman’, and ‘the thing’—might function like the pronouns ‘he’, ‘she’, and ‘it’ is entertained by Quine (1960: 102–3, 112–13). I imagine he was not the Wrst to point this out. 75 See e.g. Cooper (1979), Davies (1981), Evans (1985), Ludlow and Neale (1991), and Neale (1990). 76 The label ‘D-type’ seemed apt to Ludlow and me for those pronouns Evans claimed were E-types. First, Evans’s E-type pronouns were rigid terms whose references were Wxed by description; Ludlow and I denied the existence of E-type pronouns. (By virtue of having their references Wxed rigidly by description, on Evans’s account E-type pronouns are meant to be semantically equivalent to names introduced by description (‘let us call whoever invented the zip fastener ‘‘Julius’’ ’).) Second, Sommers (1982) had already suggested ‘D-type’ for descriptive pronouns. Third, the idea of D-type pronouns (but not under this label) was logically prior to as well as sportier and more manœuvrable than the idea of E-type pronouns. Pragmatic considerations often require D-type pronouns to be understood with large scope; but if we Wnd occurrences that may be understood with small scope, this would seem to provide convincing evidence against the E-type analysis as well as a challenge to other theories of anaphora. In fact, scope ambiguities can be found in attitude, temporal, and modal contexts. (‘A man murdered Mrs Smith. The FBI think the police think he is from out of town.’) See e.g. Davies (1981), Evans (1985), Ludlow and Neale (1991), and Neale (1990). 77 Giving ‘only one donkey’ large scope and construing ‘it’ as a variable it binds yields the wrong result. Someone uttering (9) would not be saying that only one donkey satisWes John bought x and Paul vaccinated x, for that is compatible with John having bought two donkeys which the utterance of the original conjunction is not.

Neale, This, That, and the Other

127

Consider the following progression: (10)

every man who bought only one donkey had to pay for it with cash 0 (10 ) every man who bought only one donkey had to pay for the donkey with cash (1000 ) [every man who bought only one donkey]1 had to pay for the donkey he1 bought with cash. Utterances of (10) and (100 ) may well be understood as elliptical for utterances of (1000 ).78 The interesting point here is that the description for which the pronoun ‘it’ in (10) and the incomplete description in (100 ) are taken to be elliptical may itself contain a pronoun, moreover a pronoun understood as a bound variable. That is, the truth conditions of utterances of (10), (100 ), and (1000 ) might be captured using the following formula of our semi-formal language: (11)

[everyx : man x • [just oney : donkey y] x bought y] [they : donkey y • x bought y] x had to pay for y with cash.

(This is not an LF, of course, but it may correspond quite closely to the LF of (1000 ). See below.) The description is Russellian but relativized in the sense that uniqueness is relative to choice of man who bought only one donkey. This sort of relativization should occasion no surprise, and I embraced it in Descriptions. The processes at work in interpreting an incomplete description or a descriptive pronoun are pragmatic and

78 Again, as Evans (1977) shows, giving ‘only one donkey’ large scope and construing ‘it’ as a variable it binds yields the wrong result (in this an countless other cases). Someone uttering (10) would not be saying that only one donkey satisWes every man who bought x had to pay for x with cash. The pronoun ‘it’ is not functioning as a bound variable in (10) because it does not lie within the scope of its purported antecedent ‘only one donkey’; it is a D-type pronoun, and so is eVectively an ‘unbound anaphor’. At least that is the account I gave in Descriptions, and it still seems to me the best account going. (See also Davies (1981) and Ludlow and Neale (1991).) In some languages the distinction between bound and unbound pronouns shows up lexically. (i) and (ii) are ambiguous in English between bound and donkey readings according as /his/ is bound by the subject DP or ‘merely anaphoric’ on the embedded DP: (i) (ii)

[every man who has [a son]2 ]1 loves his1=2 wife [only [John]2 ]1 loves his1=2 wife.

In Icelandic, by contrast, /his/ translates as the reXexive possessive sı´na where binding by the subject DP is intended, but as the genitive hans where ‘mere anaphora’ with the embedded DP is intended: (i0 ) a. [se´rhver maður sem a´ [son]2 ]1 elskar konuna sı´na1 (‘[every man who has [a son]]2 ]1 loves his1 wife’) b. [se´rhver maður sem a´ [son]2 ]1 elskar konuna hans2 (‘[every man who has [ason]2 ]1 loves his2 wife’) (ii0 ) a. [aðeins [Jo´n]2 ]1 elskar konuna sı´na1 (‘[only [John]2 ]1 loves his1 wife’) b. [aðveins [Jo´n]2 ]1 elskar konuna hans2 (‘[only [John]2 ]1 loves his2 wife’). For discussion, see Neale (forthcoming b).

128

I. Incomplete Descriptions

richly inferential, and it is clear hearers have no trouble coming up with (1000 ) when quizzed about (10) or (100 ).79 The relativization in (10) and countless other donkey sentences—indeed in countless non-donkey sentences (in ‘every man loves the woman he married’, for example)— is sometimes called ‘covariation’, or ‘implicit binding’. The three labels are all trying to get at the same basic point: the interpretation of the donkey pronoun ‘it’ is relativized to or covaries with objects satisfying ‘man who bought only one donkey’. The label ‘implicit binding’ is, perhaps, slightly more loaded than the other two as it may lull (or help those with an agenda lure) the unsuspecting into thinking the concept of concern is a syntactic one, that it concerns the explicit binding of an aphonic variable at LF. And that is certainly not the case. In their long discussions of this type of implicit binding, Evans (1977), Davies (1981), Ludlow and Neale (1991), and Neale (1990, 1993a, 1994) discuss the phenomenon and do not posit special variables in underlying syntax for the quantiWers to bind. The last four of these works are highly sympathetic to LF, but there is no commitment to aphonic variables at LF being so-bound, and no claim that (10), (100 ), and (1000 ) share an LF.80 So what do the LFs of (10), (100 ), and (1000 ) look like? This is something I took up in Neale (1993). Given the role LFs are supposed to play in syntactic theory, I suggest roughly the following:81 [every 1 man1 who3 [[ [just one]2 donkey 2 ]2 [S e3 bought e2 ]]1 ]1 [it]2 [S e1 paid for e2 with cash] 0 (12 ) [every 1 man1 who3 [[ [just one]2 donkey 2 ]2 [S e3 bought e2 ]1 ]]1 [the2 donkey 2 ]2 [S e1 paid for e2 with cash] (1200 ) [every 1 man1 who3 [[ [just one]2 donkey 2 ]2 [S e3 bought e2 ]]1 ]1 [the2 donkey 2 e1 owns e2 ]2 [S e1 paid for e2 with cash].

(12)

79 Neale (1990, 1993a, 1994) toyed with the idea of specifying the descriptive content of D-type pronouns using a simple algorithm and then pointed to problems suggesting retrieval was a looser pragmatic matter, perhaps guided or shaped by formal factors and strong interpretive heuristics. 80 It is odd that Stanley and Szabo´ (2000a, b) and Stanley (2000, 2002a, b) assume without argument that relativization (implicit binding) requires an actual variable at LF, given the existence of accounts of relativization in the literature that do not postulate aphonic variables at LF that the subject quantiWer binds. What Stanley and Szabo´ do argue is that there are all sorts of examples that involve relativization. But that is something people who discuss relativization already knew. For a while I thought perhaps an implicit argument was lurking behind Stanley and Szabo´’s assumption: it is impossible to entertain quantiWed thoughts without entertaining quantiWed natural language sentences containing variables that the quantiWers bind—the evidence would have to come from psychology, and I am aware of none. But even if this turned out to be true, would it be relevant? Surely I can utter a sentence X and thereby conversationally implicate something we might describe using a quantiWed sentence Y, but that does not mean X itself must contain a variable for the quantiWer in Y to bind. It is hard to believe that relativized interpretations of utterances are going to present problems for general pragmatic processes given that these processes must also be capable of revealing conversational implicatures, irony, metaphor, and so on. 81 I here assume with Evans (1977) and Quine (1960) that a relative pronoun is a device of predicate abstraction, the subscript on its relative clause indicating the predicate’s argument. Thus the subject DPs are construed as [everyx1: man(x1 ) • [lx3 [just one xa : donkey(x2 )]x3 bought (x2 )](x1 )]

Neale, This, That, and the Other

129

In (12) ‘it’ has been raised, which goes hand-in-hand with its interpretation as a quantiWer, indeed as a description, a full descriptive content which the speaker expects the hearer to come up with pragmatically, just as with the incomplete description in (120 ).

9. Incompleteness and Persistence In order to bring into the picture certain mathematical concepts of importance, let us idealize in certain ways temporarily, abstracting from the context-sensitive and elliptical ways of ordinary speech. Following Barwise and Cooper (1981), let us say that a determiner D is persistent just in case the truth of (1)

D A(s) is (are) B(s)

(e.g. ‘some ants are black’) guarantees the truth of (2)

D A0 (s) is (are) B(s)

(e.g. ‘some animals are black’) in any case where the set of things that are A0 (e.g. animals) properly includes the set of things that are A (e.g. ants). So whereas ‘some’ is persistent (witness the fact that if ‘some ant is black’ is true, so is ‘some animal is black’), ‘every’ is not (witness the fact that the truth of ‘every ant is black does’ does not guarantee the truth of ‘every animal is black’). In Descriptions, I noted in passing that simple cases of matrix incompleteness arise naturally in connection with uses of non-persistent determiners such as ‘every’, ‘no’, and ‘the’, whilst ‘derived’ or ‘inverted’ cases arise naturally in connection with uses of persistent determiners such as ‘a’ and ‘some’.82 That note exhausted what I felt needed to be said at the time because I did not then see incompleteness as particularly interesting or threatening to Russell’s Theory of Descriptions. Subsequent literature has convinced me the phenomenon is more interesting than I once thought, and while writing the present work and reviewing the footnotes that were culled from Descriptions before it was Wrst published, I came across two that went into rather more detail; realizing their importance to the current discussion, I have distilled them for use here. Simple cases exemplifying the problem of incomplete matrices arise naturally in connection with the use of quantiWers that are non-persistent, such as ‘every’, ‘no’, ‘the’; inverted cases arise naturally in connection with those that are persistent, such as

82 Neale (1990: 114 n. 45). Actually, derived cases arise naturally in connection with determiners that are not anti-persistent, an interestingly wider group. To foreshadow: (i) since ‘the’ is both non-persistent and non-antipersistent, simple and inverted cases are readily produced; (ii) perhaps the (overlapping) categories non-persistent and non-anti-persistent (rather than the (non-exhaustive) categories anti-persistent and persistent) is the one at the core of the generalized quantiWer analysis of natural language determiners.

130

I. Incomplete Descriptions

‘some’. The general problem of incompleteness, recall, is to explain how it is that a matrix A(x) is understood, in context, as satisWed by fewer objects than its superWcial form seems to require. The contrast between the implicit and explicit approaches is best summarized thus: on the implicit approach, A(x) is satisWed by fewer objects because it is understood as ranging over fewer objects; on the explicit approach, it is satisWed by fewer objects because it is understood as abbreviating A(x) • C (x). The problem has simple and inverted forms in the following sense. In the simple case, an utterance of ‘D A is B’ is understood, in context, as expressing a truth, despite the evident falsity of the formula [Dx: A(x)]B(x) (taken at face value). Examples, in context, might be the following: (3) (4) (5)

the table is covered with books [false—there are millions of tables] every student swam the Hellespont [false—millions of students have never been to Turkey] no student lives in China [false—millions of Chinese students live in China].

In inverted cases, an utterance of ‘D A is B’ is understood, in context, as expressing a falsehood despite the evident truth of [Dx : A(x)]B(x) (taken at face value).83 An obvious way of producing an inverted case is to introduce negation with large scope. Examples, in context, might be the following: (30 ) (40 ) (50 )

it is not the case that the table is covered with books [true—there are millions of tables] not every student swam the Hellespont [true—millions of students have never been to Turkey] it is untrue that no student lives in China [true—millions of Chinese students live in China].

Alternatively, given familiar relations between quantiWers, inverted cases can be generated without negation using ‘some’ or ‘a’, which are persistent: (6) (7)

some students live in China there’s a bottle in the fridge.

Suppose you are in my kitchen and ask me for a beer; I reply by uttering (7); you open the refrigerator and Wnd a bottle of ketchup but no bottle (or can) of beer. Now it is certainly true that there is a bottle in my fridge, but not true that there is a bottle of beer in there. And so there is some inclination to say that I made a false statement, in this context, when I said ‘there’s a bottle in the fridge’. 83 My use of ‘inverted’ in connection with forms of the Argument from Incompleteness is intended to mirror my use of the same word in connection with the Argument from Misdescription. In the simple case of the latter argument, used by Donnellan (1966), an utterance of ‘the A is B’ seems to be true despite the apparent falsity of its Russellian analysis. In the inverted case of the argument, used by Hornsby (1978), an utterance of ‘the A is B’ seems to be false despite the apparent truth of its Russellian analysis.

Neale, This, That, and the Other

131

Finishing oV our square, we should Wnd simple cases again if we introduce negation with large scope over persistent quantiWers; and in fact we do: (60 ) (70 )

It is not the case that some student lives in China There is not a bottle in the fridge.

I want to appeal to certain mathematical properties of quantiWcational determiners, in particular persistence, in so far as they bear on the matter of the interpretation of the determiners ‘the’, ‘this’, ‘that’, and ‘a’.84 Thoughts about incompleteness should be cast aside for a while, as I want to talk in the abstract manner of the mathematical logician prior to raising issues about, say, domain restrictions when discussing utterances of natural language sentences. For immediate purposes, then, two harmless Wctions or idealizations will be maintained to facilitate a clear exposition of the main points and avoid distracting side issues.85 First, the matrices used in the examples should be taken at face value, as requiring no completion. Second, each unambiguous English sentence should be taken to map onto exactly one truth-evaluable formula of the language of restricted quantiWcation. A formula [Dx : A(x)]B(x) contains two open formulae, A(x) and B(x). The determiner Dx combines with A(x) to form a restricted quantiWer, and that quantiWer combines with B(x) to form a formula.86 Let us say that A(x) occupies ‘Wrst’ position with respect to Dx while B(x) occupies ‘second’ position in the formula [Dx : A(x)]B(x). We can now examine the impact particular determiners have on these positions in some detail. Talk of persistent quantiWers can be extracted straightforwardly from Barwise and Perry’s talk of persistent statements. The basic idea is this: a quantiWer [Dx : A(x)] is persistent if and only if a simple sentence containing it remains true if [Dx : A(x)] is replaced by [Dx : A0 (x)], where the set of things satisfying A0 (x) properly includes the set of things that satisfy A(x), as for example the set of animals properly includes the set of ants. Thus ‘some ants’ is persistent because if, say, ‘some ants are black’ is true, so is ‘some animals are black’. It will be useful to specify this more precisely.

1. On Being ("1) A determiner D is "1 (‘one-up’, ‘upward entailing in Wrst position’, ‘persistent’), if the following inference is valid (for arbitrary A, A0 , and B):

84 The general interest of these properties to semantics and syntax is discussed in detail by Barwise and Cooper (1981). See also Westersta˚hl (1985a, 1989). 85 When I say the Wctions or idealizations are harmless, I mean only that they are harmless for the particular concerns I have—they can be very harmful in other contexts. 86 In an alternative system of ‘binary’ rather than restricted quantiWcation, Qx might combine directly with A(x) and B(x) in one move to form a formula [Qx ] (A(x), B(x)), but as a way of having our notation closer to the structure of English I have opted to treat A(x) and B(x) diVerently.

I. Incomplete Descriptions

132

("1)

[Dx : A(x)]B(x) [everyx : A(x)]A0 (x) [Dx : A0 (x)]B(x).

Thus: (8)

[somex : ant(x)] black(x) [everyx : ant(x)] animal(x) [somex : animal(x)] black(x)

Examples: ‘some’, ‘an’, ‘at least n’, ‘a few’ (unlike ‘few’). Quick test in natural language: if det is "1 then (90 ) is entailed by (9), (90 ) (9)

D animal(s) is (are) black D ant(s) is (are) black ("1)

(The determiner ‘some’ may take a singular or a plural complement (‘some ant’, ‘some ants’); either way, it is "1.) The simplest form of the problem of incompleteness arises in connection with determiners that are non-persistent, "1 (‘not one-up’). It is evident from consulting (9) "1. (Like ‘some’, the determiner ‘no’ may take a and (90 ) that ‘every’ and ‘no’ are singular or a plural complement (‘no ant’, ‘no ants’); either way, it is "1).87 The numerical determiners ‘exactly n’ and the proportional determiner ‘most’ are also "1 (unlike ‘at least n’). More importantly for our concerns, on the Russellian account we are assuming, ‘the’ is also "1. Instantiating (90 ) and (9), we get (100 ) and (10): (100 ) (10)

the animal is black the ant is black ( "1)

Unpacking these as (110 ) and (11), in Russell’s manner, makes it clear that (100 ) is not entailed by (10): (110 ) (11)

there is exactly one animal and every animal is black. there is exactly one ant and every ant is black. ( "1)

Like ‘some’ and ‘no’, the deWnite article may take a singular or a plural complement (‘the ant’, ‘the ants’); and, naturally enough, ‘the’ remains "1 when it takes a plural complement. This is worth exploring. A useful way of thinking about Russell’s Theory of Descriptions is as follows. Whereas the determiners ‘every’ and ‘all’ introduce universal quantiWcations and the determiners ‘a’, ‘an’, and ‘some’ introduce existential quantiWcations, the determiner ‘the’ simultaneously introduces both. For some purposes, it is useful to think of deWnite descriptions as complex existential phrases; for others it is more useful to think of them 87 So is ‘all’, which could, perhaps, be seen as a syntactically plural form of ‘every’, though nothing I shall say here turns on such an assumption.

Neale, This, That, and the Other

133

as complex universal phrases. For example, if one focuses on the existential character of Russell’s proposal it is easy to explain it to the novice by building upon Russell’s proposal for indeWnite descriptions because a sentence of the form ‘an A is B’ is analysed by Russell as (12), whilst one of the form ‘the A is B’ can be analysed as (13), which is logically equivalent to Russell’s preferred formula: (12) (13)

9x(Ax • Bx). 9x((Ax • Bx) • 8y(Ay y ¼ x)).

But if one focuses on the universal character of the proposal, the relationship between singular and plural deWnite descriptions comes more clearly into view. There is just a cardinality diVerence between the truth condition of an utterance of ‘the A is B’ and that of an utterance of ‘the As are Bs’: the former is true if and only if every A is B and there is exactly one A; the latter is true if and only if every A is B and there is more than one A.88 We can now conWrm that ‘the’ remains "1 when it takes a plural complement. Instantiating (90 ) and (9) again, we get (140 ) and (14): (140 ) (14)

the animals are black the ants are black ( "1)

Unpacking these as (150 ) and (15) makes it clear that (140 ) is not entailed by (14): (150 ) (15)

there are two or more animals and every animal is black. there are two or more ants and every ant is black ( "1)

So the simple form of the problem of incompleteness should arise naturally in connection with uses of both ‘the A’ and ‘the As’, which in fact it does. By contrast, the inverted problem should arise naturally in connection with uses of both ‘some A’ and ‘some As’, and for uses of indeWnite descriptions of the forms ‘an A’ and ‘a few As’. It is frequently suggested that demonstrative descriptions (complex demonstratives) of the forms ‘this A’ and ‘that A’ are special forms of deWnite descriptions. But the persistence test suggests they are, at least when used demonstratively, special forms of indeWnite descriptions. Instantiating (90 ) and (9) again, we get (160 ) and (16): (160 ) (16)

that animal is black that ant is black ("1)

In order for the intuitive test to make sense, two assumptions must be made. First, ‘that A is B’ is not true unless the object ‘that A’ denotes is an A; second, the occurrences of 88 The determiner ‘both’ is surely the dual form of ‘the’: ‘both As are Bs’ is true if and only if every A is B and there are exactly two As. Singular, dual, and plural truth conditions may be put into a set-theoretic notation that renders the relations between them transparent ‘the A is B’ is true if and only if jABj ¼ 0 and jAj ¼ 1 ‘the As are Bs’ is true if and only if jABj ¼ 0 and jAj > 1 ‘both As are Bs’ is true if and only if jABj ¼ 0 and jAj ¼ 2.

134

I. Incomplete Descriptions

‘that animal’ and ‘that ant’ in (160 ) and (16) are being used to denote the same object. (The assumptions seem harmless if we are interested merely in the codiWcation of inference.) Since (160 ) is entailed by (16) on these assumptions, ‘that’ is "1 and so patterns with ‘a’ and not with ‘the’ (which is "1).89 The same is true of the plural form ‘those’, and of ‘this’ and its plural form ‘these’. So, for example, (170 ) is entailed by (17): (170 ) (17)

these animals are black these ants are black ("1)

In summary, while ‘the’ is "1, ‘a’, ‘that’, ‘this’, ‘these’, and ‘those’ are "1.

2. On Being (#1) One way for a determiner to be "1 is for it to be #1 (‘one-down’, ‘downward entailing in Wrst position’, or ‘anti-persistent’). A determiner D is #1 if and only if the following is valid: (#1)

[Dx : A0 (x)]B(x) [everyx : A(x)]A0 (x) [Dx : A(x)]B(x)

Examples: ‘every’, ‘no’, ‘at most n’. Quick test: if D is #1 then (90 ) entails (9) (rather than being entailed by (9), as in the "1 case): (90 ) (9)

D animal(s) is (are) black D ant(s) is (are) black

(#1)

What of the deWnite article? We have already seen that it is "1. Is it "1 because it is #1? Or is it #1 as well as "1? Along with ‘exactly n’ it is #1, as demonstrated by the fact that (100 ) does not entail (10), and (140 ) does not entail (14): (100 ) (10) (140 ) (14)

the animal is black the ant is black the animals are black the ants are black

(#1) (#1)

So ‘the’ and ‘exactly n’ are both " #1 (‘one-Xat’). This should come as no surprise: ‘exactly n’ is " #1 because ‘at least n’ is "1 whilst ‘at most n’ is #1; and ‘the’ is " #1 because ‘some’ is "1 while ‘every’ is #1 (think about it set-theoretically). Certain derived partitive quantiWers are also " #1: ‘exactly n of the’, ‘half of the’, and ‘most of the’, as well as the simple proportional determiner ‘most’, as the reader can easily verify. 89 Actually matters are slightly more complicated than indicated here. One issue is highlighted by the fact that demonstrative descriptions are sometimes used in ways that require them to be understood as Russellian deWnite descriptions, and on such uses the determiner ‘that’ is "1 rather than "1. This would seem eVectively to preclude the simplest unitary accounts of the determiner ‘that’.

Neale, This, That, and the Other

135

The important point to bear in mind when we return to incompleteness is that simple cases arise naturally in connection with determiners that are "1, such as ‘the’, ‘every’, and ‘no’, and inverted cases occur naturally in connection with those that are #1, such as ‘the’ (again), ‘some’, and ‘a’—the referential commitments of ‘this’ and ‘that’ rule out signiWcant talk of incompleteness. Let us now turn to the impact of determiners on second position.

3. On Being ("2) D is "2 (‘two-up’, ‘upward entailing in second position’, ‘monotone increasing’) if and only if the following is valid: ("2)

[Dx : A(x)]B(x) [everyx : B(x)]B 0 (x) [Dx : A(x)]B 0 (x)

Examples: every and some. Quick test: if det is "2 then (180 ) is entailed by (18): (180 ) (18)

D ant(s) is (are) black or brown D ant(s) is (are) black.

("2)

4. On Being (#2) Dx is #2 (‘two down’, ‘downward entailing in second position’, or ‘monotone decreasing’) if and only if the following is valid: (#2)

[Dx : A(x)]B 0 (x) [everyx : B(x)]B 0 (x) [Dx : A(x)]B(x)

Examples, no and few. Quick test: if det is #2 then (180 ) entails (18). It is clear from the fact that ‘some’ and ‘every’ are both "2 that on a Russellian analysis ‘the’ will also be "2 (think about it set-theoretically), and the relevant test conWrms this. Instantiating (180 ) and (18) we get (190 ) and (19): (190 ) (19)

the ant is black or brown. the ant is black ("2)

And the latter entails the former, as the Russellian unpacking makes clear: (200 ) (20)

there is exactly one ant and every ant is black or brown. there is exactly one ant and every ant is black ("2)

In this respect, ‘the’ diVers from ‘exactly one’, ‘just one’, and ‘one and only one’, which are " #2.

136

I. Incomplete Descriptions

The fact that ‘a’ and ‘some’ are "2 suggests that ‘this’ and ‘that’ (and their plural forms) will be "2. Again, the relevant test provides conWrmation (making the same assumptions as before). For example, (21) entails (210 ): (210 ) (21)

that ant is black or brown that ant is black

("2).

We have identiWed ten properties. Six of them can be characterized succinctly as follows: #1 if we can move ‘down’ from truth to truth, "1 if we can move ‘up’, " #1 if neither: (#1) det animal(s) is (are) black ("1) det ant(s) is (are) black. det is #2 if we can move ‘down’ from truth to truth, "2 if we can move ‘up’, " #2 if neither: (#2) det ant(s) is (are) black or brown ("2) det ant(s) is (are) black.

det is

Like " #1 and " #2, the four further negative properties "1, #1, "2, and #2 can be characterized in terms of negation, conjunction, and #1, "1, #2, and "2, which the table takes as ‘primitive’. But it is not obvious this is the most illuminating way of organizing things, at least if meaning (as opposed to inference) is our main concern. Perhaps we are better oV viewing "1, #1, "2, and #2 as ‘primitive’ (as not allowing true up and down moves) and seeing the rest as Boolean products. As far as deWnite, indeWnite, and demonstrative descriptions are concerned, matters may be summarized thus: "1 a, this, that " #1 the "2 the, a, this, that Ladusaw (1980) proposes that negative polarity items (e.g. ‘ever’ and ‘only’) occur felicitously only in # environments. If this is correct, such items are precluded from both positions of ‘the’, which is " #1 and "2. May (1985) accepts Ladusaw’s hypothesis and a Russellian semantics for singular ‘the’, but suggests plural descriptions may contain negative polarity items in Wrst position. This would require ‘the’ to be merely "1 with a singular complement but fully #1 with a plural. Since other determiners (e.g. ‘no’, or some’) do not alter in respect of persistence in this way, something seems to be wrong here. Perhaps Ladusaw’s proposal should invoke " rather than #. Is this supported by the fact that ‘most’ is " #1? (Richard Larson has pointed out to me, issues about partitives intrude here.) The issues with demonstratives are also unclear: ‘that’ is "1 and should not license negative polarity items in Wrst position on either the original

Neale, This, That, and the Other

137

or revised proposal. Ludlow’s (2002) fresh insights on the vexed matter of the bearing of "# properties on the nature of syntactic theory are important here.

10. Sentence Ellipsis (1) and (10 ) may be said to share the same LF and diVer only at PF: (1) (10 )

John can waltz, but Paul can’t. John can waltz, but Paul can’t waltz.

In (1), the PF results from the deletion of a VP (verb phrase) under stringent conditions of ‘identity’ that guarantee its ‘recoverability’. In order to understand what is at stake here and later in this chapter, some history is helpful. The main ideas can be sketched without too much detail, although the nature of the battles of the 1960s and early 1970s means that much is still disputed. Early generative grammar posited two levels of syntactic representation, usually called deep structure and surface structure. Deep structures were essentially the products of a lexicon and phrase structure rules such as the following: S ! NP þ VP VP ! V þ NP. Phrase structure rules were supplemented by transformational rules, which were meant to bridge the gap between deep structure and surface structure, and explain the relationships between certain sentences (for example, active–passive pairs) by deleting, adding, and reordering elements in phrase markers (phrase structure trees) under clearly speciWed conditions.90 Katz and Postal (1964) put forward the hypothesis (often called ‘Katz-Postal’) that transformational rules did not alter meaning, from which it virtually followed that deletion transformations could take place only under conditions that made ‘recovery’ possible, something Chomsky stressed the same year: A transformation can delete an element only if this element is the designated representative of a category, or if the structural condition that deWnes this transformation states that the deleted element is structurally identical to another element of the transformed string. A deleted element is, therefore, always recoverable. (1964: 41)

90 Deep structures were meant to be generated by sets of context-free phrase structure rules, i.e. rules of the form, (i) a ! b1 . . . bn . These may be contrasted with context-sensitive rules of the form (ii) XaY ! Xb1 . . . bn Y, which specify that if a occurs in the context XaY it can expand to b1 . . . bn . The mapping between deep structure and surface structure was eVected by context-sensitive transformational rules. In the philosophy of language a related distinction can be found, between compositions that are what Davidson calls semantically innocent (i.e. context invariant) and compositions that are not.

138

I. Incomplete Descriptions

We are sorely in need of examples. In (2), ‘John’ is the subject of the main clause and ‘Paul’ is the subject of the embedded clause (here inWnitival): (2)

John wants Paul to leave.

Compare (2) with (3), where there appears to be no subject of the embedded clause: (3)

John wants to leave.

One early and widely discussed transformational deletion was Equi NP deletion, where the embedded clause of a verb like ‘want’ was deleted on account of being ‘identical’ to the subject of the main verb. The rough idea was that (3) was derived from the underlying form (30 ): (30 )

John wants John to leave.

(It could not be not be derived from, say, (2).) In short, we have something we might call ‘recoverability of deletion under identity’, although we need to specify exactly what is meant by ‘identity’ here. Recoverability was a fundamental condition governing deletion. To abandon it—as some generative semanticists did—was to abandon serious work on the syntax of natural language.91 Obviously there are all sorts of issues to take up, even in connection with this single example—is the deletion obligatory or optional? is identity to be construed as a formal or interpretive notion? And so on. At Wrst blush, the second question seems to have a clear answer. If I have two friends called John, even if I am being unclear, surely I am not speaking ungrammatically when I utter (30 ), intending to refer to one of the friends using the subject of the main verb, and to the other using the subject of the embedded clause. Only when I intend the main and embedded subjects to refer to the same person does Equi take place.92 So interpretive identity is required (as one would expect, assuming Katz-Postal);93 Something like formal identity also seems to be required, however, for otherwise (4) could be the result of Equi applying to (40 ) when ‘Cicero’ and ‘Tully’ share a subscript: (4) (40 )

Cicero wants to leave Cicero wants Tully to leave.

91 Eight years after Chomsky’s remark appeared in print, the premier linguistics journal Linguistic Inquiry, published a piece by Fiengo and Lasnik called ‘On Nonrecoverable Deletion in Syntax’ (1972: 528). There was no text, just the authors’ names and their institutional aYliation (they were both then at MIT) and a blank space. 92 In the right circumstances could I not use (30 ) intentionally to talk about a single individual? ‘I want Tom to leave, you want Tom to leave, Mary wants Tom to leave, Fred wants Tom to leave, Tom wants Tom to leave, but he can’t until the curtain comes down.’ Perhaps the fact that heavy stress is needed on the embedded occurrence indicates that a rule of sentence grammar is being stretched, however. 93 It was not appreciated at the time that even under interpretive identity (3) and (30 ) are not equivalent. In current theory, the subject of ‘want is an aphonic PRO. ‘John wants PRO to leave’ must be read de se, meaning PRO cannot be interpreted as a copy of ‘John’.

Neale, This, That, and the Other

139

On the assumption this is not something we want from a grammar it would seem that the operation of Equi is constrained by formal and interpretive identity, so let us use ‘identity’ in this way in this context. A type of deletion of more theoretical interest is VP deletion. VP deletion raises questions about the ‘scope’ of transformational rules and about variable binding and the interpretation of pronouns. Since the second of these will assume some importance later, it is good to get clear about the main issue right away.94 VP deletion appears to take place in the derivation of (1) from (10 ):95 (1) (10 )

John can waltz, but Paul can’t. John can waltz, but Paul can’t waltz.

The deletion is recoverable because it takes place only under identity: (1) cannot be obtained by VP deletion from (and hence cannot be understood in terms of ), say, (5) or (6): (5) (6)

John can waltz, but Paul can’t tango. John can waltz, but Paul can’t speak German.

Recoverability and Katz-Postal respected again. It is no more a ‘merely pragmatic’ fact that (1) must be read as equivalent to (10 ) than is the fact that ‘himself ’ must be read as bound by ‘John’ in (7)

John told Paul’s wife a lot about himself.96

VP deletion raises two immediate questions. First, transformational rules were meant to operate on sentences, but VP deletion appears not to respect sentence boundaries, utterance boundaries, or even speaker boundaries, witness the following natural dialogue between A and B. (8)

A: John can waltz. B: I know. It’s a shame Mary can’t.

This would appear to show that the simplest transformational account of VP deletion is in trouble, at least on a standard account of what transformational rules are, that the phenomenon is a discourse phenomenon, albeit one that is grammatically

94 I discussed VP deletion in several footnotes in Descriptions, but what I said was mostly rubbish. 95 For high-level introductions to VP ellipsis, see ch. 9 of see Heim and Kratzer (1998) and May (2002). It is common to distinguish VP ellipsis from VP anaphora. Examples involving the latter would be (i) and (ii), which contain the anaphoric elements ‘it’ or ‘so’: (i) Bill can waltz, but Tom can’t do it; (ii) Bill can waltz, and so can Tom. On the diVerence between VP anaphora and VP ellipsis, see May (2002). 96 There are good reasons for thinking that reXexives are not referential expressions but devices that combine with predicates to form larger predicates, making ‘John loves himself ’ equivalent to ‘John is a self-lover’. On the assumption that an intransitive verb ‘snores’ is understood as (lx (x snores) ), and an intransitive verb ‘loves’ as (ly(lx (x loves y) ), this will drop out on a suitable account of reXexives. See Neale (forthcoming a).

140

I. Incomplete Descriptions

constrained. As May (2002: 1095) succinctly puts it, to the extent that VP deletion can be thought of as a rule, it would appear to be more of a rule of discourse grammar than sentence grammar. There is no need to pursue this matter here. Let us simply note that syntactic theory has evolved considerably since the heyday of transformations. In Chomsky’s (1995, 2000) current set-up we might say that (1) and (10 ) have the same LF, (1l ), but diVerent PFs, the PF of the former given by something like (1p ), after VP ellipsis:97 (1l ) (1p )

[S Bill [can [VP waltz]] but [S Tom [can’t [VP waltz]]] [S Bill [can [VP waltz]] but [S Tom [can’t [VP e]]].

Katz-Postal can be transposed accordingly: interpretation remains constant under linguistic ellipsis (henceforth, this is what I shall mean by ‘Katz-Postal’). The second thing VP ellipsis brings out is an important ambiguity, independently motivated and, in eVect, predicted by Katz-Postal. There is only one reading of (9), but there are two readings of (10):98 (9) (10)

John loves his wife, but Mary doesn’t John loves his wife, but Paul doesn’t.

On one reading of (10) Paul is being said not to love John’s wife, on the other not to love his own wife. (The former is usually called the strict reading, the latter the sloppy reading.) The existence of the two readings of ‘Paul doesn’t’ as it occurs in (10) follows from the fact that on standard accounts of the interpretation of pronouns there are two distinct, but truth-conditionally equivalent readings of ‘John loves his wife’: (11) (12)

John (lx(x loves John’s wife)) John (lx(x loves x’s wife)).

The fact that there are two ways of construing (10) does not jump out at us until we look at inference patterns or consider more complex examples containing ‘John loves his wife’ as a proper part, (10) for example, or an attitude report like (13):99 (13)

Mary believes that John loves his wife.

Assuming Katz-Postal, we can now see what is going in (10). On one reading, the occurrence of ‘his’ in the Wrst clause is being used to refer independently to John; on the other reading ‘his’ is bound by ‘John’. If the former, the recovered predicate is the 97 Strictly speaking, (1p ) cannot be a PF in Chomsky’s current theory because it is still annotated for phrasal structure. I think Chomsky has been misunderstood on this matter. On the sort of pure theory Chomsky is articulating, PF is just a set of instructions to the sound system and unless empirical evidence to the contrary emerges the null hypothesis is that these instructions need no access to information about phrase structure of the sort present at LF. To say this is not to say that the instructions constituting PF have been derived without information about phrase structure. 98 Such examples are discussed extensively by Sag (1976) and Williams (1977). For recent, user-friendly discussion, see Heim and Kratzer (1998) and May (2002). 99 See Soames (1990). Similarly with examples like ‘Only John loves his wife’.

Neale, This, That, and the Other

141

one in (11); if the latter, the one in (12). Either way, we have recoverability. And we can also elegantly explain what is going on in an example like (14), which combines inference and VP deletion: (14)

Every Englishman loves his wife, but John doesn’t, so John can’t be English.100

Three Wnal, quick examples of syntactic ellipsis will give us all we need. The Wrst is stripping, where the surface form (150 ) is optionally reduced to (15): (15) (150 )

Smith took a job in Uruguay after the war, or Paraguay. Smith took a job in Uruguay after the war, or Smith took a job in Paraguay after the war

In (15), everything but ‘Paraguay’ has been deleted (‘stripped’) from the second disjunct under identity (of the material surrounding ‘Uruguay’ and ‘Paraguay’). Again, the deleted material is recoverable. From (15) we cannot recover (16), for example: (16)

Smith took a job in Uruguay after the war, or Smith collected stamps from Paraguay after the war.

Relatedly, but via a process of syntactic ellipsis called gapping, (17) can be obtained from (170 ) but not from (18): (17) (170 ) (18)

Smith went to Uruguay, and Jones to Paraguay. Smith went to Uruguay, and Jones went to Paraguay. Smith went to Uruguay, and Jones sent his wife to Paraguay.

Another example, involving two distinct deletions, might be the derivation of (19) from (1900 ), perhaps via (190 ): (19) (190 )

one coVee with milk and one without. one coVee with milk and one coVee without

100 Some very interesting examples, discussed by Elbourne (2001), appear to demonstrate the intersection of sentence ellipsis and utterance ellipsis involving descriptions. (To Elbourne they demonstrate the need to see D-type pronouns as real descriptions at LF.) The facts are truly maddening. First oV, the VP ellipsis in (i) seems Wne, but the simple recovery (ii) is not: (i) every Englishman who bought a donkey vaccinated it (/the donkey), but John didn’t, so he can’t be English (ii) every Englishman who bought a donkey vaccinated it, but John didn’t vaccinate it (/the donkey), so he can’t be English. However, a recovery that contains a full description containing a bound pronoun is Wne: (iii) every Englishman who bought a donkey vaccinated it, but John didn’t vaccinate the donkey he bought, so he can’t be English. I really don’t know what to make of this. The idea that at LF these donkey pronouns really are full descriptions containing bound pronouns might seem to provide half the story: recovery is of what is present at LF (which, after all, is where the conditions for ellipsis must be satisWed); but why can’t the description in the recovered VP collapse into a D-type pronoun the way the occurrence in the donating VP does? I Wnd this utterly baZing. Perhaps Elbourne’s packed paper contains the answer but I have not had time to study it.

142

I. Incomplete Descriptions (1900 )

one coVee with milk and one coVee without milk.101

Again the deletions are recoverable, which explains why (19) cannot be derived from (and so understood as), say, (20): (20)

one coVee with milk and one tea without sugar.102

101 I am not here concerned with the issue of whether the derivation proceeds via (190 ) or via (i) ‘one coVee with milk and one without milk’. Either way, recoverability is not violated. The example takes on more interest in cases where deletions are based on previous utterances, as in the following: Smith: Two coVees, please. Jones: Milk? Smith: One with, one without. I have no Wrm opinion of how much syntactic theory itself is supposed to contribute to our understanding of such a dialogue, but it is clear that the only recoverable interpretation of Smith’s second utterance is as a request for one coVee with milk and one coVee without milk, and this strongly suggests syntax is implicated. Such examples seem problematic for Stanley’s (2000) bold claim that it is impossible to perform a linguistic speech act by uttering something less than a whole sentence. According to Stanley, every purported counter-example involves either (i) a case where a whole sentence is, in fact, being uttered, or (ii) a case in which no linguistic speech act is, in fact, being performed. An utterance of ‘water’ made in circumstances where it is indeterminate whether to construe the remark as tantamount to an assertion that the speaker wants water or a request for water does not (because of this indeterminacy) constitute a linguistic speech act, Stanley says. Presumably he would say the same about ‘one with and one without’: it is indeterminate whether the speaker is asserting that he wants one coVee with milk and one coVee without milk rather than requesting one coVee with milk and one coVee without milk. If the speaker has not performed a linguistic speech act, why is recoverability satisWed?. 102 It is sometimes claimed that syntactic deletion is involved in the following cases, where (ib) is derived from (ia), and (iib) from (iia): a. This donkey is faster than that donkey b. This donkey is faster than that one (ii) a. This donkey is faster than the other donkey b. This donkey is faster than the other. (i)

But here recoverability is violated. While it is perfectly natural to interpret an utterance of (ib) as equivalent to an utterance of (ia), this is not forced. Suppose we have been comparing a few horses in respect of speed. After a while Smith says to me, while pointing, ‘I think I’m going to buy this one; he’s a bargain’. I reply, pointing Wrst at the horse Smith is thinking of buying, and then at my donkey, ‘Not that one, surely! This donkey is faster than that one.’ The word ‘donkey’ needs to be stressed, I think, but the example still seems to show that recoverability is violated and hence that the occurrence of ‘one’ in (ib) is no more obtained by syntactic deletion than the occurrence of ‘one’ in (iii): ‘I’ll take this one, please.’ (Imagine a shopper who has been examining several items and is aware that a sales assistant is watching. The sales assistant holds back until he is reasonably sure the shopper has made a choice, at which point he goes over and says to the shopper, ‘Can I help you, sir?’ The shopper replies by uttering (iii).) What example (i) seems to show—example (ii) is not relevantly diVerent, but notice that ‘the other’ and ‘the other donkey’ are both incomplete descriptions—is that the interpretation of ‘one’ cannot be determined automatically from grammatical considerations alone. In many contexts there will be a natural tendency to interpret an utterance of (ib) as equivalent to an utterance of (ia), but that interpretation is not forced by grammar. While there may be heuristics of interpretation that draw on syntactic information to provide default interpretations of utterances of ‘one’ and also utterances of some pronouns, when we Wnd that the hypothesized default can be over-ridden very naturally in a particular syntactic structure, we have found a structure that does not provide enough syntactic information to sustain recoverability in the required sense, and hence a structure that we cannot see as obtained by deletion. (Two interesting examples from 1960s popular music, the Wrst due to Dylan (1965), the second due to Lennon and McCartney (1966): (iv) ‘The cops don’t need you and, man, they expect the same’; (v) ‘She told me she worked in the morning and started to laugh / I told her I didn’t and crawled oV to sleep in the bath.’ (iv) involves what is often called VP anaphora because of the presence of the seemingly anaphoric element ‘the same’. (v) is particularly interesting given the various ambiguities, expectations about when sleep takes place, and other lines in the song.)

Neale, This, That, and the Other

143

It should be abundantly clear why sentence ellipsis could be of no relevant interest to anyone who agrees with Quine and Sellars that utterances of ‘the table is large’ are elliptical and as such do not counterexemplify Russell’s Theory of Descriptions. For there is nothing in the old idea that such utterances are elliptical to suggest the sentence ‘the table is large’ is itself elliptical in the syntactician’s sense.103 Consider (21)–(23): (21) (22) (23)

[S [DP the table] [VP is dirty]]. [S [DP every table] [VP is dirty]] [S [DP no table] [VP is dirty]]

The idea of recovering syntactic deletions here is absurd; indeed invoking syntactic deletion here would involve making a category mistake, and it is hardly surprising no philosopher who sees utterances of these sentences as involving utterance ellipsis maintains that, for example, the phrase marker (21) is derived from an underlying phrase marker such as (210 ), (210 )

[S [DP the table we are sitting at] [VP is dirty]].

Such philosophers are not even talking about the sentence speciWed by (21) being elliptical but about particular dated utterances being elliptical.104

11. The Argument from Anaphora We now have everything we need to discuss a syntactically sophisticated version of the Argument from Anaphora.105 Consider (1): (1)

The man in the gabardine suit is a spy. He tried to bribe me.

The original version of the argument, which traces back to Strawson (1952) in its basics, goes like this. (i) The occurrence of ‘he’ can be understood as anaphoric on the occurrence of ‘the man in the gabardine suit’. (ii) If an occurrence of a pronoun b is anaphoric on an occurrence of another expression a, then b is either a variable bound by a or a device that inherits its reference from a. (iii) The occurrence of the pronoun ‘he’ in (1) is not a bound variable. (iv) Therefore, it inherits its reference from the occurrence of ‘the man in the gabardine suit’. (v) Therefore, ‘the man in the gabardine

103 Of course, an account of how we manage to interpret elliptical utterances will be needed at some point, just as a theory of how we interpret proper names and pronouns and so on will be required, and a theory of how we grasp what a speaker is conversationally implicating by his utterance. But it is hardly an objection to the (obvious fact) that there are elliptical utterances that we do not yet have a pragmatic-inferential theory capable of making clear empirical predictions. 104 For detailed textual discussions, see below. 105 The reconstruction and the criticism of the argument are taken from Ludlow and Neale (1991) and Neale (1990).

144

I. Incomplete Descriptions

suit’ in (1) is a referring expression. Mutatis mutandis for the indeWnite description ‘a cat’ in (2): (2)

A cat is on the lawn; it looks like a stray to me.

The argument is easily defeated. If, as Kripke (1977) and Lewis (1979) suggest, a (putatively) anaphoric pronoun may refer to an individual raised to salience by the utterance of a description, then either (i) or (ii) can be rejected and an alternative account provided of the referential nature of the pronouns in (1) and (2). Or, if, as the D-type theory maintains, ‘he’ and ‘it’ are just incomplete descriptions, (1) and (2) might be seen as elliptical for (10 ) and (20 ) respectively: (10 ) (20 )

The man in the gabardine suit tried to bribe me the cat on the lawn looks like a stray to me.

In which case the pronouns are not referential, and again (ii) is rejected. Before looking at the syntactically sophisticated version of this argument, we need to pull a few things together. It is a familiar point, noted earlier, there are two readings of /Paul doesn’t/ as it occurs in (3): (3)

John loves his wife, but Paul doesn’t.

On one reading of (3) Paul is being said not to love John’s wife, on the other not to love his own wife (the strict reading and the sloppy reading). It was also noted that the existence of the two readings follows from the fact that on one plausible interpretation of pronouns there are two distinct, but truth-conditionally equivalent readings of (4). (4) John loves his wife (40 ) John lx(x loves John’s wife) (400 ) John lx(x loves x’s wife). On the reading given by (40 ), the occurrence of /his/ in (4) is being used to make free (independent) reference to John; on the reading given by (400 ) /his/ is bound by /John/, giving us the reading upon which (4) is a singular instance of (5) on its bound variable reading:106 (5)

every man loves his wife.

On their bound readings, someone uttering the singular (4) and the quantiWed (5) is predicating the same thing of John and every man, that John satisWes (6), and that every man satisWes (6): (6)

x loves x’s wife

106 For discussion, see Heim and Kratzer (1998). Evans (1977, 1980) argues that the bound/free distinction in connection with (4) and (5) can be cashed out eVectively in terms of a distinction between using a pronoun in a referentially dependent way and using it to make independent reference, at least assuming a Fregean account of quantiWcation. For discussion, see Soames (1989) and Neale (forthcoming a).

Neale, This, That, and the Other

145

The crucial feature of bound readings is that they are possible only when the pronoun lies within the scope of (is c-commanded by) its antecedent.107 Thus only free readings of the pronouns are possible in (7)–(9), which is not to say they may not be coreferential with ‘John’ by virtue of being used freely to refer to John: (7) (8) (9)

/his wife loves John/ /the woman he married loves John/ /the woman who married him loves John/

I say that only free readings are available here for two reasons. First, notice that only free readings of the pronouns are available in the quantiWed counterparts of (7)–(9): (70 ) (80 ) (90 )

/his wife loves every man/ /the woman he married loves every man/ /the woman who married him loves every man/.

Now if we are serious about giving a uniWed account of what is going on in quantiWed and singular pairs like (4) and (5), we should be just as serious about giving a uniWed account of what is going on in the pairs (70 )/(7), (80 )/(8), and (90 )/(9). Second, there is an important syntactic distinction between, on the one hand, (4) and (5), and, on the other, (7)–(9) and (70 )–(90 ). In (4) and (5), the pronoun is within the scope of the purported antecedent, making a bound reading possible. Not so in (7)–(9) and (70 )–(90 ). But surely, it might be suggested, for each of (7)–(9) there is a reading upon which the pronoun and ‘John’ are co-referential. True, but irrelevant. We are discussing the asymmetric notion of binding not the symmetric notion of co-reference, remember.108 Free occurrences of /his/, /he/, and /him/ in (7)–(9) can still be used to make free reference to whoever /John/ is being used to refer to. But on such uses they are not bound by /John/: they are merely co-referential with it. And the diVerence between, on the one hand, the singular forms (7)–(9) and, on the other, their quantiWed counterparts (70 )–(90 ), is just that in utterances of the former the pronouns can be used freely to refer to whoever ‘John’ is being used to refer to, John now being as salient as any other potential target of an independent occurrence of a free pronoun. With utterances of the quantiWed examples, by contrast, there is no corresponding individual to target (i.e. no individual who is also the intended referent of the relevant DP, which is quantiWcational). With all of this in mind, let us now look at an interesting argument for a semantically distinct referential reading of deWnite descriptions, due in its essentials to Larson 107 Whilst this is close to universally accepted where the ‘antecedent’ is a quantiWed DP, some linguists and philosophers (including Evans) hold out where the antecedent is a referring expression. Along with Reinhart (1983) and others, I think this is a serious mistake. For discussion, see Neale (forthcoming a). 108 The asymmetry point is made clearly by Evans (1977, 1980), in terms of referential dependence vs coreference.

146

I. Incomplete Descriptions

and Segal (1995), although I shall state the argument diVerently. Compare, (700 )–(900 ) below with (7)–(9) and (70 )–(90 ) above: (700 ) (800 ) (900 )

/his wife loves the man I talked to/ /the woman he married loves the man I talked to/ /the woman who married him loves the man I talked to/.

Seemingly contrary to expectations, examples (700 )–(900 ) appear to pattern with the singular examples (7)–(9), not with the quantiWed examples (70 )–(90 ), in permitting anaphoric links between the direct object and the pronoun buried in the subject DP. Since binding is out of the question—witness (70 )–(90 )—how is this to be explained? Answer: the pronoun must be functioning as a referring expression that inherits its referent from the direct object, just as in (7)–(9) where the direct object is a referring expression. But in that case the descriptions in (70 )–(90 ) must also be functioning as referring expressions. This argument is essentially a syntactically elaborate version of the original Argument from Anaphora and appears to fall short of its target for the same reasons. Using a description referentially—no one doubts referential usage, only its semantic signiWcance—can make a particular individual salient (just as using a name can), rendering that individual a natural target for an occurrence of a free pronoun. The Russellian does not deny that the pronouns in (700 )–(900 ) may be used to refer to the individual the speaker is referring to using the descriptions; at the same time, he agrees with the syntactician that these pronouns are not bound by those descriptions. There is no contradiction here. Indeed, once I get to the promised rapprochement between the Unitarian and the Ambiguity Theorist, the analysis just given will seem more compelling. If the description is used non-referentially, as it might be in a particular use of, say, (10) (10)

his prime minister walks behind the monarch.

Plausibly it is D-type that we might be on the right track seems to be supported by the following contrast: (11) (12)

his wife loves a man I was just talking to his wife loves a man.

It is easier to use the richer ‘a man I was just talking to’ referentially here than it is to use ‘a man’ so.

12. Recent Claims The distinction between sentences (i.e. sentence types) and utterances (i.e. particular dated utterances of sentences) is so central to the philosophy of language that there can

Neale, This, That, and the Other

147

be only misery in store for those who miss it. One of the philosophers who did most to impress upon us the sentence–utterance distinction was Strawson (1950), in the same article in which he presented the alleged problem for Russell’s theory raised by incomplete descriptions. Sellars (1954) rightly took issue with Strawson on this matter (in the context of a broadside against Strawson’s (1950, 1952) critique of Russell) and was himself very much alive to Strawson’s sentence–utterance distinction, talking explicitly about elliptical utterances of ‘the table is c’ and never about an elliptical sentence ‘the table is c’. So it is striking that the sentence–utterance distinction is missed entirely by Strawson (1954) in his reply to Sellars and again by Stanley (2002a) in a recent discussion of the Sellars–Strawson debate containing some rather wild claims about Sellars and Strawson, and about more recent contributors to the descriptions debates. Amongst other things, Stanley (2002a) asserts: (1) that ‘an inXuential paper of Wilfred [sic] Sellars . . . has led some philosophers of language to the view that there is a legitimate notion of ellipsis’ [my italics] a ‘process’ [my italics] distinct from syntactic ellipsis which helps us understand the gap between sentence meaning and what is said although ‘no evidence has been given by Sellars for the existence of such a process [my italics]’ (2002a: 151). (2) that on ‘the proper use of the term ‘‘ellipsis’’ ’, ‘an utterance [my italics] is elliptical just in case its surrounding linguistic context quite literally supplies the missing words’, that ‘the terms ‘‘elliptical’’ and ‘‘incomplete’’ only seem apt when the phenomenon involves words that are supplied to incomplete sentences [my italics] by linguistic context’ (2002a: 156). (3) that ‘Sellars’discussion has engendered much confusion in subsequent literature on context-dependence’ (2002a: 157). (4) that Strawson ‘with characteristic incisiveness . . . in his reply to Sellars . . . in the very same journal issue . . . gives a clear explanation of the oddity of both Sellars claim and his vocabulary’ (2002a: 155). (5) that ‘all of Strawson’s points are correct’ (2002a: 156). (6) that some recent discussion ‘takes oV from an uncritical acceptance of Sellars’ vocabulary and distinctions’ (2002a: 156). (7) that ‘Sellars vocabulary and distinctions suVer from ambiguities, unclarities, and errors, which . . . continue to disrupt the literature on the topic today’ (2002a: 156). (8) that ‘every paper on ‘‘incomplete’’ deWnite descriptions that uncritically accepts Neale’s vocabulary of ‘‘explicit’’ and ‘‘implicit’’ . . . must be examined to see whether it is vitiated by Neale’s reliance on Sellars’discussion’ (2002a: 156). These claims build on three more by Stanley and Szabo´ (2002a), who, in connection with the matter of incompleteness, claim

148

I. Incomplete Descriptions

(9) to provide an ‘exhaustive’ and ‘comprehensive survey of the space of possible analyses’ (2000a: 219) (10) to bring to bear ‘considerations which militate against all but our own proposal’ (2000a: 219) (11) to ‘consider and reject . . . the ‘‘explicit’’ approach to quantiWer domain restriction [sic] discussed . . . by Stephen Neale’ (2000a: 219). All of this seems to be in the service of a claim made by Stanley (2000): (12) that ‘all eVects of extralinguistic context on the truth conditions of an assertion are traceable to elements in the actual syntactic structure of the sentence uttered’ (2000: 391). Strong stuV. Being of a pragmatist bent, I think the chances of (12) being true are, at best, slim. But that is not my concern here.109 I am concerned with (1)–(11). And here, Stanley’s substantive claims are easily dismissed once we examine what Sellars and Strawson actually say, and what those who ‘rely’ on Sellars and ‘uncritically accept’ talk of implicit and explicit approaches to incompleteness actually say. The ‘confusion’, the ‘errors’, and the ‘unclarities’ are Stanley’s, and it is he who ‘disrupts’ the literature. Once we slice through the rhetoric and confusions, it becomes clear that those who have ‘uncritically’ accepted informal talk of ‘explicit’ and ‘implicit’ approaches to incompleteness found in the literature have nothing to fear.

13. Incompleteness, Ellipsis, Modality The modality involved in describing utterance ellipsis is not diYcult to understand, and it is surprising that Stanley and Szabo´ profess to have ‘no clear conception of [its] intended interpretation’ (2000b: 295) and ‘as a result, no clear conception of the ‘‘explicit approach’’ ’ to incomplete descriptions (2000b: 295). The modal characterization of the explicit approach goes back at least to Sellars: ‘a given utterance of it is elliptical and states what would be nonelliptically stated, for example, by ‘‘The table over here is large’’ . . . ’ (1954: 200). Given that I quote this modal characterization (2000: 286), given that I take Sellars’s (1954) discussion as the locus classicus of the explicit approach in Descriptions, it is surprising that Stanley and Szabo´ (2000b), when explicitly discussing Sellars (1954) and Neale (1990, 2000), claim the modal characterization is original with me in (2000).110 109 For discussion, see Neale (forthcoming b). 110 The modal character is also explicit in the discussion by Davies (1981) that I explicitly drew upon in Descriptions. It is also explicit in the discussions of Bach (1987, 1994) and W. Blackburn (1988), which I did not discuss (but which Stanley and Szabo´ do discuss) and in Evans’s (1982) claim, in connection with one of his examples, that ‘he (the speaker) would be able to complete the description in a uniquely appropriate way’ (1990: 100). (Even Quine (1940) seems on the verge of a modal characterization—but of course the statement that

Neale, This, That, and the Other

149

Rather more worrying is Stanley and Szabo´’s (2000b) claim not to understand the modal characterization. What is so hard? Particular utterances of (1) (1)

the table is dirty

are understood as equivalent to utterances of longer sentences the speaker might have uttered in lieu of uttering (1): one utterance might be understood as equivalent to an utterance of (10 ), another an utterance of (100 ), yet another and utterance of (1000 ), and so on: (10 ) the table over here is dirty (100 ) the table we are about to sit at is dirty (1000 ) the table you promised to clean is dirty. There is nothing mysterious here.111 This way of talking reXects the way ordinary speakers—Stanley and Szabo included, as we shall see—describe the situation when asked. What is mysterious, however, is why Stanley and Szabo´ profess incomprehension. Is the modality harder to grasp than the analogous modality in Stanley’s (1997) discussion of rigidity? If one utters a sentence containing a rigid term . . . then one has asserted something diVerent from what one would have asserted [my italics] had one uttered, on that very same occasion, any sentence diVering from the original one only in the substitution of a non-rigid term for the rigid one. (1997: 135) Consider two terms (word-types) t and t0 . We shall say that t has the same content, with respect to a context c, as t0 , just in case, for any sentence S which contains t as a constituent, an utterance u of S in c would have [my italics] the same assertoric content as an utterance of a sentence which results from S by replacing t0 for t. (1997: 135)

Or harder to understand than the analogous modality in Stanley’s (2002c) statement of his ‘Expression-Communication Principle’? For all S, S0 , c, c0 , such that c and c0 agree on all contextual features relevant for determining what is said by S and S0 , S relative to c, S0 relative to c0 , express the same proposition if and only if an

there is an enriched description the speaker could have used involves quantifying into a modal context!) Critics of the explicit approach such as Reimer (1992, 1998a), SchiVer (1995), Ostertag (1998, 1999), Devitt and Sterelny (1999), and Devitt (1997a) also allude to its modal character. Devitt and Sterelny, who endorse the explicit approach in connection with incomplete descriptions used attributively but not in connection with those used referentially, are particularly explicit: ‘The way to save the Russellian is to see [incomplete descriptions] as elliptical: speakers have in mind a longer description which, if asked, they would produce to complete the brief description uttered’ (1999: 107). Stanley and Szabo´’s claim that the modal characterization is novel with me in 2000 is also straightforwardly contradicted by Stanley’s (2002a: 157 n. 11) quotation of Sellars. 111 And certainly the people mentioned in the previous footnote do not Wnd it mysterious. Nor does Devitt (this volume) who Wnds the explicit approach ‘generally hopeless’ in connection with incomplete referential uses. Whatever it is that Devitt Wnds ‘generally hopeless’, it is not because he Wnds its modal character mysterious. He has it down: ‘an incomplete deWnite is elliptical for a longer description that the speaker could supply’.

150

I. Incomplete Descriptions

utterance of S would communicate [my italics] the same thing as an utterance of S0 in every context c00 meeting the following three conditions . . . (2002c: 329).112

Or harder to understand than the analogous modality in remarks by Stanley and Szabo´ themselves. When setting up the incompleteness problem in connection with a particular utterance of (2) made by Lisa to Max, (2)

every bottle is empty,

Stanley and Szabo´ say, ‘we can plausibly assume that by uttering [(2)] Lisa conveyed to Max the proposition that every bottle she just bought is empty’ (2000a: 231). This is just the sort of thing the advocate of the explicit approach says, of course, and it is instructive that Stanley and Szabo´ avail themselves of an explicitly richer description in just stating the general problem. The modals are not far behind. ‘Had [my italics] Lisa been more explicit, she could [my italics] have conveyed the same proposition by uttering [(3)]] instead’ (ibid.): (3)

every bottle I just bought is empty.

And there are more to come. The diVerence between the various approaches they are going to discuss (their own included) they say, ‘lies in the way they spell out the relationship between Lisa’s actual [my italics] utterance of [(2)] and her hypothetical [my italics] utterance of [(3)]’ (2000a: 232). And according to ‘one very plausible’ version of one of these approaches (the ‘syntactic ellipsis approach’), ‘although the sentence articulated by Lisa is diVerent from the sentence she would have [my italics] articulated in the hypothetical [my italics] case, there is no diVerence between what was and what would [my italics] have been uttered’ (2000a: 232).113 I shall have more to say later about the explicit but innocuous modality later. Right now I want only to point out that the idea of syntactic theory recovering the underlying syntactic structure (or LF) of (1) from any of (10 ), (100 ), or (1000 ) involves a horrible category mistake, which is surely why no advocate of the explicit approach has ever advocated it. (And, of course, it would involve positing an inWnite number of sentences (not just alphabetical variants) with the same surface form, which would Xy in the face of both syntactic theory and common sense.) By contrast, there is no category mistake (or clash with common sense) in the following idea: upon hearing an utterance of (1), B might interpret A as expressing what A might have expressed more explicitly by uttering (10 ), (100 ), or (1000 ), or any of a number of similar sentences, B exercising his inferential, pragmatic abilities and exploiting facts about the context of utterance, beliefs about the speaker, and so on. Now of course the philosopher who recognizes utterance ellipsis might consider making the following (improbable) psychological claim: when A intends an utterance 112 I do not mean to be suggesting that there is any merit to the Expression-Communication Principle, only that there is no obvious mystery about its modal character. 113 There is no inconsistency in Stanley and Szabo´’s remark as technical uses of ‘articulated’ and ‘uttered’ are in play here.

Neale, This, That, and the Other

151

of (1) to be understood as expressing what he might have expressed more explicitly by, say, an utterance of (10 ), this is because A Wrst considered the longer sentence but decided to deliver a more economical one. But even if, as seems unlikely, a philosopher were to go along with this, his position would still have no bearing whatsoever on syntactic theory. He would be making no relevant claim about the syntax of the sentence actually uttered.

14. Strawson (1950) One of the virtues of Strawson’s ‘On Referring’ is its insistence on the position (also insisted upon by Grice) that referring and saying (or stating) are primarily things that speakers (rather than words) do.114 I am highly sympathetic to this stance. With Grice and Strawson, I see utterer’s meaning (a notion to be explicated in terms of utterer’s intentions) as conceptually prior to linguistic meaning, and I see speaker’s reference (a notion also to be explicated in terms of utterer’s intentions) as conceptually prior to the notion of linguistic reference (if such a notion is needed, which is far from clear). Whilst ‘On Referring’ has many virtues, it has some serious vices—as Paul Grice once put it, ‘a brilliant paper marred only by a discussion of Russell’s Theory of Descriptions’.115 Strawson’s failure to grasp a fundamental assumption underlying Russell’s proposals is the matter I want to focus on here, for it is this failure that leads Strawson (1954) to a quite ridiculous reading of Sellars (1954). The matter concerns the distinction between sentences and utterances of sentences, and the bearers of truth and falsity.116 Russell, according to Strawson (1950: 325), fails at crucial points to distinguish the following: (A1) a sentence, (A2) a use of a sentence, (A3) an utterance of a sentence. Similarly (B1)–(B3), in which ‘sentence’ is replaced by ‘description’. And this failure, he says, is one of the things that leads Russell into error. In order to forestall confusion and a silly defence of Strawson’s reply to Sellars, let us be quite clear what Strawson intends by (A1) and (A3). Sentences, in the sense of (A1), are sentence-types; this Strawson makes clear when he says that ‘one and the same 114 For detailed discussion, see the collections of papers by Grice (1989) and Strawson (1971), and Neale (forthcoming b). Every now and then a paper that explicitly discusses Grice’s notion of what is said and claims to endorse it gets even this basic point wrong, talking relentlessly about what is said by a sentence relative to a context. 115 The major vices in Strawson’s positive proposals are discussed in ch. 2 of Descriptions, where plenty of references to earlier discussions can also be found. 116 This particular vice in Strawson’s portrayal of Russell is discussed in ch. 2 of Descriptions. See also Evans (1982). It appears to have been overlooked by some recent commentators.

152

I. Incomplete Descriptions

sentence [may be] uttered on . . . various occasions’ (1950: 325). Strawson is also clear about utterances, in the sense of (A3): ‘two men who simultaneously uttered the sentence [‘‘the king of France is wise’’] in the reign of Louis XIV made two diVerent utterances of the same sentence’ (1950: 326).117 In summary, the same sentence (e.g. ‘the present king of France is bald’) may be used on one occasion to say something true, on another occasion to say something false. Thus sentences are not the sorts of things that are true (or false). If it is insisted that we need objects to be bearers of truth (falsity), rather than just saying that what the speaker said was true (false), we can take particular dated utterances, sentences-relativeto-occasions of utterance, the propositions expressed by particular dated utterances, or what have you. There are many interesting discussions of this matter in the literature; in order to provide some continuity with the rest of the present chapter and avoid entanglements with Strawson’s use of ‘use’, let us say that particular utterances are the sorts of things that are true or false. Strawson is quite right, of course, to distinguish sentences from utterances, and to distinguish between the linguistic meaning of a particular sentence and what someone says by uttering it on a given occasion, if only because of indexical expressions like ‘I’, ‘this’, ‘present’, ‘here’, and so on. But as Sellars (1954), Russell (1959), and Grice (1981) all point out in their commentaries on Strawson, the question of how to treat indexicals is, in the Wrst instance at least, quite distinct from the question of how to treat descriptions. Of course deWnite descriptions may contain indexical components (‘the present king of France’, ‘the man who gave me this’, etc.); but all this means is that there are descriptions to which the Theory of Descriptions and a theory of indexicality apply. No substantive issue turns on Russell’s failure to separate sentences from utterances explicitly when talking about descriptions. As Evans (1982) stresses, Russell is so very obviously concerned with the proposition expressed by a particular utterance— rather than the more abstract notion of the linguistic meaning of sentence-types—that 117 Notoriously, there is some latitude in the interpretation of (A2), but I suspect it is not harmful to Strawson’s overall position. The remark just quoted is the Wrst half a longer remark: ‘two men who simultaneously uttered the sentence [‘‘the king of France is wise’’] in the reign of Louis XIV made two diVerent utterances of the same sentence, though they made the same use of the sentence’ (1950: 326). He goes on to say of ‘two men who uttered the sentence, one in the reign of Louis XV and one in the reign of Louis XIV’ that ‘each made a diVerent use of the same sentence; whereas . . . two men who uttered the sentence simultaneously in the reign of Louis XIV, made the same use of the same sentence’ (1950: 325–6). Although there is enough here to illustrate to the charitable reader what Strawson has in mind by ‘use’ in (A2), taken in an uncharitably strict way it does leave an open parameter: is it just the fact that in both cases it is Louis XIV the men are talking about that guarantees sameness of use here? Or is the simultaneity of the utterances also playing a role? It should be clear that simultaneity per se is not really the issue, it is just something Strawson invokes so as not to get sidetracked into talking about ‘is’. That is, as far as sameness of use of the expression ‘the king of France’ is concerned, it is not simultaneity that is relevant but who it is being used to talk about, so two men who utter ‘the king of France is wise’ Wve minutes apart in the middle of the reign of Louis XIV, make the same use of ‘the king of France’ for Strawson, even if his view of ‘is’ (and tense more generally) requires him to say that they did not strictly make the same use of the same sentence because they referred to diVerent moments (I have no idea whether Strawson’s view of ‘is’ (and tense more generally) does require him to say this; but in any case the matter is not remotely relevant to anything I say in this chapter).

Neale, This, That, and the Other

153

it is very diYcult to lend any sort of sympathetic ear to Strawson on this point. If Russell were pushed into being more precise, he would say that only in the following sense is the sentence ‘The king of France is bald’ equivalent to the sentence ‘there is one and only one king of France and everything that is king of France is bald’: the proposition expressed by a particular dated utterance of the former is equivalent to the proposition that would be expressed, in the same context, by a particular dated utterance, of the latter.118

15. Sellars (1954) The distinction between a sentence and a particular dated utterance of that sentence, so rightly stressed by Strawson, is respected and taken very seriously by Sellars (1954) in his criticisms of Strawson. A passage quoted earlier contains much that is important and sets the scene nicely for an examination of more of the text. Concerning the sentence ‘the table is large,’ Sellars says, a given utterance [my italics] of it is elliptical and states what would be nonelliptically stated, for example, by ‘The table over here is large’ . . . the context functions to give the statement the force, for example, of ‘The table over here is large’. (1954: 200)

There are three matters we need to focus on in connection with this passage: (i) Like Quine before him, Sellars is not using ‘elliptical’ as a predicate of sentences. This facts jumps straight out in Sellars’s wording: he explicitly uses ‘elliptical’ as a predicate of utterances. If Smith and Jones know that there are several As, an utterance of the ‘the A’ made by Smith on Monday might be elliptical, whilst an utterance made by Jones on Friday might be non-elliptical because Smith and Jones both know (and know that the other knows) that every A but one was destroyed in a Wre on Wednesday. There moral, drawn earlier, is this: wherever we Wnd talk of ‘incomplete descriptions’ in the literature, if we are really interested in meaning and communication and not merely in playing with formal systems, we would do well to take as much care as Sellars by construing it as talk of incomplete utterances of descriptions. And, of course, when we discuss Sellars’s use of ‘ellipsis’ we must always bear in mind that he is explicitly talking of utterances being elliptical and not of sentences being elliptical. (Notice the examples of utterance ellipsis in Sellars’s own prose above.)

118 Strawson’s positive proposals about descriptions in ‘On Referring’ and later works have taken quite a drubbing in the literature, particularly at the hands of Sellars (1954). For a summary of some the problems and relevant references, see Descriptions, pp. 24–8 and 53–6 nn. 22–9. Strawson’s positive proposals are not my concern here, so I restrict my attention to his misreadings of Russell and Sellars.

154

I. Incomplete Descriptions

(ii) Sellars’s way of stating his main point is explicitly modal. Indeed, the modal character jumps straight out in the passage just quoted: the elliptical utterance states what would be nonelliptically stated in the same context by an utterance of the richer sentence (1)

the table over here is large.

(iii) Sellars suggests an elliptical utterance can be used to express what would be expressed, in the same circumstances, by a complete utterance of a sentence whose words include indexical (‘ego-centric’) words not present in the words of the incomplete utterance.119 Given (i)–(iii), it is abundantly clear that Sellars is not primarily interested in sentence ellipsis in his discussion of descriptions. And it is equally clear that he is not claiming that words or phrases present at some level of syntactic representation are deleted by some sort of transformational operation in the course of deriving surface syntax.120 Now there is a feature of Sellars’s discussion that might today mislead those unable to see issues in the philosophy of language (even those with a long history) except through the lens of generative linguistics. (One should often try to see the issues through the lens of generative linguistics, in my view; my complaint is with those who can see the issues only through that lens.) In preparing the ground, Sellars starts out with a very simple and obvious example of utterance ellipsis, one that is straightforwardly explicable in terms of sentence ellipsis. That is, he discusses an example that involves the utterance of something that might be thought to be less than a whole sentence: Jones: Seven is divisible by three. Smith: Seven is not divisible by three. Jones: Seven is.

And it is quite clear he might have made his point just as well using a simple case of VP ellipsis, in the sense discussed earlier, without any heavy stress on ‘is’: 119 This idea can also be found in Reichenbach (1947). Talking of the description ‘the train’ as it occurs in ‘the train will arrive at 7 P.M.’ Reichenbach says, ‘The necessary addition then is understood. It usually consists in a reference to a preceding utterance; for instance, it may be assumed in the form ‘‘the train of which we spoke’’ ’ (1947: 258). This meshes nicely with Reichenbach’s general picture of token-reXexivity. Reichenbach goes on to make some very instructive remarks about the relationship between descriptions and demonstratives. 120 The idea of levels of syntactic representation related by transformational rules did not hit mainstream linguistics until at least 1957 when Chomsky’s book Syntactic Structures was published, and it was not until the mid-1960s that Chomsky’s ideas about syntactic structure were felt strongly in philosophy departments. Chomsky’s long manuscript The Logical Structure of Linguistic Theory had been Xoating around since 1955 but was not published until much later, by which time Syntactic Structures (1957) and Aspects of the Theory of Syntax (1965), as well as a number of Chomsky’s research papers, were already quite well known to philosophers. The idea that Sellars had in mind syntactic deletion in 1954 in connection with the sentence ‘the table is large’ is preposterous and hopelessly anachronistic. Similarly, Quine (1940) and Reichenbach (1947).

Neale, This, That, and the Other (2)

155

seven is not divisible by three, but six is.

According to Sellars, what someone says by uttering ‘seven is’—or, as he tends to put it, what an utterance of ‘seven is’ says—is a function of context, but not in the same way that what someone says by uttering ‘this is red’ is: Correctly made utterances [my italics] of the latter sentence are complete even though they say what they do by virtue of their context. A context of a certain kind (e.g., the occurrence of a pointing gesture) is part and parcel of the grammar of the referring word ‘this’. On the other hand, . . . utterances [my italics] of ‘Seven is’ are as such not complete and are only made complete by the context in which they are uttered. That the context serves in this way to complete them is as much a matter of linguistic convention as is the role of ego-centric expressions. Let us call this type of ambiguity ellipsis and say that in ellipsis the context completes the utterance [my italics] and enables it to say something which it otherwise would not, diVerent contexts enabling it to say diVerent things. (1954: 200)

Now this may not be a particularly sophisticated way of making the point—Isn’t it more natural to talk about utterers rather than utterances saying things? Shouldn’t the roles of linguistic and non-linguistic context be distinguished?—but Sellars’s basic point is clear enough. Once Sellars has made the general point that some utterances are obviously elliptical, he moves on to utterances of sentences containing the description ‘the table’, consistently continuing to apply the word ‘elliptical’ to utterances: ‘A given utterance [my italics] of [‘‘the table is large’’] is elliptical and states what would be nonelliptically stated, for example, by ‘‘The table over here is large’’ ’ (1954: 176). Surely no one could construe Sellars as believing or claiming that an utterance of (3)

the table is large

constitutes an utterance of something that is not a whole sentence. (3) is just as much a well-formed and complete sentence as ‘the table over here is large’ or ‘the table beside me is large’. Again, we see why Sellars is so careful to distinguish sentences from utterances: he is concerned with elliptical utterances, in the sense he characterizes so clearly in modal terms. ‘Seven is’ may well be an elliptical sentence, but that has no bearing on the matter at hand: elliptical utterances of sentences containing deWnite descriptions like ‘the table’. (The example of utterances of the elliptical sentence ‘seven is’ is used only to get the ball rolling; it requires a monumental act of silliness to see Sellars as claiming that ‘the table is large’ is an elliptical sentence. Utterances of both ‘seven is’ and ‘the table is large’ are elliptical, the former as a matter of linguistic convention (because it is an elliptical sentence), the latter because of the way the world is, or is at least perceived to be by A.) Once the distinction between the speech act notion of an elliptical utterance and the syntactic notion of an elliptical sentence is clear (as it should have been all along), it is easy enough to distinguish the roles of speech act context and syntactic context in resolving utterance ellipsis (and in the case of the latter, also sentence ellipsis). In a case

156

I. Incomplete Descriptions

of an utterance of ‘seven is’, B is alerted to the utterance ellipsis by (at least) the sentence ellipsis. In the case of an utterance of ‘the table is large’, by contrast, there is no sentence ellipsis. B is tipped oV by an obvious non-linguistic fact: the world contains more than one table. And it is B’s knowledge of this fact—rather than a grammatical fact— that leads him to exploit context in the search for a complete interpretation of the utterance.

16. Strawson (1954) In a 1954 reply to Sellars’s paper, Strawson’s original misunderstanding of Russell is transposed into a misunderstanding of Sellars, leading to a single misguided criticism. Strawson broods on the obvious diVerence between the sentences (or at least expressions) ‘seven is’ and ‘the table is large’, then draws all the wrong conclusions about Sellars’s discussion because he misses the fact that Sellars is concerned with the elliptical and incomplete nature of certain utterances. Concerning ‘seven is’ or ‘James is’ (uttered in response to the question, ‘Who is going to drive?’), Strawson says ‘it seems very reasonable to call [these] sentences [my italics] . . . incomplete or elliptical’ (1954: 223). He then goes on to say, If one had to justify these phrases, I think one would be inclined to say that the sentences [my italics] were formally, linguistically deWcient, that they did not come up to a certain standard of how a nonconversational English sentence [my italics] should be composed; one would point out that in their conversational setting, the deWciency is remedied by the linguistic context, that the surrounding remarks supply the missing words. (1954: 223)

This seems basically correct as far as sentences are concerned, although one can well imagine cases in which a gesture or some salient feature of non-linguistic context remedies a linguistic deWciency.121 But, to repeat, Sellars is not talking about sentence ellipsis, and Strawson’s failure to appreciate this leads him into a sequence of errors: But Sellars next suggestion, I Wnd utterly puzzling. For he says that such a sentence [my italics] as ‘The table is large’ is incomplete or elliptical in the same sense as [‘seven is’]. (1954: 223) Now according to Sellars, if . . . I were to replace the words ‘This pond’ by the words ‘the pond’ [in ‘This pond is used by children for sailing boats on’] I should be replacing a complete and non-elliptical sentence [my italics] by an incomplete and elliptical sentence [my italics]! (1954: 223) In general, Sellars thinks that a sentence [my italics] containing a singular ‘the’-phrase can be rendered nonelliptical only by supplementing the ‘the’-phrase by some phrase containing what he calls an ego-centric expression. (1954: 223) 121 See the discussion of e.g. Nausicaa’s farewell (Odyssey 8. 461–2) in Hahnemann and Neale (forthcoming). One interesting feature of examples from poetry is that metrical as well as syntactic gaps need plugging.

Neale, This, That, and the Other

157

Strawson has missed the central point of Sellars’s discussion in spectacular style because he unremittingly overlooks his own distinction between sentences and utterances, and so his ‘objection’ to Sellars—which amounts to nothing more than the (correct) observation that the word ‘elliptical’ does not seem to apply to the sentence ‘the table is large’—is stillborn.

17. Repetition In a recent discussion of Strawson’s response to Sellars, Stanley (2002a) repeats Strawson’s simple error. ‘With characteristic incisiveness,’ says Stanley, ‘Peter Strawson in his reply to Sellars in the very same journal issue . . . gives a clear explanation of the oddity of both Sellars claim and his vocabulary’ (2002a: 155). Anyone familiar with the Sellars–Strawson exchange knows this can go nowhere. Stanley presents, as a single quotation, virtually two whole pages (pp. 222–4) of Strawson’s reply, the very pages that are replete with the error already discussed. Once he has Wnished quoting, Stanley says, ‘All of Strawson’s points are correct. First, the terms ‘‘elliptical’’ and ‘‘incomplete’’ only seem apt when the phenomenon involves words that are supplied to incomplete sentences [my italics] by linguistic context’ (2002a: 156). If Sellars were talking about sentence ellipsis, who could possibly disagree!122 Stanley goes on: Second . . . once one distinguishes between the kind of processes operative in resolving the reference of occurrences of demonstratives, and the kind of processes operative in resolving elliptical sentences [my italics] (such as in Sellars’ example ‘seven is’ and Strawson’s example ‘James is’) it seems bizarre to claim, as Sellars does, that the processes at work in resolving the contextdependence of ‘The table is large’ are the latter rather than the former. (2002a: 156)

Indeed it would be bizarre to claim it, and presumably this is why Sellars does not claim it. When Stanley examines Sellars’s text himself, rather than relying on the long quotation exemplifying Strawson’s ‘characteristic incisiveness’ and ‘clear explanation’, he fails to see that the passages he has carefully selected for quotation are precisely the ones in which Sellars makes it clear he is talking about utterance ellipsis: As Sellars writes, ‘ . . . each of these dialogues contains an utterance [my italics] of ‘‘Seven is’’; and it is clear that what is communicated by these utterances [my italics] is a function of the contexts in which they are uttered.’ (2002a: 154) 122 Actually, Stanley makes a mess of even this truism when analysing Strawson’s comments. He says: [Strawson] states that Sellars is justiWed in using the term ‘elliptical’ for utterances [my italics] of sentences like ‘seven is’ and ‘James is’, where the surrounding linguistic context ‘supplies the missing words’ to the incomplete sentences [my italics], because in the ordinary sense of the word ‘elliptical’, an utterance [my italics] is elliptical just in case its surrounding linguistic context quite literally supplies the missing words. This point is about vocabulary, the proper [my italics] use of the term ‘ellipsis’. (2002a: 156)

158

I. Incomplete Descriptions

According to Sellars, ‘Correctly made utterances [my italics] of [‘‘This is red’’] are complete . . . On the other hand, the two utterances [my italics] of ‘‘Seven is’’ are as such not complete.’ (2002a: 154)

Once the quotation marks are oV, Stanley wobbles all over the road: Sellars then proceeds to argue that sentences [my italics] containing ‘incomplete’ deWnite descriptions such as ‘The table is large’, suVer [sic] from . . . what he calls ‘ellipsis’. [Footnote omitted.] That is, he claims that a sentence [my italics] such as ‘The table is large’ is incomplete in just the same way as Jones’ utterances [my italics] of ‘Seven is.’ [Footnote omitted.] According to Sellars, an utterance [my italics] is elliptical for a sentence [my italics] like ‘The table over there is large’, in just the same way as Jones’ Wrst utterance [my italics] of ‘Seven is’ is elliptical for [the sentence? SN] ‘Seven is divisible by three’. Sellars claim here is startling, as is his use of ‘ellipsis’ to describe the phenomenon of incomplete descriptions. (2002a: 154–5)

The only thing that is ‘startling’ is Stanley’s slipping and sliding between ‘sentence’ and ‘utterance’.123 Stanley berates Sellars’s use of ‘elliptical’, failing—as he must, since he is blindly following Strawson—to realize that Sellars is talking about utterances not sentences. And even if Stanley were able to demonstrate conclusively that some other feature of Sellars’s overall philosophy renders Sellars’s use of ‘elliptical’ incoherent (rather than merely diVerent from the one he and Strawson mistakenly read into the paper), this would have no bearing whatsoever on the very clear and quite reasonable uses of the same word by Quine (1940), Kripke (1977), Brinton (1977), Salmon (1982), and Bach (1981), by me in Descriptions, by any of the people Stanley claims ‘uncritically accepts Neale’s vocabulary of ‘‘explicit’’ and ‘‘implicit’’ ’, or by anyone else who wishes to talk very naturally about elliptical speech.124 The motivation for Stanley’s (2002a) digression on Sellars and Strawson is, perhaps, not entirely transparent.125 He overtly seeks (i) to berate my mention of Sellars’s (1954) paper in Descriptions when describing one of the two traditional responses to the problem posed by incomplete descriptions (the ‘ellipsis’ or ‘explicit’ response), and (ii) to berate and silence those who have used or accepted my innocuous labels (‘explicit’ and ‘implicit’) for these two traditional responses. But a little digging reveals the digression is also meant (iii) to add weight to Stanley and Szabo´’s (2000b) response to criticism of their (2000a) spectacular misrepresentation of the explicit approach to incompleteness. Here is what Stanley says:

123 For further examples of this, see the passage from Stanley quoted in the previous footnote. 124 Perhaps Stanley thinks Sellars had no right (in 1954) to use the word ‘elliptical’ in the way he did because twenty years later it would be used in a narrower way in technical work by generative linguists—presumably such a complaint would apply equally to Quine’s use of ‘elliptical’ in 1940. By these dim lights, then by parity of reasoning Quine (1960) and Geach (1962) had no right to use the ordinary word ‘pronoun’ in the way they did because a decade or so later it would be used in a narrower way in technical work by generative linguists so as to exclude the reXexive pronouns ‘himself ’ and ‘herself ’? (‘Anaphors’ is the sense of Chomsky (1981).) 125 I am grateful to Robyn Carston for pointing this out to me.

Neale, This, That, and the Other

159

Sellars discussion has engendered much confusion in subsequent literature on context-dependence. Sellars’ paper has been given new life thanks to the inXuential discussion of ‘incomplete’ deWnite descriptions in Neale (1990, pp. 95V.), from which much work on the topic in the subsequent decade borrowed its vocabulary. Neale’s discussion takes oV from an uncritical acceptance of Sellars’ vocabulary and distinctions. But Sellars vocabulary and distinctions suVer from ambiguities, unclarities, and errors, which, because of Neale’s discussion, continue to disrupt the literature on the topic today. [Footnote 10: In particular, every paper on ‘incomplete’ deWnite descriptions that uncritically accepts Neale’s vocabulary of ‘explicit’ and ‘implicit’ approaches to context-dependence must be examined to see whether it is vitiated by Neale’s reliance on Sellars’discussion.] (2002a: 157)

It should be clear from all I have said already that the confusion engendered by Sellars’s discussion is all Stanley’s, the disruption to the literature entirely his doing, and the errors all his. But let us dig a little deeper as doing so will cast everything in a clearer light and provide some perspective on the explicit approach.

18. Misrepresentation Stanley and Szabo´ (2000a) make an interesting positive proposal in connection with incompleteness. It can be stated thus: the incompleteness associated with the utterance of a DP is to be explained on the assumption that it contains an aphonic, indexical, domain variable assigned a value ‘by context’. The overarching idea is a familiar one, discussed by Westersta˚hl (1985a), for example, but Stanley and Szabo´ add a novel twist: although the variable is syntactically real, it is not attached to, dominated by, or associated with either of the quantiWcational nodes, D (‘the’) or DP (‘the table’) in ‘the table’, as one might have thought; rather, it is associated with the nominal expression N (‘table’); as they put it, the variable ‘cohabits’ the node with ‘table’. We might call this a syntactic proposal with semantic import or a semantic proposal implemented syntactically, it doesn’t matter. What is key, however, is that it has a very clear syntactic dimension. DP

D

N

the

Strictly speaking, this approach to incompleteness is a version of neither the pure explicit nor the pure implicit approach, although it draws upon features of both.126 126 This is not a problem for the labels ‘implicit’ and ‘explicit’, which merely applied to two approaches to incompleteness I detected in the literature. The labels were intended to be neither exhaustive nor exclusive (I was

160

I. Incomplete Descriptions

From the implicit approach it borrows the idea of invoking quantiWer domain restrictions.127 What makes it diVerent from the implicit approach is that, instead of implicit, background domain restrictions, Stanley and Szabo´ want foreground domain restrictions, that is, they want them to be part of the proposition expressed. And on this score, their approach is in harmony with the explicit approach. Finally, it diVers from both approaches in having a crucial (and controversial) syntactic dimension: an N node in a phrase marker is not ‘inhabited’ by just a noun: it is ‘cohabited’ by a noun and an aphonic domain variable. In my opinion, this proposal faces serious syntactic and semantic obstacles, but that is a topic for another occasion.128 My purpose here is to scrutinize what Stanley and Szabo´ say about the explicit approach in the build-up to presenting their own proposal. The proposal appears very late in their paper, after a drawn-out assault on what they see as competitors to the proposal they are going to present. Indeed most of the paper appears to be an attempt to blast the logical terrain so hard that when their theory Wnally emerges it will face no competition. Two related fantasies infect their discussion. First they identify the explicit response with an absurd syntactic thesis. Second, they claim (i) to provide an ‘exhaustive’ and ‘comprehensive survey of the space of possible analyses’ (2000a: 219), (ii) to bring to bear ‘considerations which militate against all but our own proposal’ (2000a: 219), and (iii) to ‘consider and reject . . . the ‘‘explicit’’ approach to quantiWer domain restriction [sic] discussed . . . by Stephen Neale’ (2000a: 219). In short, they claim it is impossible for any theory except their own to succeed. The reason? Their theory, and theirs alone, can capture ‘implicit binding’. For many sentences X, they claim, the binding by an overt quantiWer in X of an aphonic domain variable in X is both necessary and suYcient to capture the relevant facts, and their cohabitation theory is uniquely positioned to do the job.129 It is a seriously incomplete discussion of what is possible that omits what is actual. And, as it turns out, (i)–(iii) are false because Stanley and Szabo´ never even ‘consider’ the explicit approach. What they actually ‘consider and reject’ is something quite diVerent, a straw man they call the syntactic ellipsis approach. First, a minor (but telling) point. Strictly speaking, Stanley and Szabo´ are wrong to talk about ‘the ‘‘explicit’’ approach to quantiWer domain restriction [my italics] discussed . . . by Stephen Neale’. What I call the ‘implicit’ approach to incompleteness is the one that invokes quantiWer domain restriction. What I called the ‘explicit’ approach was a response to the incompleteness problem that does not invoke quantiWer domain restrictions. Evidently, Stanley and Szabo´ are using ‘quantiWer domain restriction’ in a not trying to bifurcate the ‘logical space’, to provide an exhaustive or comprehensive survey of the space of possible analyses, as some claim to do in this area.) 127 This is why I suggested in ‘On Being Explicit’ (2000) that it was a version of the implicit approach. That was actually a very bad way to put the point, however, because Stanley and Szabo´’s approach diVers from the implicit approach in an important way I am about to specify. 128 See Neale (forthcoming, b) 129 Pragmatists have accepted implicit binding for some time, particularly in discussions of donkey anaphora. See e.g. Evans (1977), Davies (1981), Neale (1990, 1993), and Ludlow and Neale (1991).

Neale, This, That, and the Other

161

new way, as a label for the general phenomenon of incompleteness in connection with DPs. I won’t quibble with this. Now to the main point. Before they launch their attack on what they claim is the explicit approach, Stanley and Szabo´ festoon it with a mad thesis I am conWdent no advocate has ever advanced: on the explicit approach, according to Stanley and Szabo´, an utterance of a sentence containing an incomplete description (or other DP) is an utterance of a sentence that contains an ‘unarticulated portion’ (2000a: 232), a ‘covert expression that cannot be heard’ (2000a: 233), a ‘syntactic constituent that has no phonological manifestation’ (2000a: 233). This is truly what they say. And in a footnote appended to the last of these remarks (n. 17), they refer the reader to ‘the discussion of the ‘‘explicit’’ approach in Neale (1990), pp. 95V., as well as the articles cited in [Neale’s] note 49, p. 115’. My note attributes the explicit approach to Quine (1940), Grice (1981), and others, and the sentence to which the note is appended attributes it to Sellars (1954). The alleged commitment of the explicit approach to syntactic constituents that ‘cannot be heard’ amounts to a thesis in generative grammar, Stanley and Szabo´ claim, a thesis to the eVect that expressions present at some level of syntactic representation (LF presumably) are deleted by a syntactic operation in the mapping from to surface form (PF presumably). As they see things, the superWcial syntactic structure of Sellars’s example (1) is meant to be derived syntactically from the underlying (10 ): (1) (10 )

[S [DP the table] [VP is large]] [S [DP the table over there] [VP is large]].130

In their jargon, ‘over there’ is uttered but not pronounced. This is extraordinary. To claim that philosophers who have explicitly followed Quine, Sellars, and others in talking about elliptical utterances have been arguing for, or assuming, syntactic deletion is tantamount to claiming that these philosophers know just enough syntactic theory to know what syntactic deletions are, but not enough to know that deletions must be recoverable! Certainly no one would ever charge Stanley and Szabo´ with such a thing, for they are far from ignorant about the relevant aspects of syntactic theory, including the recoverability condition. As they say of VP deletion, it is ‘a syntactic phenomenon, due to some sort of syntactic rule of reconstruction or copying, or PF deletion under a syntactic parallelism condition’ (2000a: 226 n. 9). Yet they claim syntactic deletion is the heart of the position I was discussing when I sketched the traditional explicit approach in Descriptions, the approach associated with Quine (1940), Sellars (1954), and company. This is pure fantasy. I never discussed such a position in Descriptions—the relevant passages are reproduced below—and for good reason: Descriptions was meant to be a serious (if boring) book in the philosophy of language, not a work seeking to get mileage out of demolishing straw men. To the 130 Actually the LF will have a little more structure as the description will have been raised. For the present point, we can abstract from this.

162

I. Incomplete Descriptions

best of my knowledge, no one sympathetic to the explicit approach has ever claimed that when someone utters (1), he is uttering a sentence which contains an expression (‘over there’) that is uttered but not pronounced. Quite why Stanley and Szabo´ confuse (or embroider) the explicit approach with an extraordinary thesis about syntax is one of life’s mysteries. (i) Very clear warnings against precisely such a confusion (or embroidery) are voiced by Bach (1981, 1987, 1994)—whom Stanley and Szabo´ cite in various connections—and by Ostertag (1998, 1999). (ii) Not one of the well-known exponents of the explicit approach (see below) says anything to suggest it is a syntactic thesis or has a syntactic dimension. The fact that Stanley and Szabo´ talk of ‘the ‘‘explicit’’ approach . . . discussed, for example [my italics], by Stephen Neale’ (2000a: 219), shows they have one thing right: I am not the only person to have discussed it. In their eyes, my discussion is, I suppose, representative. Of course I was far from being the Wrst person to discuss it, all I did was pin a label on it: Quine (1940), Reichenbach (1947), Sellars (1954), Vendler (1967), Donnellan (1968), Kripke (1977), Brinton (1977), Sainsbury (1979), Davies (1981), Grice (1981), Wettstein (1981), Salmon (1982), Evans (1982), Soames (1986), W. Blackburn (1988), and no doubt many others, discussed it years before I threw my hat into the ring. Stanley and Szabo´ are making at least three mistakes when they attribute to the advocate of the explicit response a story about syntactic deletion: (1) they are overlooking the fact that the earliest advocates, Quine (1940), Reichenbach (1947), and Sellars (1954), proposed the explicit approach before ideas about levels of syntactic representation and transformational deletions had been formulated and absorbed; (2) they are ignoring the fact that an examination of the relevant literature reveals no advocate of the explicit approach—from Quine in 1940, to me in 1990—ever viewed it, or even discussed viewing it, as involving a process of syntactic deletion between levels of syntactic representation; and (3) they ignore the fact that anyone who knew about syntactic deletions or syntactic ellipses in generative grammar would ipso facto know that such deletions or ellipses must be recoverable. Let us now turn to textual matters and the history of the debate after Strawson and Sellars, to works written after the rise of generative grammar. (i) Donnellan (1968) says something that might be read as a tentative endorsement of the explicit approach in connection with at least some incomplete descriptions used attributively. Perhaps he knew about deletions in generative grammar at that time, but either way there is no indication that he is suggesting syntactic deletion when he says, in connection with these attributive cases, that ‘The reply one is inclined to make on Russell’s behalf is that in the loose way of everyday speech the context is relied upon to supply further qualiWcations on the description to make it unique’ (1968: 204 n. 5). Donnellan goes on to criticize the explicit response in connection with incomplete descriptions used referentially, but nowhere does he suggest he sees it as involving syntactic deletion.

Neale, This, That, and the Other

163

(ii) Kripke’s (1977) brief discussion of incomplete descriptions appears in a paper prepared from a transcript of a talk given at MIT in the early 1970s. Although sympathetic to a unitary Russellian account of descriptions, which he was defending primarily on methodological grounds from the sorts of arguments Donnellan had mounted, Kripke suggested that ultimately such an account might be compromised by referential uses of radically incomplete descriptions such as ‘the table’, which he saw the explicit approach as struggling to capture. His wording suggests very strongly that at the time of his MIT lecture, or at least at the time of editing the transcript, the type of utterance ellipsis that Quine and Sellars were discussing was an established part of the general background for discussions of Russell’s Theory of Descriptions: The considerations I have in mind have to do with the existence of ‘improper’ deWnite descriptions, such as ‘the table,’ where uniquely specifying conditions are not contained in the speciWcation itself. Contrary to the Russellian picture, I doubt that such descriptions can always be regarded as elliptical with some uniquely specifying conditions added. (Kripke, 1977: 8)

There appears to be no intention on Kripke’s part to view this sort of ellipsis as involving syntactic deletion of the sort discussed by MIT linguists. (iii) Brinton (1977) says something that suggests utterance ellipsis was at the time widely regarded as part of the Russellian picture, if not something Russell himself had in mind: Russell claims that many or most occurrences of deWnite descriptions must be regarded as elliptical for ‘complete’ descriptions or at least for descriptions which would be thought by their users to be ‘complete’. The description to be analyzed, then, is often not the description actually used by the speaker, but some longer description. Analysis takes us beyond the speaker’s actual utterance. (1977: 402)131

Brinton’s use of ‘actually’ and ‘actual’ give his summary a standard modal Xavour: the description to be analysed in an utterance is often ‘not the description actually used by the speaker’, he says, ‘but some longer description’; we go ‘beyond the speaker’s actual utterance’; and the clear implication here is of a longer description that could have been uttered.132 (iv) Sainsbury (1979), using more or less the same language as Donnellan, says that where we have an incomplete description, ‘we treat the context as supplying a further determination of the predicate involved in the description’ (1979: 115). There appears to be no commitment to syntactic deletion in this remark (and Sainsbury assures me he did not have it in mind). 131 I have not myself come across any passage in which Russell explicitly claims that utterances of sentences containing deWnite descriptions may be elliptical for complete utterances. But Brinton is quite likely correct that Russell would have concurred (and perhaps even did concur) with Quine and Sellars. 132 As noted earlier, when people who are sensitive to the distinction between utterance ellipsis and syntactic ellipsis—as Brinton is—use expressions like ‘elliptical description’ or ‘incomplete description’, it is obvious they are using a bit of shorthand and are talking about elliptical (or incomplete) utterances of the syntactically nonelliptical (complete) descriptions like ‘the table’.

164

I. Incomplete Descriptions

(v) Bach (1981) provides a very nice example of an incomplete description used attributively and invokes utterance ellipsis in his appraisal of the situation as well as a modal fact: If the emcee of a quiz show announces, ‘The winner gets a trip to Hawaii,’ he is using ‘the winner’ attributively, since he presumably does not know (and certainly does not intend the audience to think he knows) who the winner will be. Obviously he takes ‘the winner’ as elliptical for ‘the winner of the game about to be played here,’ and had he used that description, he would have used it attributively. Having not used it, he intended the audience not to Wgure out which winner he had in mind but how the description he used was to be completed. (1981: 224)

(Notice the unproblematic use of the modal ‘had he used that description’.) Later Bach goes on to say that, there are many sentences which are almost always used nonliterally as elliptical for other sentences. For example, ‘Ed doesn’t look tired, he is tired’ would likely be used with a suppressed ‘merely’ before ‘look’ to be inferred by the hearer, since the speaker would not be stating that Ed does not look tired but is tired anyway. Similarly, if I say, ‘I drink only Scotch,’ I would be stating not that I drink nothing but Scotch but merely that the only liquor I drink is Scotch . . . The phenomenon of elliptical speech is commonplace; indeed, it often seems stilted not to suppress words that can easily be inferred . . . Using incomplete deWnite descriptions elliptically. . . is just another case of this familiar phenomenon. (1981: 238)

There is much in these passages that I Wnd conducive; and it is clear that Bach’s talk of the ‘suppression’ of words is not meant to be understood as a claim about syntactic deletion between levels of syntactic representation.133 (vi) Davies (1981) makes explicit use of the modal ‘as though’, and there is no suggestion he is entertaining the idea of syntactic deletion: descriptions which do not, as they stand, apply uniquely to any object, (even taking into account contextual limitation of the domain of quantiWcation) are often used, and are interpreted as though [my italics] they contained further descriptive material (in the form of a restrictive relative clause). (1981: 162)

(vii) Grice (1981) discusses incomplete descriptions, in connection with negation and scope: Consider utterances of such a sentence as The book on the table is not open. As there are, obviously, many books on tables in the world, if we are to treat such a sentence as being of the form The F is not G and as being, on that account, ripe for Russellian expansion, we might do well to treat it as exemplifying the more speciWc form The F which is f is not G, where ‘f’ represents an epithet to be identiWed in a particular context of utterance (‘f’ being a sort of quasi-demonstrative). 133 Bach (1987) elaborates on these ideas a few years later and echoes Sellars’s (1954) point that someone using ‘the table’ is likely using it as ‘short for some complete description like ‘‘the table nearby’’, ‘‘the table previously mentioned’’, ‘‘the table I am pointing to’’, [footnote omitted] or something of the form ‘‘the table which is G’’ ’ (p. 105). Although Bach is here discussing a version of the explicit approach to incompleteness, I should stress that on his account completion does not take place at the level of what is said, but at the level of what is meant.

Neale, This, That, and the Other

165

Standardly, to identify the reference of ‘f’ for a particular utterance of The book on the table is (not) open, a hearer would proceed via the identiWcation of a particular book as being a good candidate for being the book meant, and would identify the reference of ‘f’ by Wnding in the candidate a feature, for example, that of being in this room, which could be used to yield a composite epithet (‘book on the table in this room’), which would in turn Wll the bill of being an epithet which the speaker had in mind as being uniquely satisWed by the book selected as candidate. If the hearer fails to Wnd a suitable reference for ‘f’ in relation to the selected candidate, then he would, normally, seek another candidate. So determining the reference of ‘f’ would standardly involve determining what feature the speaker might have in mind as being uniquely instantiated by an actual object, and this in turn would standardly involve satisfying oneself that some particular feature actually is uniquely satisWed by a particular actual object (e.g. a particular book). (1981: 276–7)

This is a rich passage shall draw upon later. For immediate purposes, however, it is enough to note that Grice nowhere talks of ‘the F ’ being obtained from ‘the F which is f’ by syntactic deletion. (vii) Evans (1982) seems to follow Donnellan in connection with incomplete descriptions used attributively, providing the following example to show how the explicit approach works: travelling in a car through the United States, I might pass through a town whose roads are particularly bumpy, and in consequence say ‘They ought to impeach the mayor.’ I do not intend my audience to identify the object spoken about as one of which he has information; I intend merely that he take me to be saying that the mayor of this town, through which we are passing, ought to be impeached, and this statement is adequately represented quantiWcationally. (1982: 324)

This is a nice, straightforward statement of the unadorned explicit approach. There is no mention here or anywhere else in Evans’s discussions of descriptions or in his lengthy discussions of syntactic theory, of deriving incomplete descriptions from complete descriptions by syntactic deletion. (viii) Salmon (1982) does not seem to have syntactic deletion in mind when he says that on the explicit approach an incomplete description is ‘elliptical for some more fully descriptive phrase to be supplied by presumed shared background assumptions in the context of use, or something similar’ (p. 39) ‘elliptical for some more fully descriptive phrases Xoating in reach just overhead’ (p. 44). (ix) Nor does Soames (1986) when he talks of the explicit (or ‘traditional’) approach involving ‘contextual supplementation’ and of the hearer needing to ‘extract content from context’ (1986: 278 and 301 n. 7). (x) Bach (1987) explicitly warns against viewing the elliptical/explicit approach as involving syntactic deletion (pp. 73–4), and provides the following elegant summary of the situation, before presenting his own theory: We much prefer not to make fully explicit what we mean when that can easily be inferred, and, indeed, speech is stilted when words that can easily be Wlled in are not suppressed.

166

I. Incomplete Descriptions

Using an incomplete deWnite description in place of a complete or a more elaborate one is just one case of this widespread phenomenon. So if a speaker were to utter [‘the table is covered with dust’], surely he would not be asserting that one and only one table in the universe is covered with dust. He is likely to be using ‘the table’ as short for some complete description like ‘the table nearby’, ‘the table previously mentioned’, ‘the table I am pointing to’, [ footnote omitted ] or something of the form ‘the table which is G ’. Whenever a description ‘the F ’ is used as short for a complete description ‘the Fc ’, I will refer to the latter as its completion, which the hearer is intended to identify as such. . . . if a speaker utters [‘the table is covered with dust’ he] is implicitly asserting what could be made explicit by [‘the table I am pointing to is covered with dust’]. (1987: 104–5)

It is clear from this passage and the pre-emptive warning that Bach does not think this description of the facts presupposes syntactic deletion or syntactic ellipsis. (xi) W. Blackburn (1988) discusses incomplete descriptions very fruitfully and with great clarity.134 He suggests that Russell’s (1959) response to Strawson (1950) might be defended in the face of an example involving an incomplete description such as ‘the table’, by saying that the speaker uses the incomplete description ‘in place of some other description (perhaps ‘‘the table over here’’)’ (1988: 268), and notes that ‘such a position has been defended by Wilfrid Sellars, among others’ (1988: 268). Blackburn then goes on to discuss Sellars’s account and calls it the ‘traditional’ account: the view that when a speaker uses a sentence containing an incomplete deWnite description, the sentence is (usually) used elliptically for some other sentence containing a complete description. [Note omitted.] When this occurs, what the speaker says is determined by the non-elliptical sentence, the sentence containing the complete description. . . . the position seems to provide an accurate account of what occurs in at least some cases. Consider, for example, the following exchange: John: Who is the prime minister of this country? Mary: The prime minister is Brian Mulroney The traditional account provides a plausible account of what is going on here; it is natural to suppose that Mary is using the incomplete description ‘the prime minister’ elliptically for ‘the prime minister of this country’. . . . an incomplete description is used elliptically in place of some complete description. (1988: 268)135

Blackburn nowhere suggests the traditional/elliptical/explicit account involves syntactic deletion or syntactic ellipsis. Indeed he is as clear as can be that on this account the

134 The material in Blackburn’s paper is based on material from his dissertation, ‘Reference and Descriptions’ (University of Toronto, 1986). Had I known about Blackburn’s dissertation and paper before I had written most of Descriptions—his paper was brought to my attention by Nathan Salmon after I had Wnished all but the footnotes—I would certainly have drawn on it. My conclusions would not have been diVerent, but I would have been able to save myself some work by citing Blackburn’s arguments and observations. 135 As noted earlier, when people who are sensitive to the distinction between utterance ellipsis and syntactic ellipsis—as Blackburn is—use expressions like ‘elliptical description’ or ‘incomplete description’, it is obvious they are using a bit of shorthand and are talking about elliptical (or incomplete) utterances of the syntactically non-elliptical (complete) descriptions like ‘the table’ or ‘the prime minister’.

Neale, This, That, and the Other

167

idea involves using an incomplete description in place of a complete one, which is of course the core of the explicit approach.136 We have reached 1990 and my own discussion (and labelling) of the explicit response in Descriptions. Did I say, or even imply, that I was construing those who had made the explicit response as endorsing, or even discussing, a theory that invoked a process of syntactic deletion? And did I suggest anywhere that I subscribed to such a view? Did I ever mention such a view? Most certainly not, on all counts. First, ex cathedra: ‘explicit’ is my label, and I hereby explicitly state that the explicit response does not and never did involve or invoke syntactic deletion or syntactic ellipsis. Second, the idea is ridiculous and in any event violates recoverability—no philosopher taking generative grammar as seriously as Bach, Davies, Evans, Grice, Sainsbury, Soames, and I did was ever going to proceed down that road. Third, as should be abundantly clear by now, there is nothing in the work of those people I explicitly mentioned as putting forward the explicit response—namely, Quine, Sellars, Donnellan, Sainsbury, Davies, and Evans—to indicate commitment to, or even contemplation of, the idea that it involved syntactic deletion. Fourth, everything of relevance I said about the matter in Descriptions is contained in the following passages, which, it will come as no shock to learn, make no reference to syntactic deletion: (1) On the explicit approach (taken by, e.g. Sellars (1954)), a particular utterance of ‘the table’ might be elliptical for ‘the table over there’. [Footnote: For similar suggestions see Quine (1940), Vendler (1967), Lewis (1973), Cresswell (1973), and Soames (1986).] (1990: 95) (2) On the explicit approach, the quantiWer ‘everybody’ (as it is used on this occasion) is elliptical for ‘everybody at the dinner party I had last night’, or some such ‘narrower’ quantiWer. (1990: 95)137

136 W. Blackburn sees certain types of examples as providing diYculties for the traditional ellipsis approach, diYculties that can be overcome by a modiWcation: where there are many competing completions, none of which can be said to determinately capture what the speaker intended, no single determinate proposition is expressed: Rather, there is some more or less vague class of propositions, each determined by some expansion of the description—‘The F that is H’, ‘The F that is K’, and so on. (Contrary to what Russell suggests, it is probably a mistake to think that these descriptions are in any sense present in the mind of the speaker. We can say instead that the speaker would be prepared to fall back on these descriptions if he were asked to identify which F he meant.) According to the modiWed version of the traditional account, when the speaker says ‘The F is G ’ he asserts the truth of the propositions in the resulting class (or, perhaps, of some weighted majority of these). (1988: 271) Stanley and Szabo´ (2000a) profess not to understand Blackburn here (see below). 137 The ‘occasion’ in question in (2) was an utterance of ‘everybody was sick’ made in response to a question about a dinner party I had last night. When I talked about syntactic ellipsis in Descriptions I was careful to talk of syntactic deletion rather than ellipsis in order to keep syntactic ellipsis and utterance ellipsis separate, e.g. in my brief (and useless) discussion of VP deletion on pp. 218–19. As noted several times already, when people who are sensitive to the distinction between utterance ellipsis and syntactic ellipsis use expressions like ‘elliptical description’, ‘elliptical quantiWer’, ‘incomplete description’, and ‘incomplete quantiWer’, obviously they are using a bit of shorthand and are talking about elliptical (incomplete) utterances of syntactically non-elliptical (complete) expressions like ‘the table’ or ‘everybody’.

168

I. Incomplete Descriptions

(3) Clearly I do not mean to be asserting that everyone in existence was sick, just that everyone at the dinner party I had last night was. In some fashion or other, this is discernible from the context of utterance. [Footnote: This point is made by Quine (1940), Sellars (1954), Sainsbury (1979), Davies (1981), and S. Blackburn (1984).] (1990: 95) (4) According to the explicit approach, incomplete quantiWers are elliptical for proper quantiWers. As Sellars puts it, the descriptive content is ‘completed’ by context. (1990: 95) (5) Let’s suppose that an incomplete description is elliptical for a proper (i.e., uniquelydenoting) description recoverable from the context of utterance. (1990: 96) (6) Evans suggests that he (the speaker) would be able to complete the description in a uniquely appropriate way, and supplies a plausible completion using referential rather than descriptive material. (1990: 100) (7) Even the descriptive content of an overt description is not always Wxed by purely linguistic factors. (1990: 201)138 (8) There is simply no diVerence between saying that there is an ‘implicit reference’ to the victim and saying that the incomplete description ‘the murderer’ is elliptical for a uniquely denoting description, such as ‘the murderer of that man’ (where ‘that man’ refers to the victim), or ‘the murderer of him’ (where ‘him’ refers to the victim), or ‘his murderer’ (where ‘his’ refers to the victim), all of which contribute the same thing to the proposition expressed, viz., the descriptive condition murderer-of-b where b is the victim himself rather than some description of b. Wettstein is just mistaken in claiming that ‘the murderer’ is ‘not elliptical for some Russellian description’ (ibid.); the proposition expressed by (3) [‘the murderer is insane’] is given by (4): (4) [the x: x murdered b] (x is insane). It is unclear why Wettstein thinks the description in (4) is non-Russellian. (1990: 99)

As Zvolensky (2000) points out, the last of these passages makes it quite clear I do not have a syntactic deletion account in mind. I was responding to Wettstein’s objection to the explicit approach that there are non-equivalent richer descriptions for which an attributive utterance of ‘the murderer’ is elliptical, from which no principled selection can be made. If I had intended a syntactic ellipsis theory, I could not have said what I said in (8) by way of responding to Wettstein. My point was that it was not the words in the richer description that mattered but what the words contributed to the proposition expressed, and that on that score there could be nothing to choose between, for example, ‘the murderer of him’ and ‘the murderer of that man’ for a direct reference theorist like Wettstein, ‘him’ and ‘that man’ being used solely to refer to the dead individual b, on direct reference accounts. My use of [the x: x murdered b] in the formalism giving the truth conditions of the utterance indicates clearly that I privileged no particular linguistic completion in the scenario in question, and that the formula itself was not an LF (LF was not introduced in any case until later in the book). Was I committing myself to direct reference here? No. I was committing myself to the 138 The word ‘even’ introduces the last of these remarks, and the word ‘overt’ was italicized because I was, in that passage, commenting Wrst on the fact the descriptive content of descriptive (i.e. D-type) pronouns is not always determined by purely linguistic factors.

Neale, This, That, and the Other

169

position that the direct reference theorist can get no mileage out of an attributive use of ‘the murderer’ in the scenario at hand. Am I prepared to commit myself to direct reference? No. I am happy to accept that in this case and many others no one richer description is better than many others the speaker and hearer could easily drum up. Why people worry about this baVles me. Perhaps Stanley and Szabo´’s real target is not me but more recent advocates of the explicit response, those drawn to it since the emergence of the label ‘explicit’ in 1990. If so, then they should say so, name names, and not attribute syntactic deletion to me and my explicit ancestors. Other critics of the explicit approach such as Bach (1987), Reimer (1992, 1998a), SchiVer (1995), Ostertag (1998, 1999), Devitt and Sterelny (1999), and Devitt (1997a, b; this volume) have not claimed to see it as involving syntactic deletion or syntactic ellipsis. Indeed, some of these critics—for example, Bach and Ostertag—have explicitly stated that syntactic deletion or syntactic ellipsis is not part of the story. Here is Ostertag (1998, 1999): the notion of ellipsis appealed to is not the one familiar from syntactic theory. . . whatever the relationship is that holds between [a sentence containing an incomplete description] and [a sentence containing a completed one] it is not syntactic. (1998: 20) There seems no reason to suppose that the relation between, say, [‘the table is covered with books’] and its completion is anything like that between: ‘John left and Bill did too’ and ‘John left and Bill left’. (1999: 144 n. 3)139

One can only speculate why Stanley and Szabo´ identify the friendless syntactic deletion approach with ‘the ‘‘explicit’’ approach . . . discussed, for example, by Stephen Neale’ (2000a: 219). One might, perhaps uncharitably, see them as desperate for a target, for shortly after their misrepresentation of the explicit response they provide arguments meant to refute it as part of a build-up to their own theory. But their arguments against the ‘explicit approach’ engage at most the antecedently implausible syntactic deletion theory—which they call ‘plausible’ (2000a: 232) before setting out their arguments against it.140

139 Students are warned that it would be a mistake to think of syntactic deletion as playing a role in a theory of incompleteness by Larson and Segal (1995) in their textbook Knowledge of Meaning. Taking care not to attribute such a view to anyone, Larson and Segal point out that such an approach would involve appealing to the idea of linguistic material that is ‘semantically signiWcant but phonetically unpronounced’ (1995: 331), a locution that appears to foreshadow Stanley and Szabo´’s talk of a ‘covert expression that cannot be heard’ (2000a: 233), and a ‘syntactic constituent that has no phonological manifestation’ (2000a: 233). 140 This is not the only place where Stanley and Szabo´ construct the Ximsiest of straw men. In one note (2000a: 232 n. 16) they consider what they call the ‘ambiguity approach’to incompleteness according to which a noun such as ‘bottle’ is ambiguous, an inWnite number of distinct lexical items ‘corresponding to it’, including ‘bottle206 ’ which means the same as ‘bottle1 Lisa just bought’ (where ‘bottle1’ is ‘what we usually Wnd as the Wrst entry in any dictionary, and means what we would ordinarily expect’ (2000a: 232 n. 16)). Stanley and Szabo´ do not cite anyone as holding this ridiculous view; nonetheless they feel compelled to mount an argument against it.

170

I. Incomplete Descriptions

It is unclear whether Stanley and Szabo´ (2000b) accept they earlier misrepresented the explicit approach, as there is much that is murky in their later discussion. They start out by saying, According to Neale, we misinterpret the intended meaning of this phrase [‘the explicit approach’] in the two places in which we use it. Of course nothing in our paper rests on this point. . . . We should emphasize that, even after studying Neale’s illuminating reply, we are unclear what he means by ‘the explicit approach.’ According to Neale’s original explanation, ‘incomplete quantiWers are elliptical for proper quantiWers’ (1990, p. 95). The current explanation diVers from the ones provided in (1990) by the addition of modal vocabulary; ‘the matrix of a quantiWed NP can be interpreted, in context, as elliptical for a richer (complete) matrix that could have been used in its place (Neale, 2000). The purpose of the new formulation is meant [sic] to bring out the modality Neale sees as implicit [sic] in Sellars’s use of the term ‘elliptical’. But we have no clear conception of the intended interpretation of these additional modal expressions, and, as a result, no clear conception of ‘the explicit approach’. (2000b: 295)

This is all glib. First, frequency of the phrase ‘explicit approach’ in their paper is not the issue for they Wrst identify the explicit approach with what they call the ‘syntactic ellipsis’ theory, then use their own label many times! (The label ‘explicit approach’ and my name also appear prominently in their short abstract.) Second, quite a lot turns on their misrepresentation for they claim in their original paper to provide an ‘exhaustive’ and ‘comprehensive survey of the space of possible analyses’ (2000a: 219) and to bring to bear ‘considerations which militate against all but our own proposal’ (2000a: 219). Third, it is simply false that I see modality as implicit in Sellars’s use of the term ‘elliptical’: I see it and always have seen it as explicit in Sellars’s explicit claim, that an elliptical utterance states ‘what would be nonelliptically stated’ in the same context by a complete utterance (my italics). Stanley and Szabo´ omit to mention that in Descriptions I explicitly took Sellars’s discussion as the locus classicus of the explicit approach; and as we have seen already, Sellars explicitly uses a modal characterization (as do many others). Fourth, given their own uses of analogous modals discussed earlier, how can Stanley and Szabo´ claim to have ‘no clear conception of the intended interpretation’ of such ‘additional’ modal expressions. (The ‘additional’ modal expressions they are referring to are just ‘can’ and ‘could’ in my statement that ‘the matrix of a quantiWed DP can be interpreted, in context, as elliptical for a richer (complete) matrix that could have been used in its place’ (2000: 286).) Fifth, after all of the passages from Descriptions quoted above, after all of the quotations from Quine, Sellars, Davies, and Evans above (all of which appear in Descriptions, and some of which are explicitly modal), and after the points I made explicitly in the short piece to which Stanley and Szabo´ (2000b) are replying, it takes an act of will to be ‘unclear’ about what I mean by ‘the explicit approach’.141 Critics such as Reimer (1992, 1997), Ostertag (1998, 1999), SchiVer 141 Stanley and Szabo´ deploy this rhetorical gambit elsewhere. In a note discussing Blackburn’s (1988) ‘modiWed traditional account’ of incomplete descriptions, they correctly present Blackburn as suggesting that an utterance of a sentence containing an incomplete quantiWer matrix expresses ‘the set of propositions expressible by sentences which result from [that sentence] by expanding the quantiWer phrase . . . into an expression that the

Neale, This, That, and the Other

171

(1995), Devitt (1997a, b, this volume), Devitt and Sterelny (1999), and others seem to have had no trouble whatsoever understanding it and generating some genuine problems.142

19. Go¨delian Completions Trivially, no logical problem of incompleteness arises for a Go¨delian description (ix)(Fx • x ¼ a), the sort used in spelling out Go¨del’s slingshot argument.143 This suggests a fool-proof way of regularly interpreting utterances of incomplete descriptions used referentially. Consider the following example due to SchiVer (1995). Ferdinand Pergola is known by you and me to uniquely satisfy the matrices of many distinct descriptions (‘the author of Smells and Tickles’, ‘the man staggering up to the podium in front of us’, etc.); seeing Pergola trip, I say to you ‘the guy is drunk’; you understand my remark. According to SchiVer, no single uniquely satisWed matrix, ‘however complex’, is salient enough to underlie both my intention and your comprehension. I suggest simple rather than complex: you understand me as saying something whose truth conditions are captured by (1), where a stands for Pergola: (1)

[thex : guy(x) • x ¼ a] x is drunk.

The underlying idea is simple and natural: completion of an incomplete description used referentially is eVected non-descriptively (as non-descriptively as possible, at any rate). It is something like this that I believe once had in mind in the passage cited earlier. Before saying more, I want at least to mention a matter I cannot do justice to here because of a host of complex psychological and semantic issues it raises. One of SchiVer’s main aims is to present a dilemma for a certain theory of propositional attitude reports—the so-called hidden indexical theory—and he goes about this by Wrst presenting a dilemma for anyone wishing to hold a unitary Russellian account of descriptions in conjunction with a direct reference theory of simple indexical and demonstrative pronouns.144 SchiVer asks us to suppose that I had uttered ‘he is drunk’ speaker ‘‘would be prepared to fall back on’’ ’ (2000a: 237 n. 23) and then add, ‘We are not sure what is meant by this latter phrase’ (ibid.). It is pretty obvious what Blackburn means, as SchiVer (1995) and other commentators recognize. An apparent problem for versions of the explicit approach like Blackburn’s is mentioned by Brinton (1977): in a particular situation a speaker will often assent to alternative completions, which lead to complete matrices that happen to be true of diVerent individuals. There is no evidence that this worry is implicated in Stanley and Szabo´’s professed uncertainty about what Blackburn means. 142 The rhetorical gambit again: Stanley and Szabo´ say, ‘We avoided using Neale’s phrase ‘‘the implicit approach’’, because we had no real grasp of it’ (2000b: 295). Again, other critics appear to have had no problem. For insightful discussions of both the implicit and explicit approaches, see Reimer (1992, 1998a). 143 See Neale (2001). 144 An expression is directly referential, in Kaplan’s (1989) sense iV its contribution to the proposition expressed is just the object to which it refers. There is no commitment in Descriptions to direct reference. Indeed I talked of object-dependent (rather than singular) propositions precisely to counter any thought that I was

172

I. Incomplete Descriptions

(rather than ‘the guy is drunk’) in the example involving Pergola. And he says the Russellian has no good basis for preferring a direct reference theory of my use of ‘he’ to a theory that treats it as an incomplete description with more or less the same content as my use of ‘the guy’. This is because in the two cases under consideration there would be no discernible diVerence in my communicative intentions, and these intentions are the only psychological states relevant to determining what I would have meant by my utterance—what I would say is part of what I would mean, so it would be backed by a communicative intention. The Russellian must, it would seem, either (a) deny that my demonstrative use of ‘he’ is directly referential (and oVer an alternative treatment, presumably description-based), or (b) deny the claim of indiscernible communicative intentions, or (c) deny that psychological facts alone determine the issue of what I meant. (a) might have some plausibility. My utterance of the demonstrative pronoun ‘he’ might be interpreted as equivalent to an utterance of a description we interpret as the Go¨delian [thex : male x • x ¼ a], where a refers to Pergola. Thus, a formal representation of the truth conditions of my utterance of ‘he is drunk’ might be given by (2): (2)

[thex : male x • x ¼ a] x is drunk.

I am inclined to think this would be a more plausible line of defence than (b)—(c) is out of the question for the pragmatist—but more would need to be said about the nature of pronouns and about the relationship between formal representations of truth conditions, the LFs of English sentences, and the thoughts we seek to convey. The upshot of all this is a welcome rapprochement between those who have argued for a semantically signiWcant distinction between Russellian and referential uses of descriptions and those who have argued for a unitary Russellian theory. In a sense, everyone was right and everyone was wrong. When a description is used referentially it is being used in such a way that there is an obvious Go¨delian completion (whether the description actually needs one or not). On this view, if I use a description ‘the f’ referentially, to refer to John, then what I say entails that John is f, which seems right. If, in SchiVer’s example, I had said not ‘the guy is drunk’ but ‘the idiot is drunk’, then surely I am committing myself to Pergola’s being an idiot, and any purely referential alternative is going to have to Wnd some way of accommodating this. The diVerence between what I called referentialD and referentialN uses in chapter 3 of Descriptions can now be thought through in terms of the type of expression most plausibly substituting for a if (2) were to be rendered explicitly in English, a simple demonstrative or a name. Do we now say that it is part of the meaning or semantics of ‘the’ that on one use it invites a Go¨delian completion? Or should we see such completions as simply an abstraction over regular interpretations? I doubt it really matters how we think about this until we have some understanding of the cognitive assuming direct reference. There is no problem of principle in maintaining that a formal language (or Mentalese, for that matter) contains directly referential terms while English itself does not. On my account, of course, as far as the use of natural language is concerned it is speakers rather than expressions that refer. On the use of ‘directly referential’ and ‘rigid’ in connection with what a speaker refers to and says, see Neale (forthcoming b).

Neale, This, That, and the Other

173

mechanisms involved in interpretation, which is presumably a long way oV. Either way, the philosophical point remains intact: descriptions are Russellian, and the phenomenon of referential usage is a special case of the phenomenon of incompleteness. I said ‘a special case’ not ‘just a case’; the case is special in the sense that it is highly regular, perhaps even conventional. Which brings me to the next point.

20. The Argument from Convention Devitt (1997a, b, this volume) and Reimer (1998a) have presented an intuitive and powerful argument for an ambiguity in deWnite descriptions. I shall call it the Argument from Convention: referential uses of descriptions are common, standard, regular, systematic, and cross-linguistic; indeed so much so that it would be a bit rich to deny that such uses are conventional, a direct function of linguistic meaning in a way that referential uses of other quantiWed DPs are not.145 I think Devitt and Reimer are right about this; but rather than undermining the unitary Russellian analysis per se, the Argument from Convention undermines only the standard, wooden, Gricean explanation of referential usage (like the one I sketched in Descriptions), which amounts to no more than a generalized conversational implicature story.146 But the proposal on the table here is unruZed by the Argument from Convention. Indeed, the proposal seems to explain the convention in question, and this is why it seems to me a synthesis of everything that is right and nothing that is wrong in the old unitary Russellian theory and in semantical ambiguity theories that associate object-dependent (rather than object-independent) propositions with what is said when a description is used referentially. Moreover, it seems to do a much better job than either in yielding blueprints that place the right truth-conditional constraints on what is said, particularly where modalities are concerned.147 In this last respect it also diVers from the earlier pragmatist 145 I do not want to put too much weight on the word ‘convention’ here. Perhaps ‘Argument from Regularity’ is a better label, but I have been using ‘Argument from Convention’ for some years (against Kent Bach’s advice) so I stick with it here. For an illuminating discussion of the pitfalls involving ‘conventional’, ‘standard’, ‘regular’, and so on see Bach (1995). 146 All conversational implicatures must be calculable in Grice’s sense, even those he takes to be generalized. It is crystal clear from the Wrst sentence of ‘Logic and Conversation’ onwards that the conversational implicatures associated with uses of the words ‘and’, ‘or’, ‘if ’, ‘every’, ‘a’, ‘the’, etc.—i.e. those words corresponding to formal devices in logical theory—are generalized implicatures for Grice, the ones of philosophical importance, the ones that really bothered him (unlike the particularized ones which have no philosophical signiWcance). It is equally clear that ‘generalized’ has no theoretical import for Grice in the context of his account of the properties an implicature must have if it is to count as conversational (hence the calculability requirement) and that generalized conversational implicatures are quite diVerent from conventional implicatures. Grice is so very careful here, and it is odd that he has been misunderstood. (by e.g. Devitt (this volume)). 147 One advantage it has over the original unitary Russellian account is that it captures a reading of (i), pointed out to me some years ago by Irene Heim: (i) ‘It is always the case that the player on the left wins these games’. Traditional considerations governing scope suggest that the description may not take scope over the temporal operator, yet there would still appear to be a reading of (i) upon which the speaker could be saying of Smith (who is the player standing on the left) that he always wins.

174

I. Incomplete Descriptions

proposals of Recanati and Bezuidenhout which seek broad truth-conditional agreement with the ambiguity theorist; furthermore, unlike its pragmatist predecessors, it does not require us to view the type of proposition (object-dependent or objectindependent) expressed as semantically underdetermined, an idea that has always seemed to me to carry the whiV of ambiguity. (Given the traditional relation between proposition types and logical form, the attempt to use LFs in capturing logical forms, and minimal constraints on the compositional character of LFs, it would seem that the proposals of Recanati and Bezuidenhout will not be able to avoid positing two distinct LFs.)

21. Demonstratives as Go¨delian IndeWnites Some who have posited a semantically distinct, referential reading of ‘the f’, have been inclined to posit one for ‘a f’ too—but others have pointedly not done this. Let us suppose the following truth-theoretic axiom helps yield a blueprint for interpreting an utterance of a sentence of the form ‘a f is c’: (1)

[ak : f]c is true of a sequence s iV c is true of at least one sequence that f is true of that diVers from s at most in the k th position.

As noted earlier, the simplest cases exemplifying the problem of incompleteness arise naturally in connection with uses of " 1 determiners such as ‘the’, ‘every’, and ‘no’, whilst the most straightforward ‘derived’ or ‘inverted’ cases arise naturally in connection with uses of "1 determiners such as ‘a’ and ‘some’. The need for enrichment in connection with indeWnite descriptions is clear when they occur within the scope of negation: (2)

It is not the case that a student lives in Chicago.

If we are discussing students in my seminar, I can say something true by uttering (2), despite the obvious fact that many students live in Chicago. It is sometimes suggested that demonstrative descriptions ‘that f’ and ‘this f’ are special forms of deWnite descriptions. But the persistence test, discussed earlier, suggests that when used demonstratively they are more like indeWnite descriptions. But what does the diVerence between demonstrative and indeWnite descriptions consist in? The answer may well be that, whereas indeWnites do not signal uniqueness, demonstratives do. But doesn’t that make them like deWnites? Yes and no. The diVerence is in the way uniqueness is meant to be secured. Demonstratives are Go¨delian by nature. An act of reference is signalled as a matter of linguistic convention by the use of ‘that’ (rather than ‘a’ or ‘the’). In short, they are Go¨delian indeWnites; the blueprint for an utterance of ‘that f is c’ is basically one speciWed by (3): (3)

[anx : f(x) • x ¼ that]c(x):

Neale, This, That, and the Other

175

This would appear to deliver, by a route Lepore and Ludwig (2000) may Wnd objectionable, an analysis equivalent to one they have suggested, capturing the demonstrative and descriptive features (or conventions governing the use of ) demonstrative descriptions. The following truth-theoretic axiom would help yield the desired blueprint: (4)

[thatk : f]c is true of s iV c is true of at least one sequence that ‘that ¼ x • f’ is true of that diVers from s at most in the k th position.

There is no reason why f itself might not be another identity statement, which might suggest a treatment of ‘that name’ as equivalent to [anx : that ¼ x • name ¼ x]c(x), assuming an appropriate axiom for name, consistent with a speaker-based notion of reference, of course. There is much to discuss here, but it should be clear that, given this notion of reference, under the Go¨delian analysis the usual problems for accounts of demonstrative descriptions created by the twin pulls of rigidity and descriptive content of what is said evaporate.148

22. Relativization Again It was noted earlier that descriptions may contain pronouns bound by exterior quantiWers: (1) [every man here]1 loves the woman he1 married (2) [every man who bought only one donkey]1 paid cash for the donkey he1 bought.

148 Notice that (i) the third person personal pronouns have possessive forms (‘his’, ‘hers’, and ‘its’), (ii) demonstrative descriptions have possessive forms (‘this man’s’ and ‘that man’s’), but (iii) the demonstrative pronouns ‘this’ and ‘that’ do not have possessive forms (‘this’s’ and ‘that’s’): ‘this’s colour’ and ‘that’s shape’ are usually regarded as ill-formed and must give way to the ‘French’ forms, ‘the colour of this’ and ‘the shape of that’, respectively. Is the following the beginning of an explanation? (a) There is no overt nominal for ‘s’ to attach to in the DP ‘that’ because its syntactic structure is really [DP that [NP e]], where the Wrst element is actually the determiner ‘that’ and the second, [NP e], an aphonic nominal (b) ‘s’ cannot attach (linearly speaking) to the aphonic [NP e]; (c) it cannot attach to the determiner ‘that’ either because [NP e] is (although not present at PF) in the way; (d ) since ‘that’ is not assigned case by a verb, or preposition or other marker (such as ‘s’), Chomsky’s (1981) case Wlter is violated. (Mutatis mutandis for ‘these’ and ‘those’.) Pursing this line of thought might, as Richard Larson has pointed out to me, help to explain why we Wnd a similar situation with respect to other determiners, that freely license a ‘missing’ NP. Sentence (i) is Wne, but attempts to add the possessive fail, as in (ii), unless an overt nominal is present, as in (iii): (i) [DP many [NP e]] applied, but [DP few [NP e]] were selected (ii) [DP few[NP e]]’s credentials were good enough (iii) [DP few[NP applicants]]’ credentials were good enough. Unfortunately, this does not explain why possessive forms of some other demonstrative and indexicals are usually regarded as ill-formed. For example, although ‘yesterday’s’, ‘today’s’, and ‘tomorrow’s’ are Wne as possessive forms, ‘then’s’ and ‘now’s’ are not (although there was once a band called ‘Now’s Children’). Nor are ‘ here’s’ or ‘there’s’.

176

I. Incomplete Descriptions

The subject DPs bind-into the descriptions by virtue of wholly-binding the pronoun ‘he’. The descriptions are Russellian but relativized in the sense that uniqueness is relative to choice of man satisfying the NP in the subject DP.149 It was also noted that even utterances of incomplete descriptions and pronouns may be understood as elliptical for bound-into descriptions: (3) (4)

[every man who bought only one donkey] paid cash for the donkey. [every man who bought only one donkey] paid cash for it.

But here, of course, there is no actual variable-binding because there is no actual variable in the sentences’ LFs, at least not according to the pragmatist proposal I suggested in Descriptions and have reiterated here. And that’s why I prefer to talk of ‘relativization’, as I did in Descriptions, when talking about the interpretive phenomenon rather than ‘implicit binding’, which has been known to lead people astray. All of (1)–(4) exemplify relativization; but on the suggestion at hand only (1) and (2) make that relativization transparent through the actual binding of variables in syntactic structure. It is the task of a pragmatic theory to explain how hearers retrieve the content speakers intend (or at least something close enough to it). One of the standard uses of the English expression ‘in question’ when it is attached to a noun-phrase seems to be explicitly to invite the hearer to see the description as incomplete; indeed, it can be used to indicate relativization:150 (5) (6)

[every man who bought only one donkey] paid cash for the donkey in question. every man kissed the woman on his immediate left. At least one man married the woman in question.151

The demonstrative descriptions ‘that donkey’ and ‘that woman’ may also be used in (5) and (6), but oddity would accompany ‘that donkey in question’ and ‘that woman in question’. Why is this? And why is it that ‘a woman in question’ is odd but ‘one of the women in question’ and ‘every woman in question’ are Wne (the intrusion has to be paragogic rather than epenthetic)?

23. The Argument from Binding A number of philosophers and linguists have argued that some occurrences of deWnite descriptions function as bound variables and hence as referential expressions, which if 149 There is nothing conceptually problematic about an expression that is bound-into wholly-binding (or binding-into) another: (i) [every man here]1 thinks [the woman he1 married]2 loves [her2 mother]. 150 This is noted by Kripke in his 1973 John Locke lectures, unfortunately still unpublished. 151 Related devices are ‘under consideration’ and ‘at hand’. Unlike ‘in question’ these seem less happy in environments where relativized interpretations are called for.

Neale, This, That, and the Other

177

true would create a problem for a unitary Russellian analysis.152 The arguments are far from compelling. First, in all of the cases I have seen discussed in the literature, the Russellian analysis yields the right result once incompleteness is taken into account and the distinction between wholly-binding and binding-into is appreciated; second, in certain cases the bound variable analysis yields the wrong results while the Russellian analysis continues to deliver the right results; and third, the Russellian might actually have the germ of an explanation of why deWnite and demonstrative descriptions seem able to function in this way whilst indeWnite descriptions do not. Let me begin with the following claim by G. Wilson: deWnite descriptions have a number of related pronominal uses, and referential use [of descriptions] is the use of these pronominal descriptions in direct, singular reference. Pronominal uses . . . . contrast systematically and as a matter of semantics with instances of attributive use. (1991: 359)

Wilson’s case against the unitary Russellian position (and hence the motivation for his own theory of ‘pronominal descriptions’) turns on the unargued assumption that certain occurrences of descriptions must be treated as bound pronouns, an assumption that is simply false. Consider the following examples, on which the case rests: (1) (2)

[every scientist who was Wred from the observatory at SoWa]1 was consoled by [someone who knew [ the Wred scientist]1 as a youth]. [every scientist who was Wred from the observatory at SoWa]1 was consoled by someone who knew [him]1 as a youth.

It is Wilson’s opinion that the italicised description [in (1)], like the pronoun [‘him’ in (2)] that could replace it, is a variable bound by the quantiWer phrase that fronts the sentence. The repeated lexical material in the descriptor helps here to indicate the tie between the description and the quantiWer phrase that binds it. (1991: 360–1)

On this account, the truth conditions of what is said by utterances of (1) and (2) are given by (3), the underlined variable x inside the second quantiWer doing the work of the italicized pronoun in (2) and the italicized description in (1): (3)

[everyx ; scientist x • x was Wred from the observatory at SoWa] [somey : y knew x as a youth] (x was consoled by y).153

152 See (e.g.) Kempson (1986), Wilson (1991), Larson and Segal (1995). 153 Remember these representations are not LFs. Stan Dubinsky has suggested that the descriptions Wilson sees functioning like bound pronouns fail to satisfy a condition that bound pronouns should satisfy. Compare (i) and (ii): (i) [every scientist who was Wred from the observatory at SoWa]1 was consoled by the project leader he1 thought had made the decision to terminate him1 (ii) ? [ every scientist who was Wred from the observatory at SoWa]1 was consoled by the project leader he1 thought had made the decision to terminate [the Wred scientist]1 . (i) is a perfectly natural sentence in which ‘he1 ’ and ‘him1 ’ are interpreted straightforwardly as bound by the subject DP. By contrast (ii), which diVers from (i) only in the replacement of the pronoun ‘him1 ’ by the descrip-

178

I. Incomplete Descriptions

There are several things to note here.154 (i) For some speakers, there is an important diVerence between (1) and (2): what someone says by uttering the latter, but not by uttering the former, can entail that every scientist Wred from the observatory at SoWa was male.155 In order to forestall immediate objections to Wilson’s proposal, let us think of ‘him’ in (2) as shorthand for ‘him or her’ (or simply replace it by ‘him or her’); I will, however, return brieXy to this matter when I compare Wilson’s proposal with the Russellian’s. (ii) It is well known that demonstrative descriptions (phrases of the form ‘that f’) can be used to signify an anaphoric link in much the same way as deWnite descriptions. In (1), for example, ‘the scientist’ could just as well have been ‘that scientist’. (iii) A wooden Russellian treatment of the description ‘the Wred scientist’ in an utterance of (1) would yield an interpretation we can render as (4), which obviously fails to capture the intended interpretation of the utterance: (4)

[everyx : scientist x • x was Wred from the observatory at SoWa] [thez : Wred scientist z] [somey : y knew z as a youth] (x was consoled by y).

The reason (4) fails is obvious: it fails to relativize values of z to values of x in the way Wilson’s bound variable treatment (in eVect) does by treating ‘the Wred scientist’ as an occurrence of x. (iv) On a more subtle Russellian treatment, ‘the Wred scientist’ as it occurs in an utterance of (1) is an incomplete description that is meant to be interpreted as if it were an utterance of richer description that is bound-into. Saul Kripke seems to have spotted all of this some time ago.156 A natural enrichment manifests itself in ordinary tion ‘the Wred scientist1 ’, does seem a little unnatural. The situation is similar with (iii) and (iv), which diVer from (i) and (ii) only in the replacement of the nominative pronoun ‘he1 ’ by the description ‘the Wred scientist1 ’, and which thereby diVer from one another only in the replacement of the pronoun ‘him1 ’ in (iii) by the description ‘the Wred scientist1 ’ in (iv): (iii) [every scientist who was Wred from the observatory at SoWa]1 was consoled by the project leader the Wred scientist1 thought had made the decision to terminate him1 (iv)? [every scientist who was Wred from the observatory at SoWa]1 was consoled by [the project leader the Wred scientist1 thought had made the decision to terminate the Wred scientist1 ]. All of this suggests to Dubinsky that the occurrence of ‘the Wred scientist’ in (ii) and the second occurrence of it in (iv) do not function as variables bound by ‘every scientist who was Wred from the observatory at SoWa’ since they do not behave in exactly the same way as the occurrences of the pronoun ‘him’ in (i) and (iii), which do seem to function as bound variables. If Wilson maintains the standard line that variables are directly referential, then Dubinsky’s examples, if convincing, present a genuine problem. I am grateful to Anne Bezuidenhout for forwarding Dubinsky’s observations and questions, and for her own thoughts on the matter. 154 An alternative diagnosis is oVered by Simons (1996). 155 There is some Xuidity here. Fifty years ago, I imagine fewer people ‘felt’ the entailment (cp. ‘someone has left his pen behind’ which would have been regarded as default-neutral as to gender); and, of course, Wfty years ago there were fewer female scientists in the world, numerically and proportionately. 156 Examples similar to Wilson’s are discussed by Kripke in his 1973 John Locke lectures, and the schematic form of what I take to be the right Russellian response can be extracted from his discussion. Kripke and Scott Soames brought up the same idea in the discussion following Wilson’s presentation of similar examples at a conference on anaphora at Princeton in Oct. 1990.

Neale, This, That, and the Other

179

talk where we might Wnd the overtly relativized description ‘the Wred scientist in question’. In a representation of the truth conditions of an utterance of (1), a Go¨delian description containing variables on both sides of the identity sign gives us exactly what we want: (10 )

[everyx : scientist x • x was Wred from the observatory at SoWa] [thez : Wred scientist z • z ¼ x] [somey : y knew z as a youth] (x was consoled by y).

The matrix of [thez : Wred scientist z • z ¼ x] is understood as uniquely satisWed relative to values of x, exactly as in the donkey and other relativized examples discussed earlier. In short, the Russellian says that the incomplete description in (1) is not, pace Wilson, a bound variable, but just another incomplete description—one for which the speaker could provide a fuller description that is bound-into—a description containing a bound pronoun. It is an incomplete, relativized description whose natural completion contains an expression understood as a variable bound by the subject expression. (v) The Russellian analysis is perfectly capable of making transparent the relationship between anaphora on singular and quantiWed expressions, something Wilson rightly sees as repudiating a ‘‘pronoun of laziness’’ approach to pronouns anaphoric on singular terms. For example, an utterance of (5) will be analysed as (50 ), and an utterance of (6) as (60 ) (imagine (6) uttered in a context in which it is presupposed that all of the Wred scientists were gifted astronomers):157 (5) (50 ) (6) (60 )

Hugo Wexler was consoled by someone who knew the gifted astronomer as a youth. [Hugo Wexler x] [thez : gifted astronomer z • z ¼ x] [somey : y knew z as a youth] (x was consoled by y) every scientist who was Wred from the observatory at SoWa was consoled by someone who knew the gifted astronomer as a youth. [everyx : scientist x • x was Wred from the observatory at SoWa] [thez : gifted astronomer z • z ¼ x] [somey : y knew z as a youth] (x was consoled by y).

These straightforward analyses neatly capture the fact—which Wilson wants to capture—that the interpretation of utterances of (5) and (6) concerns the satisfaction of the same condition. According to the Russellian, the condition in question is given by (7), which is open in x: (7)

[thez : gifted astronomer z • z ¼ x] [somey : y knew z as a youth] (x was consoled by y).

Someone uttering (5) says that Hugo Wexler satisWes it; someone uttering (6) says that every scientist who was Wred from the observatory at SoWa does. 157 For convenience, I have raised the name as if it were a quantiWer. There are other ways of obtaining the desired result. See Neale (forthcoming a).

180

I. Incomplete Descriptions

(vi) The Russellian account immediately explains the semantic diVerence between (1), which contains ‘the Wred scientist’, and (6), which contains ‘the gifted astronomer’: the respective analyses (10 ) and (60 ) are not equivalent: (10 )

(60 )

[everyx : scientist x • x was Wred from the observatory at SoWa] [thez : Wred scientist z • z ¼ x] [somey : y knew z as a youth] (x was consoled by y). [everyx : scientist x • x was Wred from the observatory at SoWa] [thez : gifted astronomer z • z ¼ x] [somey : y knew z as a youth] (x was consoled by y).

According to Wilson, however, (6) is ‘little more than a stylistic alternative’ of (1) (1991: 361). Evidently, much turns on what is meant by ‘little more’. If ‘the gifted astronomer’ in (6) is meant to be a bound variable just like ‘the Wred scientist’ in (1), then Wilson’s analyses of (1) and (6) are equivalent and ‘no more than a stylistic alternative’ of (1). But if Wilson wants his analyses to be non-equivalent, he needs to tell us what ‘little more’ there is to (6). To sum up, the Russellian has a perfectly good account of why sentences can contain descriptions that appear to be functioning as bound variables—they are bound-into. Far from presenting problems for a unitary Russellian theory of descriptions, the examples discussed by Wilson serve only to emphasize the elegance and extraordinary range of Russell’s Theory of Descriptions. A question remains, however. Why is it that deWnite and demonstrative descriptions can behave in this way whilst indeWnite descriptions cannot? Larson and Segal (1995) see the contrast between the sentences in (8) and those in (9) as supporting the contention that deWnite and demonstrative descriptions may function as bound variables: (8) (9)

[[every boy]1 ’s mother] loves him1/[that boy]1 /[the boy]1 [[every boy]1 ’s mother] loves [a boy]1 / [a unique boy]1 / [exactly one boy]1 .

In fact, the strongest conclusion that can be drawn is that deWnite and demonstrative descriptions may be elliptical for bound-into Go¨delian descriptions but indeWnites may not. If the analysis of demonstrative descriptions above is on the right track the explanation is straightforword: the demonstrative description just is the requisite indeWnite.

24. Are Pronouns Ever Wholly Bound? Flushed with the success of the Russellian analysis of Wilson’s and related examples, we might get carried away. Why not interpret the pronoun ‘him’ in (1) (1)

Every scientist who was Wred from the observatory at SoWa was consoled by someone who knew him as a youth

Neale, This, That, and the Other

181

as equivalent in this context to a relativized description [thez : Wred scientist z • z ¼ x], or [thez : scientist • z ¼ x], or with attitude contexts in mind, [thez : z ¼ x]. Why not treat all pronouns traditionally treated as bound variables as devices that only contain bound variables, as descriptions that are bound-into? On such an account all anaphoric pronouns would be D-types—no natural language bound pronoun would actually be a bound variable but all would be interpreted as bound-into. How might such an idea be motivated and implemented? These are questions I take up in Neale (forthcoming a), but I can sketch part of the story here158 Recall that I am taking seriously Postal’s idea that third person pronouns are versions of the deWnite determiner ‘the’. An occurrence of a DP ‘he’, on the syntactic theory I favour, has the structure [DP [D he][NP e] ]. Since [NP e] is aphonic, its existence in a minimalist framework must be justiWed by its role at LF. So how is it interpreted? I suggest it is interpreted as a formula xk ¼ xj (k 6¼ j), meaning [DP [D he][NP e]] is interpreted as [he xk : xk ¼ xj ], assuming an axiom for he that is a trivial modiWcation of the Russellian axiom for the (as Postal’s hypothesis would anyway suggest): [he xk : f]c is true of a sequence s iV c is true of every sequence f is true of diVering from s at most in the k-th position, and there is exactly one such sequence.

The issue is f, of course. The morphosyntax of English will, in fact, insist upon the complement of ‘he’ being [NP e]. If the semantics of [NP e] is given by xk ¼ xj (k ¼ 6 j) (at least when it functions as the complement of the determiner ‘he’, [NP e] in [DP [D he][NP e]] has a deWnite role at LF. When we have a suitably placed co-indexed binder as in (2), we get a bound interpretation of ‘he’, revealed by (20 ), interpreted as (200 ): (2) [every man]1 thinks he1 is smart’ (20 ) [DP every man]1 thinks[S [DP he2 [NP e]1 ]2 is smart]] (200 ) [every x1 : man x1 ](x1 thinks([he x2 : x2 ¼ x1 ]1 (x2 is smart) ) ). If there is no such binder, we have a free occurrence of ‘he’, used to make indexical reference to some individual. If gender is seen as important, as I suggested above in connection with Wilson’s examples, then we restate the axiom for the determiner he in such a way that f is a conjunction (xk ¼ xj • male x). Talk of bound pronouns is still perfectly intelligible on this proposal. The D ‘he’ is not bound; indeed it is a binder. And the DP ‘he’ is not wholly-bound the way a bound variable is; rather it is just bound-into. The subscript on ‘he’ in (2) must now be understood as indicating that [DP [D he][NP e]] is bound-into rather than wholly bound. Trivially, we now have a ‘uniform theory’ of pronouns, something many semanticists crave. Some have laboured to produce exotic accounts of variable-binding that might 158 Coming from a diVerent direction, a proposal similar to the one I explore has been articulated nicely by Elbourne (2001), who draws upon the idea of binding situation variables inside pronominal DPs. It seems to me that our proposals have a common spirit and diVer primarily in execution and assumed ontology, although I am not sure Elbourne would agree.

182

I. Incomplete Descriptions

draw in pronouns occurring both inside and outside the scopes (as standardly deWned) of their antecedents. My suggestion is that uniformity might be found without modifying the standard notion of binding. The proposal also seems to solve a problem raised in connection with Chomsky’s (1981) Binding Theory. Principle A of that theory states (roughly) that a reXexive pronoun must be bound by something in its own clause, correctly predicting that ‘every man’ cannot bind ‘himself ’ in (3): (3)

[S [every man]1 thinks[S I like himself1 ] ]

So how does ‘every man’ manage to bind ‘himself ’ in the isomorphic structure (4)? (4)

[S [every man]1 thinks[S he1 likes himself1 ] ].

Answer: it doesn’t: ‘every man’ binds the non-reXexive ‘he’ (satisfying Principle B, which states (roughly) that a non-reXexive must be bound by something outside its own clause), and ‘he’ in turn binds ‘himself ’ (in accordance with Principle A). That is, the structure of (4) is really something like (40 ): (40 )

[S [DP every man]1 thinks[S [DP he2 [NP e]1 ]2 likes himself 2 ] ]:

Fourth, it might also help with the matter of gender: occurrences of ‘he’, ‘him’, and ‘himself ’ anaphoric on (and within the scope of ) on a quantiWer understood as say, ‘[everyx : man x]’ might be understood as ‘[he xk : xk ¼ xj • male xk ]. Whether an occurrence of ‘he’ always makes this richer contribution to what is said is, however, debatable. There is more than one way to skin a cat.

PA RT I I

The Referential–Attributive Distinction

Russell (1905) argued that deWnite descriptions are quantiWers, thereby challenging the view that the meaning (or ‘propositional contribution’) of such an expression is identical to its referent. He was thus challenging the view that deWnite descriptions are what he called ‘logically proper names’. Strawson (1950) contested Russell’s analysis of deWnite descriptions, claiming that such expressions never function as quantiWers, but are commonly (though not invariably) used as ‘referring’ expressions. In saying this, Strawson meant only that deWnite descriptions are linguistic devices used by speakers to ‘pick out’ objects or individuals about whom they wish to speak. (He did not intend to suggest that such expressions were logically proper names.) Thus, in assertively uttering, in May of 2003: (1)

The current president of the US is a republican.

the speaker (on Strawson’s view) would be using the description to ‘pick out’ a particular individual, George W. Bush, and to say of him that he is a republican. Such a speaker would not (according to Strawson) be asserting that there is a unique current president of the US, as Russell claims. He would, however, be presupposing that there is a unique such individual to whom he is referring, and the truth-evaluability of his utterance would require the truth of the presupposition. (For more on presupposition, see Part III, ‘Presupposition and Truth-Value Gaps’.) Donnellan (1966) attempted to incorporate elements of both Russell’s and Strawson’s accounts of deWnite descriptions into his own hybrid account. In particular, he argued that while deWnite descriptions can function as quantiWers, such expressions can also function as referring expressions. In saying that deWnite descriptions can function as referring expressions, Donnellan seemed to mean that they could function as logically proper names: as expressions whose meaning is their reference. (In this respect, his notion of reference was diVerent from Strawson’s.) Thus, consider an assertive utterance of: (2)

The murderer of Smith is insane.

184

II. Referential–Attributive Distinction

If the description is used as a quantiWer (‘attributively’), the ‘statement made’1 is one to the eVect that whoever uniquely murdered Smith is insane. If the description is used as a logically proper name (‘referentially’), the ‘statement made’ is one to the eVect that x, the particular individual the speaker ‘has in mind’, is insane. Russell’s theory appears to capture (or very nearly capture) the truth conditions of sentences containing descriptions used attributively. However, it does not capture the truth conditions of sentences containing descriptions used referentially. Russell’s theory is thus incomplete at best, according to Donnellan. Strawson’s theory fares no better, as it appears to capture (at best) only the referential use. In an inXuential paper, ‘Speaker’s Reference and Semantic Reference’ (1977), Saul Kripke challenged Donnellan’s claim that the referential use of descriptions undermines Russell’s Theory of Descriptions. SpeciWcally, Kripke claimed that Donnellan was mistaking a pragmatic phenomenon for a semantic one. Although descriptions can be used referentially or attributively, their meaning is the same in both cases: arguably, as speciWed by Russell’s theory. In other words, although an assertive utterance of ‘The F is G’ might be used to communicate either a singular or general proposition, what the sentence means is the same in either case, roughly: There exists a unique F and whatever is F is G. Kripke’s central argument for this position was methodological: don’t complicate your semantics beyond necessity. To suppose that deWnite descriptions can function semantically as quantiWers or as referring terms is (according to Kripke) to do just that. For the data to which Donnellan appeals—data concerning the ‘statements made’ by assertive utterances of sentences containing deWnite descriptions—can all be accounted for on the assumption that Russell’s Theory of Descriptions is a correct theory (for English). In their contributions to this volume, Kent Bach and Nathan Salmon both claim that (contrary to Donnellan’s suggestion) there is no good reason to posit a semantically signiWcant referential use. In this respect, they are both following Kripke’s lead. Both philosophers go further (as did Kripke) and claim that there are good reasons not to posit a semantically signiWcant referential use. Despite the similarity in their basic views, they arrive at those views in very diVerent ways. Bach is unimpressed with previous attempts to argue for a semantically signiWcant referential use, and having dealt with such attempts in earlier work (Bach 1987/1994), does not focus on them in his contribution to this volume. He is instead concerned to provide positive considerations in favor of his quantiWcational approach to descriptions. His central point is a seemingly paradoxical one: the distinctive quantiWcational character of deWnite descriptions helps explain how and why they can so readily be used to refer. Bach claims that singular deWnite descriptions are a special kind of quantiWer—they imply uniqueness with respect to the property of being F , but the uniqueness is not encoded in the meaning of ‘the’. When using ‘the F ’ one is pretending, in eVect, that there is only one F , namely the F one intends the hearer to think of and to take the speaker to have in 1 The locution is Donnellan’s (1966).

II. Referential–Attributive Distinction

185

mind. But that an object is the unique F (or the only salient or contextually salient F ) does not alone explain how ‘the F ’ can be used to refer to it. Also of importance is the fact that any connection between the property of being the unique F and being G is irrelevant to what the speaker is trying to convey. In using a description referentially, the speaker intends his audience to rely on the fact that he (the speaker) uttered the description to provide a basis for Wguring out which thing is being talked about. To recognize that the speaker is doing this, the audience has to realize that the property of being F has in itself no direct bearing on what the speaker is trying to convey. Bach then goes on to discuss what is involved in a referential use of a description and in doing so he distinguishes genuine reference from spurious reference (such as ‘reference’ by description). He contends that one cannot refer to an individual to which one is not connected by a chain of perception, memory, or communication. One can refer only to individuals one is perceiving, one has perceived and remembers, or one has been informed of and remembers. Based on his conclusions about the nature and semantic insigniWcance of the referential use of descriptions, Bach attempts to clarify the nature of the referential–attributive distinction. His contention is that the proper way to understand the distinction is in terms of the contrast between singularity and generality. Roughly, singular propositions are made true by the properties of a particular entity; general propositions are made true by the properties of whatever entity happens to satisfy some speciWed conceptual content. On Bach’s view, referential utterances of ‘The F is G’ are to be understood as communicating singular propositions about the entity referred to, to the eVect that it is G. Attributive utterances of ‘The F is G’ are to be understood as communicating general propositions to the eVect that whatever is uniquely F is G. Bach concludes his contribution by responding to various objections to Russell’s theory based on incompleteness and referential use, several of which are put forth by Devitt in his contribution to this volume. Salmon’s concern is with the nature of the referential use and its alleged semantic signiWcance. Like Bach, Salmon is unimpressed with arguments for a semantically signiWcant referential use, which he has addressed in earlier work (Salmon 1982, 1991). Salmon begins with a close critical look at the distinction itself. He then makes a further distinction between correct applications of deWnite descriptions, whether referential or attributive (‘Good uses’), and referential misapplications (‘Bad uses’). Salmon argues that this particular distinction discredits one of Kripke’s objections to the ‘semantic signiWcance thesis’: the thesis according to which deWnite descriptions have a semantically signiWcant referential use. The objection of Kripke’s in question involves an attempt to extend the referential/attributive distinction to proper names. Salmon claims that Kripke casts the ordinary use of proper names on the attributive side, when in fact they are more at home on the referential side. A referential–attributive distinction for proper names is then redrawn, correcting for Kripke’s alleged error. The distinction is claimed to provide insuYcient support for Kripke’s claim that referential uses are semantically insigniWcant. A more general distinction is drawn between rival philosophical conceptions of semantics: the speech-act centered

186

II. Referential–Attributive Distinction

conception coming from Wittgenstein, and the expression centered conception of Frege and Russell. It is the former conception that, according to Salmon, leads to the mistaken notion of a semantically signiWcant referential use. Salmon then turns to various characterizations of the referential/attributive distinction. Objections are raised against characterizations of the distinction in terms of the familiar de re/de dicto distinction as applied to assertions, and against the attempt to equate semantic content with speaker assertion. In raising these objections, Salmon defends the intriguing claim that one can make a de re assertion about something even while being in no position to have any de re belief about that thing. Finally, Salmon proposes and defends a new characterization of the referential–attributive distinction. This novel characterization involves an appeal to various sorts of communicative intentions. GeoV Nunberg, building on earlier work (Nunberg 1993), furthers the case against a semantically signiWcant referential use. He points out that arguments for and against semantically signiWcant referential uses of deWnite descriptions rarely if ever appeal to what linguists regard as ‘empirical’ evidence. After exposing inadequacies in some of the standard arguments for a semantically signiWcant referential use, Nunberg draws attention to some interesting empirical data that suggest that the semantic signiWcance thesis is mistaken. The data in question involve the phenomenon of ‘deferred’ interpretation. In the particular cases in question, an expression contributes a property instantiated by the entity directly denoted, rather than the entity itself. Consider, for instance, the following two sentences. Suppose that (3) is uttered on New Year’s Eve 2003. And suppose that (4) is uttered by a prisoner on death row. Then, the expressions ‘today’ and ‘I’ in the utterances of the following contribute instantiated properties rather than individuals. (3) (4)

Today is always the biggest party day of the year. I am traditionally allowed to order whatever I want for my last meal.

In (3), we take the interpretation of ‘today’ as some property that is instantiated by the day of the utterance, rather than the day itself. The obvious property is that of being the eve of the new year. In (4), the pronoun ‘I’ does not refer to the speaker, since obviously there could not be any traditions that deal speciWcally with his last meal; instead, it refers to the role the speaker exempliWes: that of being on death row. Note, however, that analogous readings are not available for referentially used deWnite descriptions: (5) (6)

Sophie’s eleventh birthday is always the biggest party day of the year. The prisoner with the shaved head is traditionally allowed to order whatever he wants for his last meal.

In particular, the description in (5) cannot be interpreted as referring to a property of the denoted day—such as the property of being the eve of the new year. Nor can the description in (6) be taken as referring to a property of the denoted prisoner—such as being on death row.

II. Referential–Attributive Distinction

187

This is problematic for the advocate of a semantically signiWcant referential use. Suppose, as the referentialist holds, that the descriptions in (5) and (6) are semantically equivalent to demonstratives. Then, there would be no way of accounting for the discrepancies between indexicals and descriptions: no reason why the descriptions should not admit of a ‘deferred’ interpretation, one in which the semantic value of the expression is some property instantiated by the object/individual directly denoted. Under the Russellian analysis of deWnite descriptions, by contrast, these discrepancies are to be expected. On such an analysis, it is not odd that the mechanism that produces a deferred reading does not come into play. The individual that the description applies to is not a constituent of the proposition, so the hearer does not have access to any of its properties, apart from those that are made explicit by the description itself. Prima facie, this seems a telling empirical argument for saying that referentially used descriptions have a Russellian interpretation, and that the sentences that contain them have the same truth conditions as sentences containing attributively used descriptions. Michael Devitt, in contrast to the three other contributors writing on the referential–attributive distinction, argues that there are indeed semantically signiWcant referential uses. Expanding on earlier work on the topic (Devitt 1981), Devitt provides six arguments in favor of the thesis that deWnite descriptions have a semantically signiWcant referential use, Wve of which apply to indeWnite descriptions as well. One argument appeals to the fact that speakers regularly use descriptions referentially, which suggests that there is a convention of so using descriptions. If so, then the convention is surely a semantic one according to Devitt. A second argument turns the tables on an argument of Kripke’s (1977). Kripke asks us to stipulate that Russell’s Theory of Descriptions is a correct theory for a hypothetical language (‘Russell English’), a language exactly like English, but stipulated to conform to Russell’s Theory of Descriptions. Russell English would be indistinguishable from the actual English language, in which deWnite descriptions are used referentially. This shows (according to Kripke) that Russell’s Theory is not refuted by the referential use of deWnite descriptions. Devitt, eVectively inverting Kripke’s argument from ‘Russell English’, asks us to stipulate a language (‘Donnellan English’) in which there is a convention of using deWnites referentially. This language would not appear any diVerent from the English language, which suggests that it may in fact be English. A third argument appeals to the fact that referential uses have a role just like complex demonstratives and similarly depend for their reference on a perceptual/causal link to an object. This suggests that referential uses are semantically akin to complex demonstratives, and many Wnd it plausible to view complex demonstratives as referring expressions. The remaining three arguments concern a wide variety of interesting semantic phenomena, including rigidity, incompleteness, and exportation of singular terms from opaque to transparent contexts. Devitt admits with Salmon that referential descriptions are not rigid in the standard sense, but claims that they are ‘weakly’ rigid, and thus contrast with attributive descriptions, which are not. Weak rigidity of an expression requires that any objects referred to in possible worlds ‘Wt’ the descriptive content of the expression.

188

II. Referential–Attributive Distinction

Thus, ‘the murderer’ used referentially to pick out Jones in the actual world refers to Jones only in those possible worlds in which he is a murderer. In contrast to Recanati, Lepore, Neale, and Bach (see their chapters in this volume), Devitt claims that incompleteness undermines a quantiWcational treatment of referential descriptions.2 In particular, he argues that while an implicit approach will work for attributively used incomplete descriptions, neither an explicit nor an implicit approach will work for referentially used descriptions. The diYculty is that the implicit approach to referential uses cannot avoid problems of ignorance and error, problems which arise when the speaker is ignorant of or mistaken about the properties of what she refers to. And Wnally, Devitt claims that referential descriptions, in contrast to attributive descriptions, export from opaque to transparent contexts just like terms standardly treated as referring expressions, thereby lending credence to the view that such descriptions are referring terms. 2 This is the one argument that does not apply to indeWnites as well as deWnites.

4 Descriptions: Points of Reference Kent Bach

1. Introduction Russell (1905) made a compelling case that descriptions, deWnite as well as indeWnite, are devices of quantiWcation, not referring phrases. Strawson (1950) and Donnellan (1966) pointed out that deWnite descriptions can be used to refer. And even indeWnite descriptions can be used to refer. All this is old hat. If ‘On Denoting’, ‘On Referring’, and ‘Reference and DeWnite Descriptions’ had not provoked decades of debate, philosophers might have just thought it obvious that the mere fact that an expression can be used to refer does not show that it is inherently a referring expression, an expression that itself refers. After all, there is an obvious need for a distinction between linguistic meaning and speaker’s meaning, and the distinction between linguistic reference and speaker’s reference is just a special case of that (Kripke 1977: 263). Invoking this distinction does not, of course, tell us which sorts of expressions are inherently referring expressions and which are merely capable of being used to refer, but it is enough to suggest that a special reason is needed to support the claim that descriptions, which obviously can be used in nonreferring ways and which have the syntactic earmarks of quantiWer phrases, nevertheless have referential readings. Consider the sentence, ‘The discoverer of X-rays was bald’. It is one thing for a speaker to be able to use the sentence to convey the singular proposition that Ro¨ntgen was bald and quite another for the sentence itself to express that proposition. The mere fact that descriptions can be used to refer does not provide much support for the claim that when so used they are semantically referential. After all, there are many things expressions can be used to do that have non-semantic explanations. For example, socalled rhetorical questions (‘Why are you so lazy?’) are statements made using interrogative sentences. No one would seriously suggest that the interrogative form has an additional declarative meaning. Indeed, it is because a rhetorical question involves the For their very helpful comments, suggestions, and corrections, I would like to thank, and refer to not by description but by name, Barbara Abbott, Michael Devitt, Robin Jeshion, JeV King, David Sosa, and Zolta´n Szabo´.

190

II. Referential–Attributive Distinction

utterance of an interrogative sentence whose semantics accounts for its literal use to ask a genuine question that it has its force: it is taken as a statement and is intended by the speaker to be taken as such, precisely because the answer to the question is obvious. Various strong pragmatic arguments have challenged the claim that deWnite and even indeWnite descriptions sometimes function semantically as referring expressions.1 Indeed, JeV King (2001) has recently made a compelling case that the best candidate for referential descriptions, namely demonstrative descriptions (of the form ‘that F ’), are quantiWcational, even in their paradigmatic referential use. The key points of section 2 combine to provide new reason for supposing that deWnite descriptions are not semantically referential. The main point is that they can readily be used to refer precisely because they are quantiWer phrases of a certain sort. Section 3 addresses the question of what it takes to use a description to refer. One can use a description to refer to something one believes to satisfy the description, but what if one is not in a position to think of that individual, if, as Russell would say, one ‘knows it only by description’? Can one refer to it anyway? If descriptions are quantiWcational and propositions expressed by sentences containing them are general, how can one use such a sentence to convey a singular proposition involving whichever individual satisWes the description in question? Also, there is the intermediate case (between referential and merely quantiWcational uses) of the so-called speciWc use of indeWnite descriptions. It illustrates one way to fall short of referring to an object (others will be discussed too): one indicates that one has a certain individual in mind without indicating which individual it is. In short, one merely alludes to it. Section 4 revisits Donnellan’s referential–attributive distinction in light of the preceding points. I will review various ways in which it has been characterized, by Donnellan himself and by others, and try to pin down what the distinction amounts to. Also, I will take up the apparent problem for Russell’s theory posed by incomplete deWnite descriptions, whether used referentially or attributively. These are descriptions, such as ‘the table’, that are not uniquely satisWed. There are too many tables (more than one is enough) for a sentence like ‘the table is covered with books’ to be made true in the way that Russell’s theory requires. Finally, I will rebut Michael Devitt’s new arguments (this volume) that descriptions have referential meanings. 1 Regarding deWnite descriptions, see Kripke (1977), Bach (1987/1994: ch. 6), Neale (1990: ch. 3; this volume), and Salmon (1991); as to indeWnite descriptions, see King (1988) and Ludlow and Neale (1991). The general strategy has been to seek a pragmatic explanation for referential uses. Kripke was the Wrst to develop such an account, arguing that the distinction between linguistic reference and speaker’s reference is just a ‘special case’ of the distinction stressed by Grice between linguistic meaning and speaker’s meaning (Kripke 1977: 263). And Grice himself had insisted that, despite having both an ‘identiWcatory’ and a ‘non-identiWcatory’ use, ‘descriptive phrases have no relevant systematic duplicity of meaning; their meaning is given by a Russellian account’ (1969: 143). However, Kripke doubted that Russell’s account can accommodate the case of incomplete descriptions, such as ‘the table’. And Devitt (this volume) has suggested some new reasons for supposing that descriptions have referential as well as quantiWcational meanings. Kripke’s worries are addressed in Point 12, and Devitt’s arguments in Point 13.

Bach, Descriptions: Points of Reference

191

Addressing these many issues requires taking a position on what it involves to refer to a particular thing (our discussion will be limited to reference to spatio-temporal things) and on what it takes to think of one. I will assume a certain conception of singular reference, either by a linguistic expression or by a speaker.2 When reference is made by the noun-phrase itself, indicative sentences in which it occurs express singular propositions, at least when they express a proposition at all.3 The referent of the expression is a constituent of that proposition. When a speaker refers, he uses a noun-phrase to indicate which thing he is trying to convey a singular proposition about. He can use a nounphrase to refer his audience to something even if the noun-phrase does not itself refer. I also assume a certain conception of singular thought. In my view, we can have singular thoughts about objects we are perceiving, have perceived, or have been informed of (Bach 1987/1994: ch. 1). We do so by means of non-descriptive ‘de re modes of presentation’, which connect us, whether immediately or remotely, to an object. The connection is causal-historical, but the connection involves a chain of representations originating with a perception of the object. Which object one is thinking of is determined relationally, not satisfactionally. That is, the object one’s thought is about is determined not by satisfying a certain description but by being in a certain relation to that very thought (token). We cannot form a singular thought about an individual we can ‘think of ’ only under a description. So, for example, we cannot think of the Wrst child born in the twenty-second century because we are not suitably connected to that individual. Strictly speaking, we cannot think of it but merely that there will exist a unique individual of a certain sort. Our thought ‘about’ that child is general in character, not singular. Similarly, we cannot think of the Wrst child born in the fourth century bc either. However, we can think of Aristotle, because we are connected to him through a long chain of communication. We can think of him even though we could 2 Some observations on the term ‘singular’, which is contrasted with both ‘plural’ and ‘general’. There is the singular/plural linguistic distinction as to grammatical number. Subject–verb agreement is sensitive to the number of the subject, and pronouns anaphoric on noun-phrases agree with them as to number. However, the quantiWer phrases, ‘each F ’ and ‘every F ’, though grammatically singular are not semantically singular. A singular/plural distinction may also be drawn as to reference. Philosophical discussions of reference to individuals generally concentrate on singular reference, but there is also the case of plural reference, reference to more than one individual at a time. In philosophy the phrase ‘singular term’ is often used in contrast to ‘general term’. This corresponds to the notational distinction in logic between individual constants or variables and predicate letters. However, singular deWnite and even indeWnite descriptions are sometimes labeled ‘singular terms’, even though they are usually represented by quantiWers. Finally there is the distinction between singular and general propositions. The phrase ‘singular proposition’, as introduced by Kaplan (1979), applies to propositions having individuals as constituents, in contrast to general propositions, which are quantiWcational. This distinction is not exclusive, however, since a proposition may be particular with respect to one argument slot and general with respect to another, as in the case of the proposition expressed by ‘She has seven children’. 3 This qualiWcation is meant to allow for cases in which the sentence occurs fails to express a proposition, either because the referring expression, though referring in character, does not have a referent or because the sentence is semantically incomplete. Regarding the Wrst sort of case, I have in mind the view that if an expression lacks a semantic value, then sentences in which it occurs fail to express propositions. Examples of the second sort include sentences like ‘John hasn’t Wnished’ and ‘She is late’, where an additional constituent is needed to yield a complete proposition (for discussion see Bach 1994: 126–33).

192

II. Referential–Attributive Distinction

not have recognized him, just as I can think of the masked man I recently saw rob my bank. Nor does being able to think of an individual require being able to identify that individual by means of a uniquely characterizing description.4 So on my conception of singular thought, there must be a representational connection, however remote and many-linked, between thought and object. A more restrictive view, though not nearly as restrictive as Russell’s, would limit this connection to personal acquaintance (via perception and perception-based memory), and would disallow singular thoughts about unfamiliar objects. A more liberal view, though one I would contest, would allow singular thought via uniquely identifying descriptions. In any case, although I am assuming the above conception of singular thought, the questions to be asked and the distinctions to be drawn, such as the distinction between referring to something and merely alluding to or merely singling out something, do not essentially depend on that conception (of course, how one uses these distinctions to divide cases does depend on one’s conception). All that is required is the assumption that one can have singular thoughts about at least some objects one has not perceived and that only certain sorts of relations one can bear to an object put one in a position to have singular thoughts about it. What is essential to our questions is the above conception of singular reference: speaker’s reference and the hearer’s full understanding of it can be achieved only by way of a singular thought of the object. In the next three sections, I will develop and defend the following thirteen points: Referring and Describing (Section 2) Point 1. Names and descriptions can both be used merely to indicate what we are speaking about. Point 2. We generally choose the least informative sort of expression whose use will still enable the hearer to identify the individual we wish to refer to. Point 3. Often the only way to refer to something is by using a deWnite description. Point 4. Using a description to refer identiWes by implicitly conveying an identity. Point 5. The distinctive quantiWcational character of deWnite descriptions helps explain how and why they can readily be used to refer, because it plays a key role in their referential use. Referring by Describing (Section 3) Point 6. With a speciWc use of an indeWnite description, one is not referring but merely alluding to something. Point 7. One can describe a (singular) proposition without being in a position to grasp it. 4 There is the further question of whether a singular proposition comprises the complete content of a singular thought. SchiVer (1978) argued that this is not possible. In my view, de re modes of presentation are also involved (Bach 1987/1994: ch. 1). Moreover, I have argued that a belief ascription whose ‘that’-clause expresses a singular proposition does not fully individuate the belief being ascribed (Bach 1997, 2000). I point out e.g. that the one ‘that’-clause in the two ascriptions, ‘Peter believes that Paderewski had musical talent’ and ‘Peter disbelieves that Paderewski had musical talent’, does not fully characterize something that Peter both believes and disbelieves. And, as I say, every case is potentially a Paderewski case.

Bach, Descriptions: Points of Reference

193

Point 8. One can describe (single out) an individual without actually referring to it. Point 9. Descriptive ‘reference’ is not genuine reference. Point 10. Various kinds of pseudo-reference are not to be confused for the real thing. The Pragmatic Status of the Referential–Attributive Distinction (Section 4) Point 11. Various contrasts may seem to capture the referential–attributive distinction, but only one really does. Point 12. Incomplete deWnite descriptions pose no real threat to Russell’s theory. Point 13. DeWnite descriptions do not have referential meanings.

2. Referring and Describing Descriptions can be used to refer. What does that involve? Quite a bit. What does it show about the semantics of descriptions? Not much.

Point 1. Names and descriptions can both be used merely to indicate what we are speaking about. We commonly talk about particular persons, places, or things.5 We refer to them and ascribe properties to them. In so doing, we are able to accommodate the fact that an individual can change over time, that our conception of it can also change over time, that we can be mistaken in our conception of it, and that diVerent people’s conceptions of the same individual can diVer. All this is possible if in thinking of and in referring to an individual we are not constrained to represent it as having certain properties. It has long been thought that proper names, along with pronouns, are the linguistic devices best suited for doing this.6 As Mill wrote, the function of proper names is not to convey general information but rather ‘to enable individuals to be made the subject of discourse’; names are ‘attached to the objects themselves, and are not dependent on . . . any attribute of the object’ (1872: 20). According to Russell, a proper name, at least when ‘used directly’, serves ‘merely to indicate what we are speaking about; [the name] is no part of the fact asserted . . . it is merely part of the symbolism by which we express our thought’ (1919: 175).7 Russell treated deWnite descriptions diVerently, of 5 I will generally use the words ‘individual’, ‘object’, and ‘thing’ as neutral terms for persons, places, things, and other concrete objects. 6 For the sake of discussion and contrary to my own view, I am going along with the widespread view that proper names are referring terms (and that sentences in which they occur express singular propositions). However, elsewhere I have defended an alternative, a version of meta-linguistic descriptivism I call the Nominal Description Theory of names, and oVer an explanation of why it seems that proper names are referring terms (Bach 1987/1994: chs. 7 and 8; Bach 2002b). I am assuming also that there are singular propositions, which have particulars among their constituents. Anyone who has qualms about them can make do with objectdependent truth conditions. 7 Notoriously, Russell supposed that one has to be in a special, uniquely intimate relation with an object in order to think of or to refer to it, a relation which, he maintained, one cannot bear to ordinary things. Mill did

194

II. Referential–Attributive Distinction

course, since the object a description describes ‘is not part of the proposition [expressed by the sentence] in which [the description] occurs’ (1919: 170). Whereas a (genuine) name introduces its referent into the proposition, a description introduces a certain quantiWcational structure, not its denotation. The denotation of a description is thus semantically inert—the semantic role of a description does not depend on what, if anything, it denotes. In contrast, a proper name ‘directly designat[es] an individual which is its meaning’ (1919: 174).8 Nevertheless, Russell allowed that proper names can not only be ‘used as names’ but also ‘as descriptions’, adding that ‘there is nothing in the phraseology to show whether they are being used in this way or as names’ (1919: 175). Interestingly, Russell’s distinction regarding uses of names is much the same as Donnellan’s famous distinction regarding uses of deWnite descriptions (see section 4). If the property expressed by the description’s matrix (the F in ‘the F ’) enters ‘essentially’ into the statement made, the description is used attributively;9 when a speaker uses a description referentially, the speaker uses it ‘to enable his audience to pick out whom or what he is talking about and states something about that person or thing’ (1966: 285). Donnellan’s distinction clearly corresponds to Russell’s. Whereas an attributive use of a deWnite description involves stating a general proposition, as with the use of a proper name ‘as a description’, a referential use involves stating a singular proposition, just as when a proper name is used ‘as a name’. And just as Russell comments that ‘there is nothing in the phraseology’ to indicate in which way a name is being used, so Donnellan observes that ‘a deWnite description occurring in one and the same sentence may, on diVerent occasions of its use, function in either way’ (1966: 281). If Russell and Donnellan are right, respectively, about proper names and deWnite descriptions, then expressions of both sorts can be used referentially (as a name, to indicate what we are speaking about) or attributively (as a description). This leaves open whether either sort of expression is semantically ambiguous or whether, in each case, one use corresponds to the semantics of the expression and the other use is accountable pragmatically from that use.10 For Russell a deWnite description, not suppose this, and neither do I. My view of what it takes to be in a position to think of or refer to something is far more liberal than Russell’s. 8 Thus the relation of a description to what it denotes is fundamentally diVerent from the relation of a name to what it refers to. This is suggested by the fact that we speak of individuals satisfying descriptions but not of them satisfying names. In Russell’s terminology, descriptions ‘denote’ what they describe but do not refer to them. And notice Russell’s use of the adverb ‘directly’ in characterizing how names designate their objects, just as Kaplan (1989a) characterizes indexicals and demonstratives as ‘directly referential’. However, given the distinction between denotation and reference, the occurrence of ‘directly’ in Kaplan’s ‘directly referential’ is redundant, and ‘indirectly referential’ is an oxymoron. Also, when Kripke (1977) speaks of the ‘semantic reference’ of a deWnite description, he can only mean its denotation. So if we distinguish reference from denotation as two diVerent species of what Kripke calls ‘designation’, then all referring expressions are rigid designators and all denoting expressions are nonrigid designators, except those that are rigid de facto, like ‘the smallest prime’ (Kripke 1980: 21). None of these terminological points addresses the question of which expressions are referring (as opposed to which ones, though not referring, can be used to refer). 9 Assume here and in general that the description occurs in a simple sentence of the form ‘the F is G’. 10 If it is, then it is arguably nonliteral, though in an innocuous sort of way (see Bach 1987/1994: 69–74). For my take on the semantic–pragmatic distinction, see Bach 1999a.

Bach, Descriptions: Points of Reference

195

whichever way it is used, is inherently a quantiWer phrase, whereas a ‘logically proper’ name is a referring term.11 Donnellan was evidently unsure whether to regard the referential–attributive distinction as indicating a semantic ambiguity or merely a pragmatic one.12 But at least this much is clear: regardless of the semantic status of the distinction between using a description referentially (as a name) or attributively (as a description), there is a deWnite diVerence between asserting a singular proposition and asserting a general one.13

Point 2. We generally choose the least informative sort of expression whose use will still enable the hearer to identify the individual we wish to refer to. Suppose you want to refer to your mother-in-law. In some circumstances, it may be enough to use the pronoun ‘she’. The only semantic constraint on what ‘she’ can be used to refer to is that the referent be female (ships and countries excepted). So its use 11 Of course Russell held that ordinary proper names are ‘disguised’ or ‘truncated’descriptions, in which case they too, contrary to appearances, are quantiWcational. 12 Recanati (1993: ch. 15) and Bezuidenhout (1997), deny that deWnite descriptions are semantically ambiguous. However, they are disinclined to treat one use as literal and to explain the other pragmatically, because intuitively they Wnd referential uses to be no less literal than attributive ones. Accordingly, they suggest that the existence of both uses is symptomatic of semantic under-determination (or what Recanati calls, borrowing a phrase from Donnellan, ‘pragmatic ambiguity’). From this it follows, implausibly, that a sentence like ‘The discoverer of X-rays was bald’ does not express a determinate proposition (Recanati allows that the meaning of such a sentence determines an ‘external’ proposition, which he distinguishes from any proposition that the sentence expresses in context, but it turns out that this is the same uniqueness proposition that is expressed on an attributive use of the description). If we wish to maintain that such a sentence does express a determinate proposition, and does so univocally, the obvious choice is a general proposition, in which case the description functions as a quantiWer phrase and only its attributive use is the strictly literal one. And, as Point 5 below suggests, the referential use, though requiring a pragmatic explanation, is partly explained by the distinctive quantiWcational character of descriptions. 13 The type of general proposition is what Strawson (1950) called a ‘uniquely existential’ proposition. I shall call it a simply ‘uniqueness’ proposition (once we opt, as below, for restricted quantiWcation notation rather than that of standard Wrst-order logic, Strawson’s label no longer applies). Russell analyzed a uniqueness proposition of this sort into a complex logical form, expressible in the notation of modern Wrst-order logic as: (LF-dd) (9x)(Fx&(8y)(Fy y ¼ x)&Gx). The key point is that the object that is the F does not appear in this proposition. So, e.g., ‘The queen of England loves roses’does not express a proposition about Elizabeth II. It means what it means whether or not she is queen of England and, indeed, whether or not England has a queen. As a quantiWer phrase, ‘the queen of England’does not refer to Elizabeth II (Russell would say it ‘denotes’ her, but for him denotation was a semantically inert relation), though of course it can be used to refer to her. Now it is commonly and justly complained that the above form distorts syntax beyond recognition. However, as Neale (1990) has stressed, such syntactic complexity is not essential to Russell’s theory of descriptions, in particular to his view that deWnite descriptions are ‘incomplete symbols’. What makes them incomplete is not that they ‘disappear upon logical analysis’, making grammatical form misleading as to logical form, but that they introduce quantiWcational, variable-binding structure, rather than individuals, into the propositions expressed by sentences in which they occur. Besides, the repugnant complexity of the above form is an artifact of the notation used in Wrst-order logic (as is that of the usual symbolizations of English sentences with ‘every’ or ‘some’). It is well-known that quantiWer phrases in general are not amenable to Wrst-order treatment (Barwise and Cooper 1981), and Neale, following Barwise and Cooper, urges the use of restricted quantiWcation as a uniform device for representing sentences containing any sort of quantiWer phrase. In the simplest case it takes the form ‘[Qx: Fx] Gx’, so that ‘the F ’ is represented as ‘[the x: Fx]’. This avoids the tortured syntax introduced by Russell’s theory.

196

II. Referential–Attributive Distinction

provides only the information that the intended referent is female. If it is to be used successfully to refer the hearer to a certain female, there must be some female that the hearer can reasonably suppose the speaker intends to be referring to. If out of the blue you said, ‘She is unforgettable’, intending with ‘she’ to refer to your beloved mother-inlaw, you could not reasonably expect to be taken to be referring to her. However, if she were already salient, by being visually present and prominent or by having just been mentioned, or you made her salient in some way, say by pointing to a picture of her, then using ‘she’ would suYce. In other circumstances, you would have to use some more elaborate expression. For example, to distinguish her from other women in a group you could use ‘that woman’, with stress on ‘that’ and an accompanying demonstration. Or, assuming the hearer knows her by name, you could refer to her by name. Otherwise, you would have to use a deWnite description, say ‘my beloved mother-in-law’. This example suggests that a speaker, in choosing an expression to use to refer the hearer to the individual he has in mind, is in eVect answering the following question: given the circumstances of utterance, the history and direction of the conversation, and the mutual knowledge between me and my audience, how informative an expression do I need to use to enable them to identify the individual I have in mind? Note that informativeness here can depend not only on the semantic information encoded by the expression but on the information carried by the fact that it is being used. Some linguists have suggested that which sort of expression one must use depends on the degree of ‘givenness’ (or ‘familiarity’ or ‘accessibility’) of the intended referent. Giving my own gloss (in parentheses) on the well-known scale proposed by Gundel, Hedberg, and Zacharski (1993), we can distinguish being: • in focus (being the unique item under discussion or current center of mutual attention) • activated (being an item under discussion or being an object of mutual awareness) • familiar (being mutually known) • uniquely identiWable (satisfying a deWnite description)14 Notice that in the Wrst two cases the referent is already an object of mutual awareness, that in the third case it is at least mutually known, and that only in the last case may it be something unfamiliar to the hearer. So we should keep in mind the distinction between keeping the hearer’s attention on something he is already attending to, calling

14 Gundel et al. (1993) also include the statuses of being referential and being type-identiWable, but both are problematic. Being referential is a property of expressions, not objects of reference (strangely, at one point (p. 277) they speak of ‘expressions which are referential but not uniquely identiWable’), and it emerges that the intended diVerence between being uniquely identiWable and merely referential turns on whether the expression being used to refer, without consideration of the rest of the sentence, suYces to enable the hearer to identify the reference. Being type-identiWable is the weakest category in Gundel et al.’s hierarchy. It does not distinguish one member of a type from another and is not even a weak way of being able to be referred to. So it really does not belong on the list.

Bach, Descriptions: Points of Reference

197

his attention to something he is already familiar with, and bringing to his attention something unfamiliar to him. Gundel et al. plausibly claim that correlated with each status is a type of expression most appropriately used to refer to items with that status: • • • •

in focus: unstressed pronominals (and, in some languages, zero pronominals) activated: stressed third-person pronouns and simple demonstratives familiar: demonstrative phrases uniquely identiWable: deWnite descriptions

For some reason Gundel et al. do not discuss Wrst- and second-person pronouns and proper names. The reference of singular Wrst- and second-person pronouns is determined by linguistic role (the speaker, the hearer), but with ‘we’ and plural ‘you’ there must be a contextually identiWable constraint on the relevant group. As for proper names, generally it is appropriate to use them only if one’s audience is familiar with the name and knows its bearer by name. Otherwise an introduction is in order. Now Gundel et al. suggest that diVerent degrees of givenness are not merely associated with but, as a matter of linguistic convention, are encoded by diVerent types of referring expressions. Perhaps they suggest this because, taking their scale to concern the cognitive status of representations in the mind of the hearer, they think this status has to be linguistically marked if it is to play a cognitive role. As I see it, however, this scale concerns the mutual (between speaker and hearer) cognitive status of the intended referent. After all, in using an expression to refer the speaker aims to ensure that the hearer thinks of the very object the speaker is thinking of, and what matters is that the expression used to refer, and the fact that the speaker is using it, provide the hearer with enough information to Wgure out what he is intended to take the speaker to be thinking of, hence to think of it himself. This is why there is a parsimonious alternative to Gundel et al.’s conventionalist view: the diVerent degrees of givenness associated with diVerent types of referring expressions are not encoded at all; rather, the correlation is a by-product of the interaction between semantic information that is encoded by these expressions and general facts about rational communication.15 On 15 There is an obvious problem with the view that degree of reference accessibility is encoded in expressions used to refer: most expressions that can be used to refer do not have to be. Unfortunately, the linguistic literature on accessibility and givenness tends to ignore the diVerence between referential and non-referential uses. If this diVerence were taken into account, it would be realized that unless the expressions in question were systematically ambiguous, which there is no reason to believe, it is incoherent to hold that they can encode degrees of accessibility just for when they are used to refer. Also, regarding the question of when it is appropriate to use a deWnite description rather than an indeWnite one, linguists often suppose (see Abbott forthcoming), that the use of a deWnite description requires familiarity with what is described and that this familiarity is somehow encoded by the deWniteness of the descriptions. But that cannot be right, since deWnite descriptions can be used attributively, in which case the speaker need not be familiar with what satisWes the description. Moreover, the requirement of familiarity, such as it is, is a simple consequence of the fact that the descriptive material in most descriptions does not apply (and is presumed not to apply) to exactly one individual. Most deWnite descriptions are not like ‘the inventor of the zipper’ or ‘the queen of England’. However, once an individual is introduced as, say, ‘a banker’, he can subsequently be referred to as ‘the banker’, i.e. the banker being discussed or the contextually relevant banker.

198

II. Referential–Attributive Distinction

this, the null hypothesis, it is because diVerent expressions are more or less informative that the things they can be used to refer to are less or more given or accessible. That is, the more accessible the referent is, the less information needs to be carried by the expression used to refer to it to enable the hearer to identify it. Notice that not only is it enough to use the least informative sort of expression needed to enable your audience to identify the individual you have in mind, it is misleading to use a more informative one. For example, in telling a story about a particular individual, it is always suYcient, once the individual is introduced, to use a personal pronoun—provided, of course, that no other individual of the same gender has been introduced in the meantime. There are stylistic or other literary reasons to use their name or a deWnite description every so often, but unless it is obvious that this is the name or a description of the individual in question, it would be inferred that reference is being made to some other individual. This inference would be made on the charitable assumption that one is not being needlessly informative (and violating Grice’s (1989: 26) second maxim of quantity).

Point 3. Often the only way to refer to something is by using a deWnite description. Suppose you want to refer to some thing (or someone). Suppose it is not perceptually present, has not just come up in the conversation, and is not otherwise salient. Suppose that it does not have a name or that you are unaware of its name or think your audience is unaware. Then you cannot use an indexical, a demonstrative (pronoun or phrase), or a proper name to refer to it. If you want to refer to it, what are you going to do? Unless you can Wnd it or a picture of it to point to, you need to use a linguistic expression, some sort of singular noun-phrase (what else?), to call it to your audience’s attention. You must choose one that will provide your audience with enough information to Wgure out, partly on the supposition that you intend them to Wgure this out, which object you are talking about. Your only recourse is to use a description. This raises the question, when you use a description, how does your audience know that you are referring to something and expressing a singular proposition, rather than making a general statement and expressing an existential or a uniqueness proposition? Although the presence of a description does not signal that you are referring—semantically, descriptions are not referring expressions—what you are saying might not be the sort of thing that you could assert on general grounds, that is, as not based on knowledge of some particular individual (see Ludlow and Neale 1991). This will certainly be true whenever it is mutually evident which individual satisWes the description in question and what is being said regarding the individual that satisWes the description can only be supposed to be based on evidence about that individual. For example, if my wife says to me, ‘The DVD player is broken’, I can’t not take her to be talking about the actual DVD player of ours. On the other hand, if before we decided

Bach, Descriptions: Points of Reference

199

on a DVD player she said, ‘The DVD player had better not cost more than $500’, clearly she would be making a general statement pertaining to whichever DVD player we buy (notice that the corresponding demonstrative description, ‘that DVD player’, is usable only in the latter, non-referential case). Also, its being mutually evident which individual satisWes a description will generally be suYcient for a referential use, since there will usually be no reason for the hearer not to be taken as making a singular statement about that individual. This applies especially to descriptions of occupiers of social positions or practical roles, such as ‘the boss’ or ‘the freezer’. Moreover, if the description is incomplete, as in these cases, and there is no mutually salient or obviously distinctive completion in sight, then the hearer, at least if the boss or freezer in question is mutually familiar, can only take the description as being used referentially.16 But if ‘the F ’ is incomplete and it is obvious that the hearer is unfamiliar with the relevant F, then a (referential) use of ‘the F ’ must be preceded by an introduction of the relevant F.

Point 4. Using a description to refer identiWes by implicitly conveying an identity. In using a noun-phrase to refer to a certain individual, you aim to do two things: to get your audience to think of that individual and to take that individual to be the one you are thinking of, hence the one you are referring to. As in Point 3, if you do not have a name at your disposal for what you wish to refer to, and an indexical or a demonstrative pronoun or phrase won’t do the trick, the only available sort of noun-phrase left is a description. You must describe what you have in mind if you are to get your audience to think of it too and to take you to be referring to it. You do this by exploiting a presumed identity between the individual you wish to refer to (thought of by means of the mental counterpart of a singular term) and that which satisWes the description (perhaps as contextually restricted—see Point 12). To appreciate how this works, let us take the perspective of the audience and ask what is involved in recognizing the intention behind a referential use. We should keep in mind that recognizing a referential intention is part of recognizing a communicative intention and that this involves looking for a plausible explanation for the fact that speaker said what he said, partly on the Gricean basis that he intends one to do so (see Bach and Harnish 1979: 4–8, 12–15, 89–93). According to Russell’s theory, a sentence of the form ‘The F is G’ expresses a general (uniqueness) proposition. Then if you utter such a sentence but use the description 16 Why mutually salient or familiar? Obviously it is not enough for the intended referent to be salient or familiar merely to the speaker, if it is not salient or familiar to the audience and if this is not evident to the speaker (etc.). A speaker presumes that it is when using an expression that requires it to be; if his presumption is mistaken, the referent will not be evident to his audience. So, in general, when I say that something is salient or familiar, I will mean that it is mutually so. Similarly, when I speak of something’s being obvious in a conversational situation, I will mean mutually obvious.

200

II. Referential–Attributive Distinction

referentially, what you say is a general proposition but what you mean is a singular one.17 But how and why does the hearer take you to be doing that? So, for example, if you uttered, ‘The mailman is dangerous’, I would take you to be asserting not a general proposition but a singular one (about the mailman, i.e. the neighborhood mailman, our mailman). Why would I do that? Well, I am acquainted with the mailman and presumably so are you. Besides, that a certain individual is the mailman has nothing to do with his being dangerous. To suppose that it does would be to take you to be stating something for which you have no evidence (you would be violating Grice’s (1989: 27) second maxim of quality). So I have no reason to suppose, as if you were unfamiliar with the mailman, that you are making a general statement, the content of which is independent of which man is the mailman. It is unreasonable for me to suppose that you do not have in mind the particular man who delivers mail in our area and that you are not basing the belief you are expressing on information or evidence derived from him. So I have positive reason to think that you have in mind, and intend me to think you have in mind, a certain individual who satisWes the description you are using. Now for me to think you intend this requires my understanding what you are saying, namely, that the mailman is dangerous, but that is a general proposition. So in attributing to you the intention to be making a singular statement, I am in eVect explaining why you said what you said, and thereby able to infer what you meant. If you are using the description to refer and I am taking you to be doing so, we must have ways of thinking of the individual in question, the mailman, in some other way than as the mailman. Presumably we both remember him by way of a memory image derived from seeing him. In thinking of him via that image, you take him to be the mailman and use the description ‘the mailman’ to identify him for me, which triggers my memory of him. We both think of him, via our respective memories of him, as being the mailman. This Wts with how Mill describes the functioning of a proper name in thought as an ‘unmeaning mark which we connect in our minds with the idea of the object, in order that whenever the mark meets our eyes or occurs to our thoughts, we may think of that individual object’ (Mill 1872: 22). Though not ‘unmeaning’, a deWnite description can play a similar role. In using a description referentially, you are using it in lieu of a sign for the object. You are not making a general statement about whatever satisWes the description but a singular statement about some unnamed, undemonstrated, and otherwise unsalient individual. You are thinking of a certain individual by means of the mental counterpart of a singular term but there is no suitable linguistic singular term for you to use. In using the F (in ‘the F is G’) you are implicitly indicating that this individual is the F and 17 Here and throughout I am assuming a distinction between saying and meaning or stating, a distinction which I have tried elsewhere to vindicate (Bach 2001a). It corresponds to Austin’s distinction between locutionary and illocutionary acts. This distinction is often blurred, e.g. by Donnellan (1966), whenever he suggests that in using a description referentially rather than attributively, one is saying something diVerent, allegedly because the content of the description does not enter into what is said. See Point 11 below.

Bach, Descriptions: Points of Reference

201

that it is this individual that you are stating to be G. What matters to your assertion is that this individual is G, not that it is (the) F. So the content of the description is inessential to what you are stating (though not to what you are saying). In eVect, you are using ‘the F ’ to instruct the hearer to think of, and take you to be talking about the individual that is the F.

Point 5. The distinctive quantiWcational character of deWnite descriptions helps explain how and why they can readily be used to refer, because it plays a key role in their referential use. So far as I know, no one has ever argued that the mere fact that a quantiWer phrase can be used to refer shows that it semantically refers, hence that it has a referential as well as a quantiWcational meaning. No one has ever argued that because, say, ‘three thugs’ might be used (say in ‘Three thugs are outside’) to refer to the three thugs that the speaker knows the hearer is expecting, this phrase has a referential reading. So why suppose that descriptions have such readings? Why isn’t it enough simply to appeal to the distinction between what a speaker says and what a speaker means and, in particular, to take into account the distinction between linguistic reference and speaker’s reference? For one thing, as Devitt (this volume) points out, unlike other sorts of quantiWer phrases that can be used to refer, deWnite descriptions are regularly so used. This is one of his reasons, to be rebutted in Point 13, for attributing referential meanings to deWnite descriptions. However, as I will now suggest, the distinctive quantiWcational character of singular deWnite descriptions actually helps account for the fact that, despite being quantiWer phrases, they can readily be used to refer. They are quantiWer phrases of a distinctive kind and, moreover, they imply uniqueness (besides, as Point 3 indicated, sometimes there is no singular noun-phrase of any other sort suitable or even available to use to refer to the intended object). DeWnite descriptions are quantiWer phrases of a special sort. In general, sentences of the form ‘Q F is G’ or ‘Q Fs are G’ say how many Fs are G. That is, they express propositions whose truth requires that the set of Gs include a set of Fs of the size speciWed (precisely or vaguely), by ‘Q’ (‘one F ’, ‘ten Fs’, ‘a few Fs’), and in some cases a set of no greater size (‘exactly one F ’, ‘exactly ten Fs,’ ‘few Fs’). Any set of Fs of the right size will do (universal quantiWers are a degenerate case, because there is only one such set). The quantiWer phrase ‘Q F(s)’ indicates how many Fs have to be G for the sentence to be true. DeWnite quantiWer phrases do more than that: they indicate how many Fs there are. In ‘The nine planets have elliptical orbits’, ‘the nine planets’ indicates how many planets there are (nine) as well as how many must have elliptical orbits for the sentence to be true. In ‘Both authors of Principia Mathematica were English’, ‘both authors of Principia Mathematica’ indicates how many authors of Principia Mathematica there were (two), in ‘the inventor of television died in poverty’, ‘the inventor of television’ indicates how many inventors of television there were (one), etc. Even the

202

II. Referential–Attributive Distinction

vague ‘the few surviving condors’ and ‘the many Enron employees’ fall, in a vague way, into this category.18 When using a deWnite quantiWer phrase, one is not just saying how many Fs are G, one is eVectively saying (or at least implying) how many Fs there are. When a sentence of the form ‘Q F is G’ or ‘Q Fs are G’ contains such a quantiWer phrase, there is one particular individual or set of individuals to which the predicate ascribes a property. This individual or set of individuals does not enter into the proposition expressed by the sentence—the proposition is general, not singular—but this proposition is made true or false by facts involving this individual or set of individuals. Accordingly, the question can arise of which individual(s) this is. The speaker may or may not have any idea which one(s) this is, but there is a determinate one (or set) to be thought of. In this way the possibility of referring can arise without any special stage setting. Now singular deWnite descriptions are a special case, for they imply uniqueness.19 Notice, however, that this uniqueness is not encoded in the meaning of ‘the’. As Richard Sharvy observed, recognizing that singular deWnite descriptions have something in common with plural ones, ‘the primary use of ‘‘the’’ is . . . to indicate totality, implication of uniqueness is a side eVect’ (1980: 623). That is, it is a kind of universal quantiWer, and the side eVect is the result of combining ‘the’ with a nominal in the singular. Sometimes the nominal itself implies uniqueness, as with descriptions containing superlatives (‘the shortest spy’) or indicative of unique positions (‘the queen of England’) or accomplishments (‘the inventor of the Segway’). When it does not, you may need to use ‘the one F ’, ‘the only F ’, or even, to forestall any suspicion that there is more than one F, ‘the one and only F ’. Nevertheless, ‘one’ and ‘only’ are redundant in these phrases, since uniqueness is already implied by the combination of singularity and totality. Just as ‘both Fs’ is equivalent to (the ungrammatical) ‘all two Fs’, so ‘the F ’ is equivalent to (the ungrammatical) ‘all one F ’.20 18 Notice only deWnite quantiWer phrases can be complete or incomplete. A deWnite ‘Q F ’ is complete iV the number of Fs is in fact the number (however vaguely speciWed) indicated by ‘Q’. So ‘the ten Fs’, ‘both Fs’, and even ‘the few Fs’ can be complete or incomplete. Of course, any quantiWer phrase can be used with some implicit restriction intended by the speaker, but that does not make it incomplete. For example, you could say, ‘Some students Xunked’, and mean that some students in your current logic class Xunked (that class). But ‘some students’ is not incomplete, and ‘some students in my current logic class’ is not complete in the above sense, for it does not indicate how many students are in the class. 19 It is commonly supposed that singular possessive descriptions also imply uniqueness, but this is far from clear. I can refer to one of my brothers as ‘my brother’, introduce one of my friends as ‘my friend’, and complain that I sprained ‘my ankle’, even though I have two ankles. Can these descriptions be included in the category of incomplete deWnite descriptions, as Neale (forthcoming) would insist, or are they genuine instances of what Abbott (forthcoming) calls the ‘uniqueness problem’? Szabo´ (2000) and Ludlow and Segal (this volume) oVer various examples intended to show that even singular descriptions introduced by ‘the’ do not imply uniqueness. However, such examples have been explained away by Abbott (1999) and by Birner and Ward (1994), who also point out that such non-unique uses tend not be productive and seem to have a kind of generic character. Compare e.g. ‘John went to the hospital’ with ‘John went to the hospital near his house’ or with ‘A Wre truck went to the hospital’. 20 It seems to me, contrary to Szabo´ (2000), that there is nothing linguistically odd about the fact that the simplex quantiWers ‘both Fs’ and ‘the F ’ have complex semantic speciWcations. It is predictable that they are special cases, because of the unavailability of ‘all two Fs’ or ‘all one F ’.

Bach, Descriptions: Points of Reference

203

Uniqueness is implied even if ‘the F ’ is incomplete, that is, even if there is more than one F. This does not show that the description contains some hidden modiWer that would make it complete or some phantom variable of domain restriction. The situation is much simpler. As Point 12 explains, in using an incomplete deWnite description, one is not using the description is a strictly literal way, even though one is using ‘F ’ literally. What one means is distinct from what one says—it goes beyond the semantic content of the sentence one is uttering—since one is not making fully explicit what one means. If one is using the description attributively, what one means is expressible by some more elaborate description, for example of the form ‘the F that is G’. If one is using it referentially, one is referring to a certain F, an F that is salient or contextually relevant. One is not using ‘the F ’ to mean ‘the salient F ’, ‘the contextually relevant F ’, ‘the F I have in mind’, or anything of the sort. Rather, it is as if there is only one F, at least for the present purposes of the conversation. In eVect, one is pretending that there is only one F, namely the F one intends the listener to think of and to take one to have in mind. That an object is the unique F (or the only salient or contextually relevant F ) does not alone explain how ‘the F ’ can be used to refer to it. There is also the fact that any connection between the properties of being the unique F and being G is irrelevant to what you are trying to convey. Consider cases in which there is a connection, for example, ‘The next pet I get will be a cat’, said long before you have decided on a pet, or ‘The next president will be a Democrat’, said before anyone has any idea which Democrats will even be running. In cases like these you could also have included ‘whatever it is’ or ‘whoever he is’ to indicate that you do not know which individual satisWes the description. In such cases, your audience does not have to identify some individual as the F. In using a description referentially, however, you intend your audience to rely on the fact that you uttered the description as a basis for Wguring out which thing you are talking about, by way of identifying what satisWes that description. To recognize that you are doing this, the audience has to realize that the property of being F has in itself no direct bearing on what you are trying to convey. They have to recognize that you are using ‘the F ’ to enable them to think of a certain individual and to instruct them to take you to be talking about a certain individual that is the F. In sum, the quantiWcational character of a deWnite description plays a role in its referential use. The speaker thinks of a certain object, takes that object to be the F, and uses ‘the F ’ to refer to it. The speaker, on hearing ‘the F ’, thinks of a certain object that he takes to be the F, and takes that to be what the speaker is referring him to. His thinking of that object may depend in part on the fact that it is the only plausible candidate for what the speaker has in mind in using ‘the F ’. Moreover, the property of being F plays an essential role in this process. Indeed, if the speaker uses ‘the F ’ to refer to a certain object that isn’t F, he is speaking falsely or not literally. Or he could be exploiting the hearer’s false belief that the object is F. In any case, if he uses ‘the F ’ to refer to an object that isn’t F, then even if what he means is true, what he says is false—

204

II. Referential–Attributive Distinction

unless the actual F has the ascribed property. So consider this dialogue between two people who watched TV one morning in May 2002, one of whom has confused two diVerent gesticulating curly-haired guests: A: The inventor of the Segway, that young guy we saw on the Today Show this morning, is very clever. B: No, that was Jason Stanley. Dean Kamen is the inventor of the Segway, he’s older, and he was on Good Morning America. A: Oh yeah, Dean Kamen. Well, like I said, the inventor of the Segway is very clever.

Both of A’s uses of ‘the inventor of the Segway’ are referential, Wrst to Stanley and then to Kamen. He meant two diVerent things, even if both are true, but what he said was the same each time, a general proposition made true or false, but not about, Dean Kamen. In both cases he conveyed an identity (Point 4), Wrst a false one, which B recognized as such, and then a true one. In both cases the quantiWcational character of the description was operative.

3. Referring by Describing So far I have just assumed that to refer to something one must be in a position to have singular thoughts about it and that the propositions one attempts to communicate in the course of referring to it are singular with respect to it (I have assumed also that being in a position to have a thought about a particular (spatio-temporal) thing requires being connected to that thing, via perception, memory, or communication). The question to be taken up now concerns whether it is necessary to be in such a position in order to refer to something. Put it this way: can one communicate a singular proposition about something one is not in a position to have singular thoughts about? The following points address diVerent aspects of this question and certain issues underlying it. They illustrate various ways in which uses of descriptions, indeWnite as well as deWnite, can come close to being referential but fall short. As we have seen, in using a description ‘the F ’ referentially (in uttering a sentence of the form ‘The F is G’), one is trying to convey that a certain thing, to be identiWed as the F, is G. One is communicating the singular proposition that this thing is G, so that understanding one’s utterance requires grasping that proposition. In order to appreciate how seemingly referential uses of deWnite descriptions can fall short of actually being referential, I will Wrst take up the case of speciWc uses of indeWnite descriptions, a phenomenon of interest in its own right.

Bach, Descriptions: Points of Reference

205

Point 6. With a speciWc use of an indeWnite description, one is not referring but merely alluding to something. IndeWnite descriptions can be used nonspeciWcally, referentially, or speciWcally.21 In the very common nonspeciWc (or purely quantiWcational) use, there is no indication that the speaker has any particular thing in mind, as with a likely utterance of (1), (1)

A martini will cheer you up.

This is a clearly quantiWcational use of an indeWnite description. No singular proposition is being conveyed or even in the oYng. As for the referential use, it is relatively rare, as it is with quantiWcational phrases generally. Consider an utterance of (2), for example, made while watching a wellknown philosopher on TV discussing how the Fed should control inXation. (2)

A philosopher thinks he’s an economist.

It is obvious that the speaker is not making a general statement, about no philosopher in particular. He is clearly talking about the philosopher he is seeing on TV. Perhaps he is implicating that there is something incongruous about a philosopher pontiWcating on the economy, but it is clear that he is talking about this particular philosopher. What is distinctive about the speciWc use of an indeWnite description is that the speaker communicates that he has a certain individual in mind, but he is not communicating which individual that is—he does not intend you to identify it. Suppose someone says, (3)

A famous actress will be visiting us today.

Unless he thinks this is the sort of day for a visit by a famous actress, presumably he has a particular one in mind (he could have made this clear by including the word ‘certain’ (or ‘particular’), as in: (30 )

A certain famous actress will be visiting us today.

He could even explain why he is not specifying which actress it is: (3)

a. A famous actress will be visiting us today, but I won’t tell you who. b. A famous actress will be visiting us today, but I want her to be a surprise.

21 They can also be used predicatively, as when one refers to an object in one way and then describes it as a such-and-such, as when you say of the thing in your hand, ‘This is a pomegranate’. It is arguable that this is not really a quantiWcational use but a distinctively predicative use, no diVerent in kind from saying, ‘This is red’. It is merely because phrases containing singular common nouns require (in English) an article that one cannot say, ‘This is pomegranate’ (one could say the equivalent of this in Russian). GraV (2001) has argued that all uses of indeWnite descriptions are fundamentally predicates and even extends her arguments to deWnite descriptions. Her account also covers generic uses of both deWnite and indeWnite descriptions, as in ‘The tiger has stripes’ and ‘A philosopher is not in it for the money’.

206

II. Referential–Attributive Distinction

In a speciWc use, the speaker indicates that he is in a position to refer to a certain individual, but is not actually doing so. He is not identifying or trying to enable the hearer to identify that individual—he is merely alluding to her. He has a certain singular proposition in mind but is not trying to convey it. So what must the hearer do in order to understand the utterance? It would seem that she must merely recognize that the speaker has some singular proposition in mind, about a certain individual of the mentioned sort, in this case a famous actress.22 It might be objected that a speciWc use of an indeWnite description is a limiting case of a referential use, not mere allusion but what might be called ‘unspeciWed’ reference. After all, can’t the hearer, recognizing that the speaker has some individual in mind, at least think of that individual under the description ‘the individual the speaker has in mind’? But, as Point 9 will say, descriptive ‘reference’ is not genuine reference. Besides, the speaker is not really referring the hearer to that individual and, in particular, does not intend her to think of the individual he has in mind under the description ‘the individual you (the speaker) have in mind’ or in any similar way. He is merely indicating that he has a certain unspeciWed individual in mind. That is, he is not referring but merely alluding to that individual. To appreciate why this is, consider a situation in which the speaker has in mind one F among many and proceeds to say something not true of that individual. Suppose a teacher has a particular student in mind when she utters (4)

A student in my ethics class threatened me yesterday.

but does not specify which student (and does not expect that to be evident). Obviously the words ‘a student in my ethics class’do not refer to the student she had in mind, for if some student threatened her but she is mistaken about which one, (4) would still be true. So (4) expresses a general proposition. Even so, since she does have a certain student in mind, could she be using this indeWnite description to refer to that student? Even if what she said is a general proposition, is what she meant a singular proposition, about the student she had in mind? No, because the audience could understand her perfectly well without having any idea which student she has in mind. They understand merely that she has a certain student in mind, the one she is alluding to. Indicating that one has a certain unspeciWed individual in mind does not suYce for referring to that individual. If it did, then any way of coming to know that someone else has some individual in mind would be enough to put one in a position to think of and to refer to that individual oneself. 22 I do not mean to suggest that ‘a certain F ’ is always used to indicate that the speaker has a particular unspeciWed individual in mind. He might merely have in mind some unexpressed restriction on ‘F ’. For example, one might say, ‘A certain contestant will go home happy’, without specifying that whoever wins the contest in question will go home happy. Similarly, an utterance of the quantiWed ‘Every author loves a certain book’ could be made true if every author loves, say, the Wrst book he wrote.

Bach, Descriptions: Points of Reference

207

Point 7. One can describe a (singular) proposition without being in a position to grasp it. My discussion of understanding a speciWc use of an indeWnite description illustrates that it is one thing to entertain a singular proposition and another thing merely to know that there exists a certain such proposition. Russell’s famous discussion of Bismarck illustrates how this can be. He operates with a notoriously restrictive notion of acquaintance, but this is not really essential to the distinction he is drawing. I agree with Russell that we cannot have singular thoughts about individuals we ‘know’ only ‘by description’, but I will not assume that these individuals are limited to those with which we are acquainted in Russell’s highly restrictive sense. They include individuals we are perceiving, have perceived, or have been informed of and remember. So although Russell’s choice of example (Bismarck) would have to be changed to be made consistent with a much more liberal notion of acquaintance, I will use it to illustrate his distinction. Russell contrasts the situation of Bismarck himself, who ‘might have used the name [Bismarck] directly to designate [himself ] . . . to ma[k]e a judgment about himself ’ having himself as a constituent (1917: 209), with our situation in respect to him: when we make a statement about something known only by description, we often intend to make our statement, not in the form involving the description, but about the actual thing described. That is, when we say anything about Bismarck, we should like, if we could, to make the judgment which Bismarck alone can make, namely, the judgment of which he himself is a constituent. [But] in this we are necessarily defeated. . . . What enables us to communicate in spite of the varying descriptions we employ is that we know there is a true proposition concerning the actual Bismarck and that, however we may vary the description (as long as the description is correct), the proposition described is still the same. This proposition, which is described and is known to be true, is what interests us; but we are not acquainted with the proposition itself, and do not know it, though we know it is true. (Russell 1917: 210–11)

The proposition that ‘interests us’ is a singular proposition, but we cannot actually entertain it—we can know it only by description, that is, by entertaining a general (uniqueness) proposition which, if true, is made true by a fact involving Bismarck. But this general proposition does not itself involve Bismarck, and would be thinkable even if Bismarck never existed. The diVerence in type of proposition is clear from Russell’s observations about the statement ‘I met a man’. What do I really assert when I assert ‘I met a man’? Let us assume, for the moment, that my assertion is true, and that in fact I met Jones. It is clear that what I assert is not ‘I met Jones’. I may say ‘I met a man, but it was not Jones’; in that case, though I lie, I do not contradict myself, as I should do if when I say I met a man I really mean that I met Jones. It is clear also that the person to whom I am speaking can understand what I say, even if he is a foreigner and has never heard of Jones.

208

II. Referential–Attributive Distinction

But we may go further: not only Jones, but no actual man, enters into my statement. This becomes obvious when the statement is false, since there is no more reason why Jones should be supposed to enter into the statement than why anyone else should. Indeed, the statement would remain signiWcant, though it could not possibly be true, even if there were no man at all. (Russell 1918: 167–8)

The same points apply when a deWnite description is being used. Imagine that Russell met Whitcomb Judson, the man who invented the zipper. Then if Russell had said, ‘I met the inventor of the zipper, but it was not Whitcomb Judson’, he would have spoken falsely but would not have contradicted himself. And, although what Russell said in uttering ‘I met the inventor of the zipper’ would have been made true only in the event that he had met Judson, someone could have understood what Russell said without ever having heard of Judson. For no actual man, Whitcomb Judson or anyone else, entered into Russell’s statement. His statement would have been the same even if Sigismond Zipschitz had instead invented the zipper.

Point 8. One can describe (single out) an individual without actually referring to it. Kaplan suggests that one can use a description to refer to something even if one is not in a position to have a singular thought about it or, as he would say, even if one is not ‘en rapport’ with it. He asks rhetorically, ‘If pointing can be taken as a form of describing, why not take describing as a form of pointing?’ (1979: 392). I will explain why not. First consider the following example of Kaplan’s ‘liberality with respect to the introduction of directly referring terms by means of ‘‘dthat’’ ’, which ‘allow[s] an arbitrary deWnite description to give us the object’ (1989a: 560).23 (5)

Dthat [the Wrst child to be born in the twenty-second century] will be bald.

‘Dthat’ is a directly referential term and, as Kaplan explains in his ‘Afterthoughts’, ‘the content of the associated description is no part of the content of the dthat-term’ (1989b: 579); it is ‘oV the record (i.e., oV the content record)’ (1989b: 581). So ‘dthat’ is not merely a rigidiWer (like ‘actual’) but a device of direct reference.24 It is the actual object (if there is one) uniquely satisfying the description, and not the description itself (i.e. the property expressed by its matrix), that gets into the proposition.25 23 Since I am discussing Kaplan, I will here use his term ‘direct’ to modify ‘reference’, although, as observed earlier (n. 8), it is redundant. An expression, like a deWnite description, that merely denotes an object does not refer to that object, in the sense that the object is not a constituent of propositions expressed by sentences in which the expression occurs. 24 As Kaplan explains (1989b: 579–82), certain things he had previously said, and even his formal system (in Kaplan 1989a), could have suggested that ‘dthat’ is an operator on deWnite descriptions, with the content of the associated description included in the content of the whole phrase. This would make ‘dthat’ a ‘rigidiWer’ rather than a device of direct reference. 25 To stipulate that any phrase of the form ‘dthat [the F ]’ refers ‘directly’ does not, of course, mean that it is guaranteed a referent. It means only that the referent, if there is one, is a constituent of the singular proposition (if there is one) expressed by a sentence in which the phrase occurs.

Bach, Descriptions: Points of Reference

209

Not only does Kaplan’s ‘liberality’ impose no constraint on the deWnite description to which ‘dthat’ can be applied to yield a directly referring term, it imposes no epistemological constraint on what one can ‘directly refer’ to.26 However, even if we concede that any deWnite description can be turned into a directly referring term, so that a sentence containing the ‘dthat’ phrase expresses a singular proposition about the actual object (if there is one) that uniquely satisWes the description, it is far from obvious that the user of such a phrase can thereby refer to, and form singular thoughts about, that object. Kaplan seems to think this ability can be created with the stroke of a pen. Consider, for example, whether one can refer to the Wrst child born in the twentysecond century. Assume that nearly one hundred years from now, this description will be satisWed (uniquely). Then there is a singular proposition involving that individual, as expressed by (5).27 Without ‘dthat’ (and the brackets), (5) would express a general (uniqueness) proposition, the one expressed by (50 ), (50 )

The Wrst child to be born in the twenty-second century will be bald.

Now can one use the description ‘the Wrst child to be born in the twenty-second century’ referentially, to refer to that child? Kaplan thinks there is nothing to prevent this, that it is a perfectly good example of pointing by means of describing. However, what enables one to form an intention to refer to the individual who happens to satisfy that description? If one is prepared to utter (50 ) assertively, surely one is prepared to do so without regard to who the actual such child will be—one’s grounds are general, not singular. For example, one might believe that the Wrst child born in the twenty-second century is likely to be born in China and that Chinese children born around then will all be bald, thanks to China’s unrestrained use of nuclear power. But this only goes to show that one’s use of the description is likely to be taken to be attributive. Unless one were known to be a powerful clairvoyant, one could not plausibly be supposed to have singular grounds for making the statement. Nevertheless, Kaplan thinks that one could intend to use the description referentially anyway, as if putting the description in brackets and preceding it with ‘dthat’ could not only yield a term that refers to whoever actually will be the Wrst child born in the twenty-second century but could enable a speaker to refer to that child. However, it seems that one is in the same predicament as the one Russell thought anyone other than Bismarck would be in if he wanted to refer to Bismarck. 26 e.g. Kaplan writes, ‘Ignorance of the referent does not defeat the directly referential character of indexicals’, from which he infers, ‘a special form of knowledge of an object is neither required nor presupposed in order that a person may entertain as object of thought a singular proposition involving that object’ (1989a: 536). He goes on to explain that he is objecting to what he calls ‘Direct Acquaintance Theories of direct reference’. Here he seems to be conXating thinking of an object with referring to one, but he soon explains that he has ‘only denied the relevance of these epistemological claims to the semantics of direct reference’ (1989a: 537). 27 A singular proposition is not only object-involving but object-dependent, in that it would not exist if its object-constituent did not exist (at some time or other). So a singular proposition exists contingently. This does not imply that it exists only when its object-constituent exists. Singular propositions are not temporal just because they are contingent.

210

II. Referential–Attributive Distinction

Would it help to have the tacit modal intention of using the description rigidly, or even to insert the word ‘actual’ in the description?28 Referring to something involves expressing a singular proposition about it, but rigidifying the description or including the word ‘actual’ would not make its use referential. Even though the only individual whose properties are relevant to the truth or falsity of the proposition being expressed (even if that proposition is modal) is the actual F (if it exists), still that proposition is general, not singular. This proposition may in some sense be object-dependent, but it is not object-involving. The property of being the actual may enter into the proposition, but the actual F does not. The upshot of these observations is that one can use a description to describe or, as I will say, single out something without actually referring to it. The fact that there is something that satisWes a certain deWnite description does not mean that one can refer to it. If a diVerent individual satisWed the description or you are describing a hypothetical situation in which that would be the case, you would have singled out that individual instead (see Point 9). Nevertheless, you can use the description just as though you were introducing the thing that satisWes it into the discourse. You could, for example, use pronouns to ‘refer’ back to it. You could say, ‘The Wrst child to be born in the twenty-second century will be bald. It will be too poor to use Rogaine.’ Giving it a name will not help. You could dub this child ‘Newman 2’ (just as Kaplan dubbed the Wrst child to be born in the twenty-Wrst century ‘Newman 1’), but this would not enable you to refer to it or to entertain singular propositions involving it. In this, as Russell might have said, ‘we are necessarily defeated’. Even though (50 ) does not express a singular proposition, the proposition which ‘interests us’ but which we cannot entertain, one can still use the sentence to describe that proposition. It might be objected that in calling this ‘singling out’ rather than ‘referring’ to an object, I am not making a substantive claim but am merely engaging in terminological legislation. I would reply that anyone who insists on calling this ‘reference’ should either show that a singular proposition is expressed or explain why expressing a general, object-independent proposition should be thought to involve reference. One possible reason is taxonomic: if we are to maintain that indexicals and demonstratives are inherently referring expressions and not merely expressions that are often used to refer, allowances must be made for the fact that we sometimes use them to do what I would describe as merely singling out an object, that is, an object that the speaker is not in a position to have singular thoughts about. For example, one could use ‘he’ or ‘that child’ to single out the Wrst child born in the twenty-second century. But the question is whether this counts as genuine reference. Indeed, one can use such expressions without even singling out an individual, as in, ‘If a child eats a radioactive Mars bar, he/that 28 This is related to Hintikka’s suggestion, as quoted by Kaplan (1979: 384), that the referential–attributive distinction ‘is only operative in contexts governed by propositional attitudes or other modal terms’. Evidently, Hintikka’s idea was that if there is no modal or other operator whose scope interacts with that of the description, the distinction between what actually satisWes the description and any possible satisWer of it does not come into play.

Bach, Descriptions: Points of Reference

211

child will be bald’. The mere fact that philosophers are in the habit of calling indexicals and demonstratives ‘referring expressions’does not justify doing this.

Point 9. Descriptive ‘reference’, or singling out, is not genuine reference. In summing up his account of the referential–attributive distinction, Donnellan concedes that there is a kind of reference, reference in a ‘very weak sense’, associated with the attributive use of a deWnite description (1966: 304). Since he is contrasting that use with the referential use, this is something of a token concession. Reference in this very weak sense is too weak to count as genuine reference, for one is ‘referring’ to whatever happens to satisfy the description, and one would be ‘referring’ to something else were it to have satisWed the description instead. This is clear in modal contexts, such as in (6): (6)

The next president, though probably a man, could be a woman.

The speaker is not asserting of some one possible president that he or she will probably be a man but could be a woman, say if he had a sex-change operation before her inauguration. In this context, the description is understood to fall within the scope of ‘could’. The speaker is allowing for diVerent possible presidents, some male, some female, only one of whom will actually be the next president. Surely this is not reference, not even in a very weak sense. Even so, we might wonder whether there is a weak sort of reference that still counts as genuine reference. Let us call this ‘reference under a description’ or simply ‘descriptive reference’. Can one use a description to refer to an individual without having some independent, nondescriptional way of thinking of and intending to refer to a certain object? That is, can one intend to convey a singular proposition about a certain individual that one is not in a position to have singular thoughts about? As we saw in Point 8, it will not help to use a rigidiWed description, say of the form ‘the actual F ’. This guarantees that no object other than the actual F could be the one on whose properties the truth or falsity of one’s statement depends, but it does not yield reference. One may be singling out the object that satisWes the description, hence describing the relevant singular proposition about that object, but one is not referring to that object or expressing that proposition. Here is a possible but dubious route toward counting descriptive reference as genuine reference. Philosophers have often referred to singular thoughts as de re thoughts or attitudes. However, they have also used the term ‘de re’ for a type of attitude ascription (contrasted with ‘de dicto’ ascriptions). Moreover, de re attitude ascriptions can be true even when no de re (i.e. singular) attitude is being ascribed. So, for example, suppose that even though there is no suspect in the case, the mayor has claimed that Smith’s murderer is insane. Smith’s murderer, on hearing this, could say, ‘The mayor thinks that I am insane’, without imputing to the mayor any de re attitudes about him. It is easy to overlook the fact that not all de re attitude ascriptions are ascriptions of de re attitudes because of the structural ambiguity of the phrase ‘de re attitude ascription’, in

212

II. Referential–Attributive Distinction

which ‘de re’ can modify either ‘attitude’ or ‘ascription’. Just as there is a diVerence between ‘Tibetan history-teacher’ and ‘Tibetan-history teacher’, so there is a diVerence between ‘de re-attitude ascription’ and ‘de re attitude-ascription’.29 A de re attitudeascription can refer to a certain object without implying that there is a certain object the ascribee has an attitude about. The point carries over to indirect quotation. Having read that the police have said, ‘Smith’s murderer is insane’, Smith’s murderer could say to a conWdant, ‘The police say that I am insane’. In reporting this, Smith’s murderer is referring to himself. It may seem that he is reporting that the police are referring to him too (with their use of ‘Smith’s murderer’), but they are not. It is sometimes suggested that giving a ‘descriptive name’ (Evans 1982: 31) to an individual whose identity is in question enables one to refer to that individual. Suppose that Smith’s murderer left some vitriolic hip-hop lyrics under a rock at the scene of the crime and the media came to call him ‘Rock the Rapper’. Would using this name put people in a better position to refer to Smith’s murderer? It would not, not even if sentences containing that name expressed singular propositions. Using such sentences would not enable one to have singular thoughts about Smith’s murderer and thereby be in a position to refer to him. Again, propositions cannot be put within one’s grasp with just the stroke of a pen.30

Point 10. Various kinds of pseudo-reference can be confused for the real thing. We have already seen several examples of this. Point 6 denied that speciWc uses of indeWnite descriptions are referential. And Points 8 and 9 denied that singling out (descriptively ‘referring’ to) something counts as genuine reference. Singling out whatever uniquely satisWes a certain deWnite description is not to refer to it. Here I will discuss discourse reference and brieXy comment on Wctional reference and failed reference, recognizing that each deserves much more extensive discussion. Reference and Discourse ‘Referents’ It is well known that unbound pronouns, as well as deWnite descriptions, can be used anaphorically on indeWnite descriptions, as in these examples: 29 Some would deny the diVerence. In Searle’s view e.g. the very idea of de re belief is an illusion ‘arising from a confusion between reports of beliefs and features of the beliefs being reported’ (1979b: 157). But Searle maintains that belief about an object is always under a description, hence not really singular. He does not recognize the category of singular thoughts and the possibility of de re ways of thinking of things. 30 It has even been suggested that giving a (descriptive) name to something whose existence is in question enables us to refer to it if it does exist. Suppose it is speculated that the perturbations in the orbit of Pluto are caused by a small but very dense planet beyond Pluto whose entire surface is very dark, making this planet invisible even to the largest telescope. Nobody knows whether there is such a planet, but astronomers take this speculation seriously and decide to call this planet ‘Goofy’. Accordingly, they believe that Goofy, if it exists, aVects the orbit of Pluto. Even so, it does not follow that they can have de re (singular) thoughts about it and use the name ‘Goofy’ to refer to it. Of course, they can single it out and describe singular propositions about it, provided it exists. Otherwise, they are just going through the motions.

Bach, Descriptions: Points of Reference (7) (8) (9) (10)

213

Mary met a man today. He/The man was bald. A girl went to yesterday’s Giants game, and she/the girl caught a foul ball. If there were a unicorn there, Macdonald would have seen it/the unicorn. Every farmer owns a donkey. He rides it on Sundays.

In (7) and (8), the pronoun (and the deWnite description) can be used to refer. If the speaker is using ‘a man’ speciWcally in uttering (7), he could use ‘he’ (or ‘the man’) to refer to the man he thinks Mary met that day. If, on the other hand, the speaker is not in a position to refer to such a man, he could only use ‘a man’ nonspeciWcally and could not use ‘he’ (or ‘the man’) referringly; the most he could intend to convey is the general proposition that Mary met a man that day who was bald. Similarly, if the speaker of (8) is using ‘a girl’ speciWcally, he could use the pronoun ‘she’ (or ‘the girl’) to refer to the girl he thought went to the game, but not if he were using ‘a girl’ nonspeciWcally. However, it seems that the pronouns (and the deWnite descriptions) in (9) and (10) cannot be used to refer at all. In (9) ‘it’ is not being used to refer to an unspeciWed (and presumably nonexistent) unicorn, and in (10) ‘it’ functions quantiWcationally, ranging over the diVerent donkeys owned by the diVerent farmers. Despite the fact that the pronouns and the deWnite descriptions in cases like (7) and (8) need not, and in cases like (9) and (10) cannot, be used to refer, many semanticists have attributed ‘discourse referents’ to them. I am not suggesting that they seriously believe that discourse referents are real referents, but this only makes it puzzling why they use this locution. Here is how Karttunen introduced the phrase: Let us say that the appearance of an indeWnite noun-phrase establishes a discourse referent just in case it justiWes the occurrence of a coreferential pronoun or a deWnite noun-phrase later in the text. . . . We maintain that the problem of coreference within a discourse is a linguistic problem and can be studied independently of any general theory of extra-linguistic reference. (Karttunen 1976: 366; my emphasis)

He explains further, In simple sentences that do not contain certain quantiWer-like expressions,31 an indeWnite NP establishes a discourse referent just in case the sentence is an aYrmative assertion. By ‘establishes a discourse referent’ we meant that there may be a coreferential pronoun or deWnite noun-phrase later in the discourse.32 IndeWnite NPs in Yes–No questions and commands do not establish referents. (Ibid. 383)

So the ‘coreferential pronoun or deWnite noun-phrase later in the discourse’ can, thanks to the discourse referent ‘established’ by the indeWnite NP, have a discourse referent even if, as in our examples, it is not used to refer. 31 Karttunen is excluding negative quantiWcational phrases like ‘no donkey’ and negative cases like ‘Bill didn’t see a unicorn. It/The unicorn had a gold mane’. 32 In other words, each link in an ‘anaphoric chain’ (Chastain 1975) is treated as having a discourse referent, even if intuitively it does not refer. It should not be supposed, as Chastain (1975) and many others have, that when the links in the chain, the expressions anaphoric on the indeWnite description, are used to refer, the indeWnite description itself refers. In most such cases the indeWnite will be used speciWcally, merely to allude to something.

214

II. Referential–Attributive Distinction

The notion of discourse referent has inspired a great deal of theorizing in semantics, including discourse representation theory (DRT), dynamic semantics, and the study of reference accessibility. However, as is implicitly conceded by Karttunen’s deWnition (‘[it] can be studied independently of any general theory of extra-linguistic reference’), discourse reference is no more reference than our relation to our perceptual experiences (as opposed to objects perceived) is perception. The basic problem is simply this: a chain of ‘reference’ is not a chain of reference unless it is anchored in an actual (‘extralinguistic’) referent. Despite the widespread use of the phrase ‘discourse referent’ in some semantic circles, so-called discourse referents are not literally referents. The pronouns in the examples like those we have just considered are not used referringly and, as King (1987) suggested, must be viewed quantiWcationally. They are used as surrogates for deWnite descriptions, descriptions which if present in place of the pronouns would not be referential. These pronouns are what Neale (1990: ch. 5) calls ‘D-type pronouns’. The basic idea is that the pronoun is used elliptically for a deWnite description recoverable from the matrix of the antecedent indeWnite description (Bach 1987/1994: 258– 61). Neale develops a detailed account of how D-type pronouns work in a wide variety of cases. It is essential to this account that the descriptions implicit in the use of D-type pronouns are not construed as referential, even when they are used referentially. To see the signiWcance of looking at such pronouns in this way, let us return to two of our examples. Consider (8) again: (8)

A girl went to yesterday’s Giants game, and she caught a foul ball.

Suppose that the speaker merely heard that a girl had caught a foul ball at yesterday’s game. Then he does not have any particular girl in mind and is not using ‘a girl’ speciWcally. Even so, it might be thought that ‘she’ refers to a certain unspeciWed girl (the ‘discourse referent’) who went to the Giants game the day before. But how could this be? The Wrst clause of (8) expresses a general proposition, but what about the second clause? Does it express a singular proposition about a certain girl who went to yesterday’s Giants game? Suppose this clause is not true, hence that no girl who went to yesterday’s Giants game caught a foul ball. Then which girl would the second clause of (8) be about, and what singular proposition would it express? Answer: no girl, and no proposition. On the view that ‘she’ is referential in the second sentence of (8), that sentence would express a singular proposition if and only if it is true! Surely which proposition a sentence expresses, or that it expresses any, cannot depend on whether or not it is true. The situation is similar with a quantiWed sentence, as in (10), (10)

Every farmer owns a donkey. He rides it on Sundays.

Suppose there are farmers who own more than one donkey. In that case, what does it take for the second sentence in (10) to be true? Its truth does not require every farmer to ride on Sundays every donkey that he owns, but nor does it require merely that every

Bach, Descriptions: Points of Reference

215

farmer ride on Sundays one donkey that he owns. So it is not clear what it would take for the second sentence in (10) to be true when there are farmers who own more than one donkey. However, DRT and dynamic semantics have been motivated by the thought that this is clear, and they have tried to make sense of sentences like the second one in (10) in terms of the notion of discourse referent. Fictional Reference This is far too big a topic to take up in any detail here. Any serious discussion has to distinguish Wctional reference (reference in a Wction) and reference (outside the Wction) to Wctional entities. Reference in a Wction does not count as genuine reference, at least if it is to Wctional persons, places, and things (in a Wction, there can be genuine reference to real persons, places, and things). If Salmon (1998) is right in claiming that Wctional entities are real, albeit abstract entities, then we can genuinely refer to them. Otherwise, we can only pretend to. In my view, both Wctional reference and reference to Wctional entities involve special sorts of speech acts, but there is nothing special about Wctional language itself (Bach 1987/1994: 214–18). Failed Reference A speaker can intend to refer but fail in the (communicative) sense that his listener does not identify the individual he intends her to identify. This is of no special interest here. More interesting is the case in which the speaker intends to refer to a certain individual but there is no such individual. This does not aVect any claim being made here, so long as it is understood that in this case there is no singular proposition about that individual and no such proposition for the speaker to convey. In such a case the speaker can use an expression to refer but not successfully, since there is no individual being referred to. In such a situation, even if the speaker sincerely utters ‘a is F ’ and means it literally, there is nothing he believes to be the F. Similarly, if he sincerely utters ‘the F is G’ and is using ‘the F ’ referentially but he fails to have anything in mind, there is nothing he is referring to. In this case, his referential intention cannot be fulWlled, and full communication cannot be achieved. There is nothing for the hearer to identify, and no singular proposition for her to entertain. The best the hearer can do is recognize that the speaker intends to convey a singular proposition of a certain sort. The speaker has the right sort of intention, to be speaking of some particular thing, but there is no thing for him to succeed in referring to. A diVerent situation would arise if the speaker merely made as if to refer to something, perhaps to deceive the hearer or perhaps to play along with the hearer’s mistaken belief in the existence of something. In this case, although the speaker does not intend to refer to something, he does intend to be taken to be. He can succeed in communicating if he is taken to be referring to the individual the hearer mistakenly believes in. But since there is no such thing, there is no singular proposition to be grasped.

216

II. Referential–Attributive Distinction

4. The Pragmatic Status of the Referential–Attributive Distinction We have seen that descriptions can be used to refer not in spite of but partly because they are devices of quantiWcation (of a certain sort) and that there are restrictions on what counts as using them to refer. Various uses of descriptions that may seem to pass as referential fall short of that. Now I will apply certain of the observations made so far in order to clarify the referential–attributive distinction and bolster the case for the pragmatic status of referential uses. In what follows, we should keep in mind Point 3, that frequently one uses a description to refer because no other sort of expression is suitable to refer to what one wishes to talk about. And, as Points 4 and 5 suggested, in using a description to refer one expects the hearer to take one to be relying on information derived from and speciWc to the particular object that in fact satisWes the description. On the other hand, in using a deWnite description attributively, one does not purport to have (or at least to be relying on) any information speciWc to the actual satisWer of the description. One is in the same epistemic position that one would be in if something else uniquely satisWed the description.

Point 11. Various contrasts may seem to capture the referential–attributive distinction, but only one really does. When he introduced the referential–attributive distinction, Donnellan (1966) tried to explicate it in various ways, in each case drawing an apt contrast that intuitively seems to mark the diVerence. These and a couple of other contrasts are worth reviewing. As we will see, suggestive as these various contrasts are, only one does the job. Being What the Speaker has in Mind versus Being Whatever Fits the Description Donnellan suggests that in using a description referentially, your statement is about whatever or whoever you have in mind, whereas with an attributive use your statement is (weakly) about whatever or whoever satisWes the description. This test seems plausible, but it is not as reliable as it sounds.33 First, it suggests that the speaker cannot know who satisWes the description, when of course he can, as Donnellan himself points out. So, for example, you could utter ‘Schmidt’s editor is insane’ and be using ‘Schmidt’s editor’ attributively even though you could know who Schmidt’s editor is. For you might not intend your statement to depend on this knowledge, say if you made it on the basis of the many strange scrawls on Schmidt’s manuscript. Then what you are stating would not be aVected if it turns out that you are mistaken about who Schmidt’s editor is. 33 The diVerence may seem to be captured by the contrasting eVects of inserting ‘whoever he is’ and a ‘namely’ phrase after the description, as in ‘Schmidt’s editor is insane’: (i) Schmidt’s editor, whoever he is, is insane; (ii) Schmidt’s editor, namely Ross, is insane. However, as the following considerations in the text suggest, this test is far from airtight.

Bach, Descriptions: Points of Reference

217

Donnellan sometimes describes the relevant contrast as between believing something about ‘someone in particular’ and about ‘someone or other’. However, as he notes, it is possible to think de re of the satisWer of a description even if one is using the description attributively. For example, you might be personally acquainted with Andrew Wiles and yet, in uttering ‘The man who proved Fermat’s Last Theorem must be a genius’, use the description attributively. It is irrelevant to what you mean that you believe Andrew Wiles to be the man who proved Fermat’s Last Theorem or even that he actually did. Also, it is possible to use a description referentially without in a certain sense knowing who satisWes it. In fact, we can distinguish two sorts of knowing who, only one of which is relevant here. There is the context- or interest-relative sort of knowing who, whose slipperiness has been well documented by Boe¨r and Lycan (1986). Not knowing who someone is in this way is compatible with knowing who they are for the purpose of referential identiWcation. You can use ‘the masked man over there’ to refer to a certain masked man, and your listener can know who you are referring to, without either of you knowing who (of the other sort) that man is. Suppose that neither of you could recognize him without his mask or know any distinguishing characteristics of his. Even so, you can refer to him then and there and your listener can know who you are referring to. Although neither of you have any way of identifying him beyond the circumstances of utterance, you both can identify him for purposes of reference. Description Inessential versus Description Essential Donnellan suggests that a description used referentially is inessential to what is said, that is, that the referent’s satisfying the description is not necessary for the truth of what is said. Presumably he suggests this because there is a particular individual in question, and because which one it is is Wxed independently of the property expressed by the matrix of the description. I would agree that the description (i.e. this property) is inessential to what is asserted, but I would insist that it is essential to what is said. In the referential case there is something that the speaker intends to refer to even if it does not in fact satisfy the description he is using (or, if the description is incomplete, some intended completion of it) and even if some other individual does. But this does not show that the description is inessential to what the speaker says. It shows only that the speaker, in uttering something of the form ‘the F is G’, could have ascribed the same property (being G ) to the same object (the one he takes to be the F ) even if that object did not satisfy the description. The description is essential to what he says, because clearly he would have said something diVerent if he had used a diVerent, nonsynonymous description. For example, if he utters, ‘The man drinking a martini is inane’, and the man he is referring to is holding a martini but not drinking it, then what he is saying is true only if some other man is the one drinking a martini and he is inane. If advised that the man he intended to refer to has not touched a drop, he might even revise his utterance and say, ‘The man holding a martini is inane’.

218

II. Referential–Attributive Distinction

The situation is diVerent if the speaker uses a description he knows not to be true of the intended referent but believes that the hearer thinks it is. For example, suppose you say ‘The man holding a martini is insane’ when you believe that his glass is Wlled with tequila but assume your audience is unaware of this. In this case, in which you are exploiting your audience’s ignorance, you do not mean what you say. However, you intend your audience, in order to identify the referent, to take you to mean what you say. Speaker’s Referent versus Semantic Referent Kripke (1977) invokes this contrast in the course of arguing against the semantic signiWcance of the referential–attributive distinction. If a description is being used referentially, the speaker’s referent may be distinct from the semantic referent (what satisWes the description), whereas with an attributive use they cannot be distinct, since the semantics of the description determines the relevant individual. This may seem right, but Kripke’s distinction is both misleading and inadequate. It is misleading because deWnite descriptions, being quantiWcational phrases, do not have semantic referents at all, at least not in the way in which genuine singular terms do (see n. 8). In Russell’s jargon, they merely denote (where denoting is a semantically inert property, so that what is denoted does not enter into the proposition expressed). More importantly, as we saw above, deWnite descriptions can be used attributively even if they are used in a nonliteral way. When so used, the speaker’s ‘referent’ is not the semantic referent. Primary Aspect versus Secondary Aspect Searle (1979b) suggests that the diVerence between the referential and the attributive use boils down to whether some description other than the one actually used is primarily relevant to the speaker’s utterance. If some other one is primary, then the one actually used expresses merely a ‘secondary’ aspect under which the speaker is referring. In that case, he has some ‘fallback’ description in mind, under which the hearer is to think of the referent. There are two problems here. First, Searle assumes that one can think of an object only under a description and not in some de re way. He does not seriously consider the possibility that attitudes can be de re. Rather, he insists that only attitude reports can be de re and thinks that the view that attitudes themselves can be de re is based on a confusion. Perhaps he thinks this confusion arises from the structural ambiguity of the phrase ‘de re belief report’ (see Point 9), but he fails to show that there are no de re attitudes (singular thoughts). The second problem is that, even if one uses a certain description with a fallback description in mind, one that expresses the primary aspect, one could still be using the former description attributively. One might use ‘the F ’, believing that the F is the H and being prepared to fall back on ‘the H ’. However, this does not mean that if one had used ‘the H ’ instead, one would have used it referentially. For example, looking at the

Bach, Descriptions: Points of Reference

219

way a certain car is parked you might utter, ‘The owner of this car must have been drunk’. You have no idea who the owner is but you have assumed that the owner of that car was the driver who parked it. Although you are prepared to fall back on the description ‘the driver who parked this car here’, surely you were not using the original description referentially. ‘Right’ Thing versus No ‘Right’ Thing Donnellan suggests that ‘in the referential use as opposed to the attributive, there is a right thing to be picked out by the audience and its being the right thing is not simply a function of its Wtting the description’ (1966: 304). Obviously the description could not be merely a device for getting one’s audience to pick out the referent unless there were a right thing for them to pick out, something which one wishes to speak about independently of how one is referring to it. That is indisputable, assuming that there is something the speaker intends to refer to, for obviously the reference can succeed only if there is some such thing and only if the hearer identiWes it. But there is a sense in which an attributive use also involves a ‘right’ thing, namely the thing that satisWes the description, though of course this is not what Donnellan means. The point of his contrast here is really that the speaker has some independent way of thinking of, and of intending the hearer to think of, the referent, as something other than the (unique) satisWer of the description being used. This seems right so far as it goes but, as it stands, the contrast between there being a ‘right’ thing and there not being one boils down to the diVerence between using and not using the description referentially. Particularity versus Generality This contrast gets to the heart of the referential–attributive distinction. Taking Russell’s theory of descriptions to apply only to attributive uses, Donnellan observes that with them ‘we introduce an element of generality which ought to be absent if what we are doing is referring to some particular thing’. But ‘this lack of particularity is absent from the referential use of deWnite descriptions precisely because the description is here merely a device for getting one’s audience to pick out or think of the thing to be spoken about, a device which may serve its function even if the description is incorrect’ (1966: 304). This seems right. After all, in using a description referentially one is using it to refer to something, and that involves conveying a singular proposition. In using a descriptive attributively, one is not referring—one is conveying a general proposition. Not surprisingly, the referential–attributive distinction is just an instance of the distinction between using a noun-phrase to refer and using it quantiWcationally. The essential diVerence between a referential and an attributive use of a deWnite description is that with a referential use one is expressing a singular proposition about a certain object (unless there is no actual referent), whereas with an attributive use one is expressing a general proposition. This is old news, but it is important not to confuse this with any of the other ways we have considered of trying to capture the referential– attributive distinction.

220

II. Referential–Attributive Distinction

To sum up. With a referential use, one must be thinking of the F in some de re way, hence in some other way than as the F. In using ‘the F ’ referentially, one is, in eVect, instructing the hearer to think of the F, and that requires him also to think of the F in some other (de re) way than as the F. With the attributive use, one need not be in a position to think of the F in any de re way, and even if one does, it is irrelevant. One is simply saying something about whatever is the F, which one need not be in a position to think of in any way at all (or at least in any other way than as the F, in so far as that counts as thinking of it at all). In the referential case, it is this other way of thinking of the F that matters. That way of thinking determines the object being referred to, the object that is a constituent of the singular proposition the speaker is trying to convey. Notice that most referential uses involve incomplete deWnite descriptions, descriptions that are insuYciently restrictive to identify a unique individual. The property expressed by the matrix of the description serves merely to distinguish one’s intended referent from anything else that, under the circumstances, one might be talking about. So if you say, ‘The table is covered with books’, using ‘the table’ is enough to distinguish the one table in the vicinity (or within the attention of you and your audience) from other objects in your presence. If there were more than one table there and it did not stand out in some blatantly obvious way, using ‘the table’ would not enable your audience to Wgure out which individual you are talking about and would be inappropriate—you would have to point to the relevant table and use ‘that table’.

Point 12. Incomplete deWnite descriptions pose no real threat to Russell’s theory. Some singular deWnite descriptions, either by necessity or as a matter of fact, are uniquely satisWed. Most of the ones that we ever have occasion to use are not. Even so, they imply uniqueness. As mentioned earlier, this uniqueness is not encoded in the meaning of ‘the’—‘the’ indicates totality, as in plural deWnite descriptions. The implication of uniqueness is a side eVect of combining ‘the’ with a nominal ‘F ’ in the singular, to yield ‘the F ’. So what happens when, as is typical, we use a description that is incomplete and we believe it to be? Even if we are using ‘the’ literally and ‘F ’ literally, we do not mean that there is exactly one F. Rather, we are in eVect making a pretense of uniqueness, at least for current conversational purposes. This is a perfectly ordinary phenomenon and no diVerent in kind from various other simplifying pretenses we implicitly make in everyday conversation. For example, we often pretend that the immediate environment or situation is all that is relevant to statements we make. We utter sentences like ‘It’s not raining now’, ‘I guess nobody’s home’, and ‘I haven’t eaten’ as if they expressed much more restricted propositions than the much more farreaching propositions they literally express. Similarly, we use incomplete deWnite descriptions, like ‘the book’ and ‘the table’, as if there is no other book and no other table than the ones we are talking about. So if, in order to tell someone where to Wnd a certain book, you say ‘The book is on the table’, for the immediate purpose of the conversation the book and the table in question are the only ones that come into consideration.

Bach, Descriptions: Points of Reference

221

The incompleteness of ordinary deWnite descriptions poses no real problem for Russell’s theory.34 That a description ‘the F ’ is incomplete does not aVect the proposition expressed by any sentence in which it occurs and does not aVect what a speaker says in uttering such a sentence; it can only aVect what the speaker means. Russell’s theory does not have to treat uses of incomplete descriptions as either elliptical for complete ones or as containing implicit but semantically relevant restrictions on their domains. In particular, if a speaker uses an incomplete description referentially, he is attempting to convey a singular proposition involving an individual that is F (assuming he is using ‘F ’ literally). It does not matter that the description is incomplete, because the speaker does not intend the description by itself to provide the hearer with the full basis for identifying the referent. But Russell’s theory, being silent about referential uses anyway, does not say that it does. Nor must Russell’s theory say that ‘the F ’ is somehow elliptical for ‘the salient F ’, ‘the aforementioned F ’, ‘the F under discussion’, ‘the relevant F ’, ‘the F I have in mind’, or anything of the sort, just because it is used to refer to some such F. The attributive case may seem to pose more of a problem for Russell’s theory. If a speaker is using an incomplete deWnite description attributively, he intends to convey a uniqueness proposition, just as Russell’s theory would predict, but this is not the uniqueness proposition expressed by the sentence he is uttering; it is the one that would be expressed if the description were made complete by being expanded in a suitable way. One apparent problem is that diVerent verbal means of doing this may yield the very same proposition. However, this is not a genuine problem, since what matters is which proposition the speaker intends to convey, not which words he would use to convey it (were he to do so in a fully explicit way). A more serious objection, made by Kripke (1977) and by Wettstein (1981), is that there are cases in which the speaker does not have a determinate uniqueness proposition in mind but nevertheless communicates successfully. For this reason, and despite treating the referential–attributive distinction pragmatically, Kripke questions whether a Russellian approach can accommodate the case of incomplete descriptions. Whereas Grice supposed that all this approach needs in order to handle a description like ‘the table’ is ‘suitable provision for unexpressed restrictions’, such as ‘the table in this room’ (1969: 142), Kripke ‘doubt[s] that such descriptions can always be regarded as elliptical with some uniquely specifying conditions added’ (1977: 255), and Wnds it ‘somewhat tempting to assimilate such descriptions to the corresponding demonstratives (for example, ‘‘that table’’)’ (1977: 271). However, this is a temptation to be resisted. Even if deWnite descriptions can sometimes be used like demonstrative descriptions, they are semantically and syntactically diVerent. And obviously Grice’s example of an incomplete description used referentially is not a demonstrative just because it contains one. But the main point is that Russell’s theory does not (falsely) 34 For more thorough defense of this point see Bach (1987/1994: 103–8, 124–6) and Neale (1990: 93–102; this volume).

222

II. Referential–Attributive Distinction

entail that if an incomplete description is used referentially, there must be some intended restriction on the description. What matters is the speaker’s intention to use the description to refer to a certain particular individual, in the way discussed above, to some salient or contextually relevant F. Whereas Kripke’s worry concerns referential uses of incomplete deWnite descriptions, Wettstein’s concerns the attributive. The worry is that Russell’s theory predicts that if no determinate restriction on the description is intended, the speaker’s utterance has no determinate content. But it may well have determinate content, although this content is not, according to Wettstein, of the sort required by Russell’s theory.35 Now Wettstein’s examples all involve complex descriptions containing expressions used referentially: ‘his murderer’, ‘the murderer of Smith’, and ‘the murderer of the man on the couch’. But such examples show merely that a deWnite description can be relativized, and this, as Neale has pointed out (1990: 99–100), is perfectly compatible with Russell’s theory. It does not apply only to ‘purely qualitative’ descriptions but is equally applicable to descriptions containing referential elements or even variables. Finally, there is a common objection, registered for example by Devitt (this volume, n. 36), that it is ‘implausible’ that sentences containing incomplete deWnite descriptions do not literally express truths. This is not accurate, because for any sentence with a truth-value, either it or its negation is true (besides, Devitt should not appeal, as he sometimes does, to ‘intuitions about truth conditions’ without saying truth conditions of what—what speakers mean have truth conditions too). The point, presumably, is that the proposition, true or false, expressed by the sentence (according to Russell’s theory) is not the one that intuitively is expressed. The Russellian should reply by saying that such sentences, because their propositional contents (or truth conditions) are what Russell’s theory says they are, are typically not used in a strictly literal way (even when all their constituent words are being used literally). This view may initially seem counter-intuitive, but that is only because of a misunderstanding of the kind of nonliterality at work here.36 In any case, there is a psychological reply to this common objection. Yes, it seems problematic for Russell’s theory to predict that a sentence like ‘The table is covered with books’ is false when the sentence can straightforwardly be used to make a true 35 In some cases of attributively used incomplete descriptions, there may be no intended expansion of the description at all. But what these cases illustrate is not a problem for Russell’s theory but a common problem in communication. When it is indeterminate what the speaker’s intention is, communication cannot really succeed, since nothing counts, strictly speaking, as understanding the speaker correctly. Practically speaking, though, this may not matter. In the case of attributively using a description without a determinate restriction, there is no problem as long as one and the same individual satisWes all the plausible candidates for the completion of the description. Then it makes no practical diVerence precisely which uniqueness proposition the speaker wishes to convey or which one the hearer takes him to convey. 36 As I have previously argued, seemingly semantic intuitions confuse what speakers mean with the semantic contents of expressions (Bach 2002a), and, in particular, there is a kind of nonliterality that is compatible with each of the words in a sentence being used literally (Bach 2001b). Sentence nonliterality, as I call it, is not like metaphor, metonymy, and other Wgurative uses of words. Even though we may not intuitively think of this phenomenon as nonliterality, because no speciWc words are being used Wguratively, it is a way of not being literal. For the speaker says one thing but intends to convey something distinct from that.

Bach, Descriptions: Points of Reference

223

statement—intuitively, one has no sense that there is any false proposition in the air. But there is an explanation for this: as soon as one hears (or reads) ‘the table’, because of its obvious incompleteness one immediately takes it either to be used referentially or with an implicit completion. One never actually computes the proposition expressed by the sentence in which it occurs. Moreover, it is a misconception to suppose, as for example Stanley and Szabo´ do (2000a: 235–6), that the literal falsity of a sentence like ‘The table is covered with books’ has anything to do with how it manages to be used to make a true statement. This is not a case of Gricean quality implicature. It is not like an utterance of the obviously false ‘Ken Lay cried all the way to the bank’, used to convey how happy Lay was with his ill-gotten gains. What triggers the hearer’s search for something other than what is said is not its obvious falsity but its lack of relevant speciWcity. To sum up, there are two basic cases. In the referential case there is an implicit narrowing of things under consideration or relevant to the conversation, and ‘the F ’ is used to refer to the only F among them. In the attributive case there is an implicit, contextually relevant narrowing of the property expressed by ‘F ’; ‘the F ’ is used to single out whichever individual uniquely has that narrowed property. In both cases the speaker intends the hearer to take such a narrowing into account in Wguring out what he means. If no such narrowing is evident to the hearer, she will not Wgure out what the speaker means. If she recognizes the use as referential, she will wonder which F the speaker is talking about; if she recognizes the use as attributive, she will wonder what unique case of an F is in question.

Point 13. DeWnite descriptions do not have referential meanings. Various reasons have been given for supposing that descriptions have referential readings, but they are commonly, though not universally, thought to have been adequately rebutted. Michael Devitt (this volume) revisits several of these reasons and oVers some new ones. He is perfectly aware of the distinction between what a speaker says and what a speaker means and, in particular, of that between linguistic reference and speaker’s reference. Even so, he is not convinced that special conventions or referential meanings are not needed to enable descriptions to be used referentially. Why am I convinced? Not because such conventions or meanings would not help explain how or why a given use of a description is taken referentially rather than attributively. After all, no ambiguity thesis has to explain how hearers manage to disambiguate the allegedly ambiguous expression. Still, it is important to appreciate what goes on in the case of deWnite descriptions. Whether or not there are referential conventions or meanings for them, how does a hearer Wgure out which way, referentially or attributively, a description is being used? Compare situations in which the following sentences might be uttered: (11)

The man drinking a martini is in the corner nearest the bar.

224

II. Referential–Attributive Distinction (12)

The man in the corner nearest the bar is drinking a martini.

Regarding the Wrst situation, imagine that you are talking by cell phone to a fellow spy who is at a party and you are helping him Wnd an informant at the party, whose identity neither of you knows. You and he mutually know that the informant is there and will be drinking a martini. You have been advised where the informant is supposed to be, and you pass on this information by uttering (11). You are using the subject description attributively, merely enabling your cohort to identify the individual who satisWes it (i.e. its obvious completion). In the second situation, you are both at the party, you see a man in the corner nearest the bar and, since you and your cohort mutually know that the informant is at the party and is to be drinking a martini, you utter (12). In this case, you are using the subject description referentially. You are asserting a singular proposition about the man in the corner nearest the bar, and are enabling your cohort to identify the individual you have in mind by identifying him as the man in the corner nearest the bar. Notice, however, that this identiWcation involves the quantiWcational meaning of the description. So if deWnite descriptions are ambiguous, their ambiguity is most extraordinary: one of their senses (the quantiWcational) plays a role when the other (the referential) is operative. An alternative account was presented in section 1: the distinctive quantiWcational character of deWnite descriptions, together with their implication of uniqueness, helps explain how, without any special stage setting, they can readily be used to refer (Point 5), in part because there may be good reason to suppose that the speaker is implicitly conveying an identity between something he is thinking of and what satisWes the description (Point 4). In eVect, he is instructing the hearer to think of and take him to be talking about whichever individual is the F. Besides, even though descriptions are devices of quantiWcation, they are often the only sort of singular nounphrase available to use to refer to what one wishes to refer to (Point 3). Now Devitt oVers some interesting arguments in defense of the thesis that descriptions have referential meanings, operative when they are used referentially. In his view, ‘the core of the referential meaning of a description token is its reference-determining relation to the particular object that the speaker has in mind in using the description’.37 37 Presumably Devitt is not equating the reference of an expression token with what a speaker refers to in using the expression, for there would be no point in calling speaker reference ‘token reference’ and in regarding that as semantic. Here I will not address Devitt’s view that expression tokens have semantic properties of their own, and not just ones inherited from their types. However, I do think this view is problematic. Indeed, I regard token semantics as, well, token semantics (Bach 1987/1994: 85–8; 1999a). It is odd that he singles me out for ‘restrict[ing] semantics to the study of types’ (n. 4), for it is he who departs from the now common practice, endorsed by such prominent semanticists as Kaplan, Salmon, Soames, Richard, and Braun, of treating semantic reference as a property of types. This practice allows for referring expressions, notably ‘pure’ indexicals, to have references relative to contexts. I do not deny this, but only that deWnite descriptions, even when used to refer, have (semantic) references. Note that on this view, as Kaplan (1989a: 522–3) points out, an expression type can refer relative to a context even if not uttered/used in that context. As Kaplan argues, strange consequences result from attributing acts of utterance (or tokens) semantic properties of their own. For example, an utterance (or token) of the sentence ‘I say nothing’ cannot express a truth, but the sentence (type) is true relative to any context in which it is not uttered (1989b: 584). In any case, my discussion of Devitt’s arguments will not rely on my doubts about token semantics.

Bach, Descriptions: Points of Reference

225

Devitt does not oVer a complete account of referential meanings and, in particular, maintains ‘neutrality’ about the precise role of ‘F ’ in ‘the F ’. In this regard it is not clear that his discussion precludes an account on which ‘the F ’ (on its alleged referential reading) means the same as ‘the particular ‘‘F ’’ I am talking about/thinking of ’, a description that is uniquely satisWed by the unique object the speaker has in mind. If this, or anything like it, counted as a referential meaning, then Devitt’s account would be vulnerable to an objection based on Neale’s point (mentioned in Point 12) that a deWnite description containing a referring term, in this case the indexical ‘I’, is nevertheless amenable to Russellian treatment as a quantiWer phrase. At any rate, I will address only those of Devitt’s arguments that might adversely aVect my pragmatic approach to referential uses of deWnite descriptions.38 In particular, I will not be rebutting his arguments II, IV, and V.39 His main argument for referential descriptions is argument I, from regularity, according to which regularized uses of expressions, even if not their original uses, correspond to conventional meanings of those expressions. He bolsters this one with argument III, which likens referential uses of deWnite descriptions to deictic uses of demonstrative pronouns and phrases. And he rests his case with argument VI, which purports to show that when descriptions are exportable from opaque contexts they are referential. Since space limitations preclude giving Devitt’s arguments the detailed attention they deserve, I will focus on the main one, argument I, and respond brieXy to III and VI. Argument I aims to rebut Kripke’s (1977) and Neale’s (1990: ch. 3) implicit assumption that, because referential uses could be explained pragmatically, they should be so explained. In rebutting this assumption, Devitt takes his opponents to be likening referential uses to particularized conversational implicatures, whose understanding requires a fancy inference on the part of the hearer. It is not clear to me that this is Kripke or Neale’s view, but it is deWnitely not mine, as Devitt acknowledges (see his nn. 9 and 12). In any case, argument I does indeed challenge the analogy with particularized implicatures. It exploits the fact that not only can we use deWnite descriptions referentially, we regularly do so (hence no need for special stage setting). This 38 Also, I will not take up Devitt’s arguments as applied to indeWnite descriptions. For one thing, at least in my opinion, most of his examples of referential uses of them are really cases of speciWc uses (see Point 6). In these cases, the listener cannot engage in ‘reference-borrowing’ in order himself to refer to the individual in question, for there is no reference to borrow. And once these sorts of cases are disallowed, it turns out that referential uses of indeWnite descriptions are not regular but relatively rare, like referential uses of other quantiWer phrases, and these, Devitt grants, do not have referential meanings. 39 Argument II aims to neutralize Kripke’s (1977) well-known pragmatic argument from Russell English (as opposed to Donnellan English), where it is stipulated that deWnite descriptions are strictly quantiWcational. But just as he insists that Kripke’s argument is ‘unlikely to seem persuasive to someone who has not already accepted the case against referential deWnites’, Devitt concedes that his own argument is ‘unlikely to seem persuasive to someone who has not already accepted the case [for referential deWnites]’. As for his weak rigidity argument (IV), Devitt acknowledges that it does not provide independent support for the referential meaning thesis, since his opponent will treat referentially used descriptions as involved merely in the speaker’s expression of weakly rigid thoughts. And his argument from incomplete descriptions (V) assumes that his opponent treats uses of incomplete descriptions as either elliptical for complete ones or as containing implicit but semantically relevant restrictions on their domains. It should be clear from Point 12 above that I do neither.

226

II. Referential–Attributive Distinction

is not true of other sorts of quantiWer phrases that can also be used to refer, but generally only in special circumstances. I agree with Devitt that this is an important diVerence, but why does he take this as strong evidence that descriptions have referential meanings? His main point is that, even if there is a historical explanation for how deWnite descriptions, considered as quantiWer phrases, came to be used referentially, this does not show that they have not thus acquired a separate referential meaning. He likens this situation to the case of dead metaphors, such as ‘kick the bucket’ and ‘spill the beans’, which have acquired an idiomatic phrasal meaning distinct from their compositionally determined meaning. He suggests that even in cases where the phrasal meaning is inferable from the compositional meaning, this is still a distinct, conventional meaning. Devitt’s dead metaphor analogy is dubious. To endorse the Russellian claim that deWnite descriptions are quantiWer phrases, not referring terms, and that their attributive use is their only strictly literal use, does not commit one to the view that their attributive use came Wrst and their referential use somehow developed later. It does not require the supposition that there was a time when deWnite descriptions were not used referentially. However, Devitt is correct to argue that the mere derivability of one use from another does not show that the second use is not a conventional use in its own right. Even so, he does not consider the possibility that the quantiWcational meaning of descriptions plays a role in referential uses and in how hearers recognize uses as referential. Moreover, if their quantiWcational meaning does play such a role, then Devitt’s view implies that deWnite descriptions are semantically ambiguous in a most peculiar way: one of their meanings, the quantiWcational one, would be operative whenever the other meaning is. No other semantic ambiguity is like this, not even when formerly Wgurative uses of certain locutions became conventionalized in their own right. For example, if you hear an utterance of ‘Kirk kicked the bucket’, you can understand the idiom ‘kick the bucket’ to mean die, without having a clue about the once-living metaphor (whatever that was) from which it derived. In contrast to these examples, one cannot understand a referential use of a deWnite description without grasping its literal, quantiWcational meaning. At least so I have claimed. To address this claim, Devitt would need to show either that the quantiWcational meaning does not play a role in referential uses or, granting that it does, that this does not render his ambiguity claim problematic. Argument I is also vulnerable to a familiar objection. The claim that descriptions have referential as well as quantiWcational meanings seems committed to a massive cross-linguistic coincidence. For what could be the source of descriptions’ ‘systematic duplicity of meaning’ (Grice 1969: 143) but the semantic ambiguity of the deWnite article? But if that is the source of their double use, then the thesis that deWnite descriptions have referential meanings misses the cross-linguistic generalization that, in any language that has a deWnite article, deWnite descriptions have double uses. It would be a remarkable fact that an ambiguous word (‘the’ in this case) in one language should have

Bach, Descriptions: Points of Reference

227

translations in numerous other languages that are ambiguous in precisely the same way. In any case, argument I is directed only to the analogy of referential uses with particularized conversational implicatures. I take referential uses to be akin to generalized implicatures. Both are cases of standardization, pragmatic regularities that obviate the need for full-blown inferences on the part of the hearer (of course Grice would insist that the complete inference be ‘capable of being worked out’ (1989: 31), of being calculable, even if it does not need to be calculated). As Grice explains, in the case of a generalized conversational implicature, ‘the use of a certain form of words in an utterance would normally (in the absence of special circumstances) carry such-and-such an implicature’ (1989: 37; my emphasis). The audience still has to infer what the speaker means from what he says, but the inference is attenuated, not requiring the more elaborate reasoning needed for recognizing particularized implicatures. Acknowledging the diVerence, Devitt laments that ‘the absence of a [full-blown] Gricean derivation is, sadly, not suYcient to make a use semantic’ (n. 12). In this way, he concedes that argument I does not undercut the analogy of referential uses with generalized implicatures. It is important to see that if argument I were eVective against this analogy, it would prove too much.40 This regularity argument would lead to the conclusion that in sentences (13), (14), and (15), the words ‘and’, ‘believe’, and ‘some’ have special conventional meanings. (13) (14) (15)

It’s worse to go the hospital and get sick than to get sick and go to the hospital. I don’t believe astrology is bogus, I know it is. Michael drank some of my wine if not all of it.

In (13) ‘and’ indicates a causal relation in its Wrst occurrence and a temporal relation in its second. These uses of ‘and’ are quite regular, but there are (well-known) Gricean explanations for them. Yet Devitt’s argument implies that these uses, being regularized, are conventionalized, hence that ‘and’ is ambiguous in at least three ways. Similarly, it implies that the sense of ‘believe’ in (15) entails ‘does not know’, contrary to the popular philosophical opinion that knowledge implies belief, and that the sense of 40 Devitt also oVers a clever argument that to deny that certain regularities of use involve semantic conventions is to slide down a slippery slope to the conclusion that ‘there are no conventional meanings at all: it is pragmatics all the way down’. Such ‘Gricean Fundamentalism’ is, of course, absurd. However, the fact that Devitt has not identiWed where the Gricean draws the line between semantics and pragmatics does not show that there is no line to be drawn (see Bach 1999a for how I draw it). I do not deny that there are problem cases (see Bach 2001a: s. 5, for examples), but Devitt’s reductio, as well as argument I (as I explain in the text), conXates the distinction between conventionalization and mere standardization or pragmatic regularity. Moreover, this ‘Gricean fundamentalist’ reductio is based on a bad analogy. In a given communication situation the hearer’s job is to Wgure out how the speaker is using a certain expression, not how it could have come to be used in that way. Whenever there is a pragmatic regularity of using that expression in a certain way and similar expressions in relevantly similar ways, this job is virtually eVortless. For further discussion see the references later in this paragraph.

228

II. Referential–Attributive Distinction

‘some’ in (14) entails ‘not all’, contrary to elementary logic. As I have long stressed (see e.g. Bach and Harnish 1979: ch. 9; Bach 1995), we need to observe the distinction between conventionalization and mere standardization or pragmatic regularity, as illustrated by these and numerous other examples (see Levinson 2000). Whereas conventionalization makes possible a certain sort of use, standardization merely facilitates a certain use. By turning a speaker meaning not determined solely by the linguistic meaning of the words being used into what Levinson calls a ‘default meaning’, it streamlines the hearer’s inference to what the speaker means. A pragmatic regularity is not a matter of convention. In arguing that referential uses of deWnite descriptions are like literal deictic uses of demonstrative phrases (argument III), Devitt simply assumes that the latter have referential meanings. A recent book by JeV King (2001) suggests otherwise, arguing that such demonstrative phrases are quantiWcational. Even if he is wrong about that, King provides compelling linguistic data to show that deWnite descriptions are fundamentally diVerent from demonstrative phrases in that speakers’ intentions do not contribute to determining the semantic contents of uses of deWnite descriptions.41 Also, there are clear pragmatic diVerences between deWnite and demonstrative phrases. For example, compare the following uses of ‘that’ and ‘the’. In the case of (16) suppose that there are many books on the table and the speaker points to one of them. (16)

a. I haven’t read that book. b. #I haven’t read the book.

Clearly the utterance of (16b) is pragmatically anomalous. As for (17), suppose that several physicists have been mentioned successively in the conversation. (17)

a. That physicist is a genius. b. #The physicist is a genius.

In uttering (17a) the speaker is referring to the last physicist mentioned. It is not clear what he would be doing if he uttered (17b). These examples show that the conditions for using an incomplete deWnite description to refer are not the same as those for using a demonstrative phrase. For example, in contexts when one would use ‘that F ’, with stress on ‘that’, along with a demonstration to pick out one F from others, it would be not only odd but ineVective to use ‘the F ’. This is not surprising, considering that singular deWnite descriptions imply uniqueness and demonstrative descriptions do not. Devitt’s argument VI is based on the claim that ‘exportation works when the exported term is referential because then the opaque ascription requires that 41 For details see King (2001: 67–78; 133–9). As King explains and illustrates with various examples, this contrast between deWnite descriptions and demonstrative phrases applies not only to their referential uses but also to their nonreferential and even to their bound or what he calls ‘quantiWcation in’ uses. King uses the existence of these other uses and various facts about them to support his contention that demonstrative phrases have a uniWed, quantiWcational semantics.

Bach, Descriptions: Points of Reference

229

the believer be, like the speaker, en rapport with [have a singular thought about] the object of belief ’. The short Russellian response to this argument is that it is analogous to the following dubious argument for an exclusive as well as an inclusive sense of ‘or’: whereas ‘Michael drinks wine or Michael drinks beer’ and ‘Michael drinks wine’do not entail ‘Michael does not drink beer’, ‘Michael is in New York or Michael is in Sydney’ and ‘Michael is in New York’do entail ‘Michael is not in Sydney’. The trouble with this argument, of course, is that the second inference is just as invalid as the Wrst, even though inferences of that sort do not lead from truth to falsity (people cannot be in two places at once). Similarly, what argument VI shows is that when a description is used referentially (and there is actually something to which the speaker is referring), exporting a description from an opaque context cannot turn a truth into a falsehood. It does not show that the description has a referential as well as a quantiWcational meaning. The fact that exportation sometimes works and sometimes doesn’t does not show that descriptions have diVerent meanings in the two cases, with an inference rule operative in one case and not in the other. It is just a symptomatic by-product of the diVerence between the two uses.42 Devitt is well aware of the distinction between semantics and pragmatics and of the correlative distinction between what a speaker says and what a speaker means. He is not attributing referential meanings to deWnite descriptions arbitrarily, and he is right to insist that the possibility of a pragmatic explanation does not establish its correctness. Even so, it does not seem that his arguments succeed in undercutting the claim that referential uses are a matter of pragmatic regularity, not semantic convention. Moreover, there is a presumption in favor of pragmatic explanations. There is good methodological reason to appeal to independently motivated principles of rational communication, rather than to special features of particular expressions and constructions, in order to explain linguistic phenomena in as general a way as possible. In the present case, the fact that the referential use of deWnite descriptions is a cross-linguistic phenomenon suggests that it demands a pragmatic explanation, not a semantic one. Moreover, the denial that descriptions have referential meanings does not require crudely wielding Grice’s semantic model of Ockham’s razor and baldly appealing to parsimony. Rather, as I have suggested, the fact that descriptions are quantiWcational phrases of a certain sort helps explain their referential use. That is because their quantiWcational meaning plays a key role in their referential use. 42 If I were to respond to the details of argument VI, I would argue, as I have previously (Bach 1987/1994: 195–202), that one can referentially use a description embedded in the that-clause of a belief report without ascribing a singular thought, and one can ascribe a singular thought without using the description referentially. But I think it suYces to point out that argument VI can be reinterpreted in a way consistent with Russellian pragmatics.

5 The Good, the Bad, and the Ugly Nathan Salmon

I. One of the most important achievements in philosophy in the latter half of the last century was a movement in the philosophy of language, spilling over into metaphysics, epistemology, and the philosophy of mind. This movement has come to be known as the theory of direct reference. Keith Donnellan’s 1966 classic ‘Reference and DeWnite Descriptions’—spotlighting its famous distinction between the referential and the attributive use of deWnite descriptions—is an early and important precursor to the direct-reference theory, in its contemporary incarnation.1 Ironically, that article argues for a direct-reference theory on its least promising turf. During the Wrst half of the twentieth century, a broadly Fregean account of meaning and reference was generally held for all linguistic terms. Then the direct-reference theorists began exposing how badly the Fregean picture Wt certain sorts of terms. Especially and most obviously, the Fregean account failed for the logician’s individual variables. But it failed also for such common expressions as proper names, indexicals, pronouns, natural-kind terms, and more besides (phenomenon terms like ‘heat’, color words like ‘red’, artifact terms like ‘pencil’)—perhaps most, or even all, simple (or single-word) terms. Even after this chip, chip, chipping away of once cherished doctrine into scrap, one might still suppose that, if there are any terms for which the traditional, Fregean perspective is at least more-or-less correct, they are deWnite descriptions. One might suppose this, that is, but for Donnellan’s ground-breaking article. Donnellan argues instead that even deWnite descriptions are routinely used in a manner that ‘comes closer to performing the

As a student I had the privilege through most of the 1970s of taking a number of courses and seminars in philosophy from Keith Donnellan, in whose honor the present chapter was written. I am grateful to the participants in my seminar at UCSB during Fall 1993 (especially Ilhan Inan) for fruitful discussion of the issues presented here, and to Alan Berger for comments. I am grateful to Donnellan both for correspondence in connection with the seminar and more generally for his many contributions to my own philosophical development. 1 See also Donnellan (1968, 1978). A pioneering direct-reference theorist, Donnellan has made additional important contributions to the literature on the theory, especially in Donnellan (1972, 1974).

Salmon, The Good, the Bad, and the Ugly

231

function of Russell’s [logically] proper names’ (1966: 303), hence a use sharply out of sync with the traditional Fregean picture.2 A number of direct-reference theorists—including Barwise and Perry (1983: 149– 56 and passim), Devitt (1981b), Kaplan (1979, 1989b: 583–4), Recanati (1989a, 1993: 277–99), and Wettstein (1981, 1983)—have favored the broad outlines of Donnellan’s account. Others—notably Kripke (1977), in a farsighted and still underappreciated critique—have balked at Donnellan’s attempt to extend the notion of direct reference that far, seeing the distinction between referential and attributive use as fundamentally pragmatic in nature, with no special semantic signiWcance.3 Interestingly, however, Kripke (1977: 6–7, 22) concedes, in eVect, that he too is inclined to embrace a direct-reference theory for the most common type of deWnite description by far: the so-called incomplete deWnite description. Taking the hard line, I have argued (with special reference to Wettstein’s arguments) that going even this far is a mistake.4 I maintain that deWnite descriptions in English (and in Hebrew, Italian, Japanese, etc.), even referentially used incomplete ones, are not Russellian logically proper names. I say they are more like Fregean terms—or perhaps generalized quantiWers (as Russell thought)—but, I claim, devices of direct reference they are not. So goes the controversy within a controversy within a controversy. It is not to my purpose here to rehearse the arguments that I have given elsewhere against the thesis of semantic signiWcance. Rather I shall explore a host of philosophical issues raised by the semantic-signiWcance thesis itself, by Donnellan’s endorsement of it, by Kripke’s criticism of it, and more generally by various attempts to characterize the distinction between referential and attributive. These issues, which concern such things as de re belief and related matters, have applications in the theory of knowledge and the philosophy of mind that go well beyond the philosophy of language. First up: what is this controversial thesis of semantic signiWcance? Suppose that a speaker, Brown, utters the sentence ‘Smith’s murderer is insane.’ For some versions of the debate, the description ‘Smith’s murderer’ should be replaced by its ‘incomplete’ variation ‘the murderer’. In either case, the central question concerns whether Brown’s 2 If deWnite descriptions go the way of direct reference, is there anything left for the Fregean account to cover? Yes: the comparatively rare attributive uses of descriptions. Also, multi-worded phrases, like ‘middle-aged, stocky man with horn-rimmed glasses, a greying beard, and a balding head’, and . . . whole sentences! But if sentences may be fruitfully thought of as designating truth-values—as Frege and others have taught us—is it not so that they are typically used referentially, rather than attributively? Something to ponder. 3 See also Donnellan (1978). Putting the controversy in terms of the presence or absence of ‘semantic signiWcance’ may be misleading. As Kripke (1977: 21) suggests, speaker reference like anything else can become semantically relevant simply by virtue of being conversationally salient—whenever an expression (such as a deictic pronoun) is invoked that relies on conversational salience to secure reference. The point of the thesis of semantic signiWcance is this: When a deWnite description is used referentially (at least if the description is ‘proper’, in the Russellian sense, and used referentially for the object ‘denoted’, in Russell’s sense), the fact that it is so used is what directly determines that its semantic content is the object referred to by the speaker—by contrast with this being indirectly determined by means of some interceding phenomenon, like conversational salience, which directly determines semantic content. This is the sense in which, according to the thesis, the referential–attributive distinction has special semantic signiWcance. 4 In addition to Wettstein (1981, 1983), see W. Blackburn (1988) and Salmon (1982, 1991).

232

II. Referential–Attributive Distinction

use of the description (‘Smith’s murderer’ or ‘the murderer’) aVects which proposition is semantically expressed by the sentence with respect to Brown’s context. There is no (relevant) quarrel if Brown uses the description attributively. The consensus is that the sentence then expresses the proposition about Smith (at least indirectly about him), that whoever murdered him single-handedly is insane. The controversy turns on the question of what the semantic content of the sentence is if Brown uses the description referentially, but correctly (let us say) for Smith’s lone killer. According to the thesis of semantic signiWcance, ‘Smith’s murderer is insane’ then semantically expresses a proposition not at all about Smith, but instead a proposition about the murderer, that he is insane. Those of us who maintain that Donnellan’s distinction has no special semantic signiWcance contend that the semantic content of the sentence, with respect to the relevant context, is completely unaVected by Brown’s referential use. It still expresses the proposition about Smith. Donnellan and his followers thus endorse something along the lines of the following theses: ðSSa Þ

ðSSr Þ

If a speaker utters ‘Smith’s murderer is insane’ in an appropriate manner in a context c, then the speaker uses ‘Smith’s murderer’ attributively in c iV ‘Smith’s murderer is insane’ expresses the proposition that whoever single-handedly murdered Smith is insane as its English semantic content with respect to c. If a speaker utters ‘Smith’s murderer is insane’ in an appropriate manner in a context c, then the speaker uses ‘Smith’s murderer’ referentially for x in c iV ‘Smith’s murderer’ semantically refers in English to x with respect to c and ‘Smith’s murderer is insane’ expresses the singular proposition about x that he/she is insane as its English semantic content with respect to c.

Here the phrase ‘to utter in an appropriate manner’ means to utter a sentence as a sentence of a particular language in a normal way with assertive intent—by contrast with reciting a line in a play, conveying a message by secret code, etc. A singular proposition about an individual x is an ‘object-involving’ Russellian proposition that is about x by virtue of x ’s occurring directly as a constituent. Let us call the conjunction of the two theses ‘SS’. It is a thesis to the eVect that a deWnite description is indexical, expressing diVerent semantic contents with respect to diVerent contexts, depending (at least in some instances) on whether it is used referentially or attributively. Donnellan (1966) was not completely clear on this last point, leaving some readers to speculate that he conceived of his distinction as a lexical ambiguity rather than as a type of indexicality. This has led to some misplaced criticism. It should be noted that Donnellan (1966: 297) explicitly denied that deWnite descriptions are semantically ambiguous. And indeed, his contrasting notion of ‘pragmatic ambiguity’ seems to correspond very closely to the contemporary notion of indexicality, or perhaps to a

Salmon, The Good, the Bad, and the Ugly

233

special kind of indexicality.5 Donnellan’s account may thus be insulated to some extent against Kripke’s (1977: 18–20) appeal to H. P. Grice’s ModiWed Occam’s Razor principle that one should avoid ‘multiplying senses beyond necessity’.6 On the other hand, there is probably a worthy objection, analogous to Kripke’s plea for semantic economy, against positing indexicality beyond necessity—or at least beyond what is suYciently plausible on independent grounds.7 One of Kripke’s central objections is easily adjusted to target the indexical rather than the lexical-ambiguity version of the semantic-signiWcance thesis.8 So modiWed, it runs something like this: Donnellan’s distinction generalizes to cover proper names in addition to deWnite descriptions. For example, just as one may use the description ‘Mary’s husband’ referentially for someone who is not in fact legally married to Mary, one may also mistakenly use the name ‘Jones’ in reference to Smith, having mistaken him in the distance for Jones. Yet it is not plausible in the least that a proper name shifts in semantic reference with the context, depending on whether there is, over and above the speaker’s general intention always to use that name for the person so named, a particular person (or other object) whom the speaker has in mind and whom, on this particular occasion, the speaker means by the use of the name. Just as the fact that a name may be misapplied on a given occasion does not mean that the semantic reference of the name shifts to erase the mistake, nor does the semantic reference of a description shift to accommodate misapplication of the description.

II. It is important to note that the thesis of semantic signiWcance primarily concerns the semantic content of deWnite descriptions (or what is sometimes called the contribution toward ‘truth conditions’, or the ‘intension’), rather than the semantic reference. It is the thesis that the proposition expressed by a sentence containing a deWnite description (or the question of whether the sentence is true with respect to a given possible world), as opposed to the reference (with respect to the actual world) of the description itself, depends crucially on the use made of the description. Donnellan contends that the referent of a deWnite description d the we shifts with the context even when its matrix w is indexical-free. SpeciWcally, he maintains that a referentially used deWnite description 5 Donnellan has conWrmed in personal correspondence (Oct 1993) that the indexicality conception has always been his view of the matter. 6 See Grice (1969: 142–3; 1978: 118–20). 7 Cf. Recanati (1989a, 1993: 277–99) and Salmon (1991: 95 n. 6). 8 Kripke Wrst presented the objection, targeting the lexical-ambiguity version of the semantic-signiWcance thesis, in Kripke (1980: 25 n.)—in what is easily seen, in retrospect, to be a compressed summary of the not yet written ‘Speaker’s Reference and Semantic Reference’. See Kripke (1977: 13–21). Kripke’s formulations speak of ambiguity where the version presented here speaks instead of reference shifting with the context.

234

II. Referential–Attributive Distinction

refers (with respect to the context of utterance) to the person or object meant by the speaker even if that person or object does not Wt the description. This has proved highly controversial. But as Wettstein (1981) pointed out, the thesis of semantic signiWcance does not actually require Donnellan’s controversial contention. It is enough if the proposition semantically expressed when a deWnite description is used referentially for the right entity is the corresponding singular proposition about that entity. And indeed, Wettstein (1981: 243–4) maintains that the referential–attributive distinction is semantically signiWcant while not endorsing Donnellan’s more controversial claim. One can maintain a version of the semantic-signiWcance thesis that does not make Donnellan’s additional claim by weakening the controversial thesis SSr into the following: SSr0 If a speaker utters ‘Smith’s murderer is insane’ in an appropriate manner in a context c, then the speaker uses ‘Smith’s murderer’ referentially for Smith’s murderer in c iV ‘Smith’s murderer’ semantically refers in English to Smith’s murderer with respect to c and ‘Smith’s murderer is insane’ expresses the singular proposition about Smith’s murderer that he/she is insane as its English semantic content with respect to c. 0

Let SS 0 be the conjunction of theses SSa and SSr . It is neutral on the question of what happens when the speaker uses ‘Smith’s murderer’ referentially for someone other than Smith’s murderer. Donnellan’s thesis SSr supplements SS 0 to provide an answer to that 0 question. Indeed, I believe SSr is the only natural complement to SSr with regard to the question at hand. But SSr represents Donnellan at his most deWant, prompting at least one follower to retreat to the neutral version of the semantic-signiWcance thesis. The mere possibility of the more neutral version of the thesis of semantic signiWcance demonstrates that the objection from Kripke sketched at the end of the previous section stands in serious need of repair. First let us calibrate the referential– attributive distinction more Wnely. One should distinguish among three types of uses of deWnite descriptions: (i) correctly applied referential uses, that is, the use of a deWnite description d the we referentially for the person or object that satisWes w (or the person or object that satisWes a suitable expansion of w, in case the description is incomplete); (ii) incorrectly applied referential uses, that is, the use of d the we referentially for someone or something that does not uniquely satisfy (a suitable expansion of ) f; and (iii) attributive uses. Donnellan’s distinction is between (i)-cum-(ii) on the one hand, and (iii) on the other. The distinction-within-a-distinction between (i) and (ii) reveals a further interesting distinction, perpendicular to the referential–attributive distinction. Let us say that a use of either type (i) or type (iii) is a Good use, and that a use of type (ii) is Bad. The point made above may now be rephrased by saying that the thesis of semantic signiWcance does not require Donnellan’s complementary claim that a deWnite description, when used Badly, semantically refers to the entity meant by the speaker. A less deWant version of the thesis conWnes itself to Good uses, holding that the semantic content of a description with respect to such a use depends on whether that

Salmon, The Good, the Bad, and the Ugly

235

use is of type (i) or of type (iii). An eVective objection to the semantic-signiWcance thesis must expose some diYculty with the latter claim. Kripke’s Jones/Smith argument is aimed at Donnellan’s more full-blooded version of the thesis of semantic signiWcance which asserts SSr . In the example, a misapplied use of a proper name is contrasted with a correctly applied use of the name, where the former is analogous to a Bad use of a deWnite description, the latter to a Good use. It is argued that the Bad use cannot aVect the semantic reference of the name. This pays no attention to the question of semantic content, and hence inevitably misses the neutral version of the semantic-signiWcance thesis. Even when evaluated in this light, however, the argument is Xawed—and not merely because it leaves the door open for the neutral version of the semanticsigniWcance thesis. The principal defect is that Kripke has not succeeded by his Jones/ Smith example in extending the referential/attributive distinction to proper names. His discussion presupposes that typical correctly applied uses of a name are the analogue of the attributive use of a deWnite description. Correct uses of names are indeed Good, but they typically bear a much stronger kinship to Good uses of type (i) than to those of type (iii). The contrast between the correct use and the misuse of ‘Jones’ is roughly analogous to the distinction among referential uses between (i) and (ii). Since Kripke has not demonstrated a genuinely attributive use for a name, his Jones/Smith example does not adequately replicate the full grounds for the semantic-signiWcance thesis. For the purposes of Kripke’s objection, it still needs to be shown that a proper name can have contrasting uses analogous to, and as diVerent as, the referential use of a deWnite description (encompassing (i) þ (ii)) and the attributive. As we shall see in section V below, what Kripke actually provides is a distinction between uses that are, in a certain sense, automatically Good, and uses that are either Bad or only accidentally Good. This comes close, but still falls signiWcantly short of capturing Donnellan’s distinction. Inevitably, there are competing, non-coextensive ways of generalizing Donnellan’s distinction for deWnite descriptions to extend it to proper names. My point is not that one of these extensions is right and the rest are wrong. (This way of putting things threatens to ignite a dispute that is largely terminological.) The point is rather that a natural and plausible extension—one that aspires to capture and respect what is conceptually and philosophically at the core of Donnellan’s distinction—will cast our commonplace uses of ordinary names on the referential side rather than the attributive. And this is something Kripke’s generalized distinction evidently fails to do. I believe it is relatively uncontroversial that proper names are at least normally used referentially. Indeed, in his initial characterization of the distinction, Donnellan likened the referential use of a deWnite description to the use of a name, or at least of a ‘logically proper name’: Furthermore, on Russell’s view the type of expression that comes closest to performing the function of the referential use of deWnite descriptions turns out, as one might suspect, to be a

236

II. Referential–Attributive Distinction

proper name (in ‘the narrow logical sense’). Many of the things said about proper names by Russell can, I think, be said about the referential use of deWnite descriptions without straining senses unduly. (Donnellan 1966: 282)

The crucial question for the purpose of Kripke’s argument is: can a proper name be used instead in something more like the manner of an attributively used deWnite description? In order to construct a plausible and relatively clear-cut example of such a use, one is naturally led to consider the sort of cases that Kripke (1980: 54–60, 70, and passim) discusses under the rubric of Wxing the reference of a name by a description, that is, examples like Kaplan’s (1969: 228–9) introduction of the term ‘Newman 1’ as a name for whoever will be the Wrst child born in the twenty-second century.9 Very well, here is a proper name that is used attributively (if used at all). But can this name be used referentially? It can, though presumably not by us. Unless and until we take ourselves to have someone in mind who Newman 1 will be, we are powerless to bestow upon the proper name so deWned what Russell (1988: 21) described as ‘the direct use which it always wishes to have, as simply standing for a certain object, and not for a description of the object’.10 Imagine, then, that Newman 1’s future parents will be avid followers of the philosophical debates of the latter half of the twentieth century, and will decide that Kaplan has spared them the anxiety of Wnding the right name for their child. They will be able to use the name referentially. Voila`: a referential–attributive distinction for proper names. But now Kripke’s intended argument encounters a serious obstacle. The problem is that some philosophers would maintain, and indeed it is not at all implausible, that the parents’ future use of ‘Newman 1’ and our present use diVer in semantic content. In fact, judging from his more recent writings, it is not clear that Kripke himself is prepared to insist (as I am) that, despite the obvious diVerence in Xavor between the two uses, the name ‘Newman 1’ is semantically univocal.11 Finally, suppose this roadblock is somehow circumvented. Even if the case is successfully made that ‘Newman 1’ retains the same semantic content regardless of whether it is given a Good referential use or an attributive use, it is still open to the semantic-signiWcance theorist to argue, not implausibly, that this precisely reXects the 9 I owe the point that such reference-Wxing ‘deWnitions’ plausibly give rise to attributive uses for names to my former student, Ilhan Inan. 10 In Kaplan’s later writings, ‘Newman 1’ has been changed into a name for the Wrst child to be born in the twenty-Wrst century rather than the twenty-second. See e.g. Kaplan (1979: 397). No reason for the change was given. Perhaps having recognized Inan’s point (see the immediately preceding note), an indulgent Kaplan was simply growing impatient to give the name the direct use which it always wishes to have. (The change in example is accompanied by a radical change in view regarding what one can do with the name: see below.) Since the various controversies that surround reference-Wxing stipulations were not resolved before the turn of the century, I am granting us a small reprieve by reverting to Kaplan’s original example. I shall alter some quotations below accordingly. 11 I have in mind certain passages in the preface to Kripke (1980: 20–1), and especially in Kripke (1988: 146–7 n. 43 and 44). Kripke explicitly proclaims his neutrality on such issues.

Salmon, The Good, the Bad, and the Ugly

237

semantic gulf—which Kripke himself (1980: 55–8 and passim) insists upon—that separates the name from the description that Wxes its reference.

III. There is no dispute concerning the legitimacy of the referential/attributive distinction. The bone of contention concerns its signiWcance, or lack of signiWcance, for semantics. Given the existence of this controversy, one cannot simply take (an appropriate generalization of ) the conjunction of theses SS—or alternatively, the conjunction of theses SS 0 —as a neutral characterization of the distinction. Fortunately (and wisely), Donnellan (1966) provides distinct characterizations of the distinction. In the opening section, he characterizes it in terms of another distinction, that between the Russellian ‘denotation’ of a deWnite description and what a speaker refers to in using an expression. In ‘On Denoting’, after presenting his theory of descriptions, Russell explains his notion of denotation for deWnite descriptions as follows (using the word ‘proposition’ where nowadays we would probably use the word ‘sentence’): Every proposition in which ‘the author of Waverley’ occurs being explained as above, the proposition ‘Scott was the author of Waverley’ (i.e. ‘Scott was identical with the author of Waverley’) becomes ‘One and only one entity wrote Waverley, and Scott was identical with that one’ . . . Thus if ‘C ’ is a denoting phrase [i.e. deWnite description], it may happen that there is one entity x (there cannot be more than one) for which the proposition ‘x is identical with C ’ is true, this proposition being interpreted as above. We may then say that the entity x is the denotation of the phrase ‘C ’. Thus Scott is the denotation of ‘the author of Waverley’. (Russell 1905: 169)12

Similarly, then, the Russellian denotation of the description ‘Smith’s murderer’ is deWned as being the person who actually murdered Smith, if there is exactly one such person, and nothing otherwise—irrespective of whom the speaker might have in mind and mean, on a particular occasion, in using the phrase. The referential–attributive distinction may then be explained by saying that in a referential use of a deWnite description, but not in an attributive use, there is someone or something the speaker has in mind and to which the speaker refers using the description (and which the speaker’s assertion is thereby directly about), independently of its satisfying, or its not satisfying, the particular conditions that would make it the denotation, in Russell’s sense, of the description used. 12 An important aspect (all too often ignored) of Russell’s theory in ‘On Denoting’ is that a deWnite description d the we , though allegedly having no ‘meaning in isolation’, is nevertheless said to ‘denote’ the object that satisWes its matrix w, when there is only one such object, and to ‘denote’ nothing otherwise. This semantic relation is not simply an idle wheel in Russell’s philosophy; it carries a vitally important epistemological payload. It is through denoting, in this sense, that we are supposed to form beliefs and other thoughts ‘about’—and thereby to gain crucial cognitive access to—the many and varied objects so important in our lives but with which we are not directly acquainted (in Russell’s sense). See especially the Wrst two paragraphs of Russell (1905).

238

II. Referential–Attributive Distinction

Interestingly, Donnellan’s initial characterization of the referential–attributive distinction thus closely parallels Kripke’s later characterization of a more general distinction, of which Donnellan’s is supposed to be a special case, in terms of the Gricean distinction between speaker reference (what the speaker refers to) and semantic reference (what the expression refers to). The parallel is striking, but it is also very likely misleading. It is my impression—based on numerous lectures and discussions, as well as his writings—that Donnellan presupposes what I call the speech-act centered conception of semantics. On the speech-act centered conception, semantic attributes of expressions— like a singular term’s referring to an object, or a sentence’s expressing a proposition— somehow reduce to, are to be understood by means of, are derived from, or at least are directly determined by, the illocutionary acts performed by speakers in using those expressions, or perhaps the illocutionary acts that would normally be performed in using those expressions. This contrasts with an expression centered conception, which I favor, according to which the semantic attributes of expressions are not conceptually derivative of the speech acts performed by their utterers, and are thought of instead as intrinsic to the expressions themselves, or to the expressions as expressions of a particular language and as occurring in a particular context. The expression centered conception takes seriously the idea that expressions are symbols, and that as such, they have a semantic life of their own. The expression centered conception need not deny that semantics, at least for a natural language, may be ultimately a result or product of speech acts, rather than (or more likely, in addition to) the other way around. But the expression centered conception marks a deWnite separation between semantics and pragmatics, allowing for at least the possibility of extreme, pervasive, and even highly systematic deviation between the two. The speech-act centered conception is more reductionist in spirit. The expression centered conception is the received conception in the tradition of Frege and Russell. With their emphasis on artiWcial or idealized languages, it is they more than anyone else who deserve credit for cultivating the expression centered conception among contemporary philosophers of language. Wittgenstein focused, in contrast, on spoken, natural language in his impenetrable but seemingly penetrating diatribe against the expression centered conception. Whether or not he himself subscribed to the speech-act centered conception, it is he—with his inXuential slogan that ‘meaning is use’—who must bear the brunt of responsibility for that rival conception. If Donnellan subscribes to the speech-act centered conception, he is not alone. I fear it may be the dominant conception—especially among philosophers with a propensity toward nominalism, physicalism, anti-realism, or other reductionisms, and among those, like Donnellan, who trace their scholarly lineage to Wittgenstein. Anyone whose lineage traces back to Wittgenstein can trace it a step further to Russell. And there are indeed clear elements of both traditions manifest in Donnellan’s thought on reference and related matters. Still, his commitment to the speech-act centered

Salmon, The Good, the Bad, and the Ugly

239

conception might explain Donnellan’s unwavering endorsement of the stronger version of the semantic-signiWcance thesis. The speech-act centered conception cannot distinguish correctly between the semantic content of a sentence with respect to a given context and the content of the assertion, or assertions (statements, utterances), normally made by a speaker in uttering the sentence in that context. If this interpretation (which is somewhat speculative) is correct, then Donnellan conceives of speaker reference as, at least implicitly, a semantic, rather than a pragmatic, notion. Furthermore, he then conceives of Russell’s notion of denotation for deWnite descriptions as a nonsemantic notion, since it does not concern (at least not directly) acts of speakers’ reference normally performed with descriptions. This interpretation seems to be conWrmed by the subsequent discussion in Donnellan (1978). There he adopts Kripke’s terminology of ‘speaker reference’ and ‘semantic reference’, but he does not equate what he means by the latter with Russellian denotation, vigorously arguing instead that ‘semantic reference’ depends on, and is determined by, speaker reference, which may be other than the Russellian denotation.13 It is the expression centered conception, and the general Frege–Russell tradition, that is the natural habitat of the distinction between speaker reference and semantic reference (as well as such other Gricean distinctions as that between speaker meaning and sentence meaning). My own view—well within the Frege–Russell tradition—is that Donnellan’s apparent cataloguing of speaker reference as semantic and of Russellian denotation as non-semantic gets matters exactly reversed. It is just one piece of evidence of the extent to which the speech-act centered conception presents a seriously distorted picture of what semantics is, enough so that I am tempted to say that those in the grip of that conception, when applying such semantic terms as ‘refer’ and ‘express’ to expressions, are not talking about anything semantic at all.14 In any event, from the perspective of the expression centered conception it would be dangerous to take Donnellan’s characterization of the referential/attributive distinction in terms of speaker reference and denotation at face value. 13 Donnellan sometimes appears to allow for semantic reference in the absence of speaker reference. On the one hand, Donnellan (1978: 30, 32) says that in using a deWnite description attributively, the speaker does not refer to anything even if the description happens to be proper, in the Russellian sense. But he also seems to say (on the same p. 32, and in the same paragraph) that an attributively used description itself refers to its Russellian denotation. This leads to the curious position that when a speaker uses a proper description attributively, the description refers but the speaker does not. And this does not Wt well the speech-act centered conception of semantics. But coupled with Donnellan’s more central position that when a description is used referentially for something x, both the description and the speaker refer to x even if x is not the Russellian denotation, nor does this curious position exactly Wt the expression centered conception—or any other conception that I can think of. Though I am uncertain what to make of Donnellan’s assertions here, I believe that a careful reading reveals that appearances are deceptive, and that Donnellan (through a carefully placed occurrence of the subjunctive ‘would’) deliberately avoids any commitment to the claim that an attributively used proper description has semantic reference. I may be wrong. 14 See Salmon (1995: 18–19 n. 27).

240

II. Referential–Attributive Distinction

IV. Donnellan (1966: s. III, 285–9) alternatively characterizes the referential–attributive distinction in terms of what a speaker asserts (states, says) and the de-re/de-dicto distinction. He does not use the actual terms ‘de re’ and ‘de dicto’, nor any other arcane terminology for the latter distinction, but he clearly appeals to it. The central idea may be illustrated by a pair of theses paralleling those comprised by SS. Let us call the conjunction of the following theses ‘DT ’, for ‘Donnellan’s Thesis’: (DTa )

(DTr )

If a speaker utters ‘Smith’s murderer is insane’ in an appropriate manner in a context c, then the speaker uses ‘Smith’s murderer’ attributively in c iV the speaker, in uttering ‘Smith’s murderer is insane’ in c, asserts de dicto that whoever single-handedly murdered Smith is insane (and does not assert of anyone, de re, that he/she is insane). If a speaker utters ‘Smith’s murderer is insane’ in an appropriate manner in a context c, then the speaker uses ‘Smith’s murderer’ referentially for x in c iV the speaker, in uttering ‘Smith’s murderer is insane’ in c, refers to x and asserts of x, de re, that he/she is insane (and does not assert de dicto that whoever single-handedly murdered Smith is insane).15

I believe that many philosophers—including many who reject SS—would take DT to be analytic, by means of an appropriate generalization that literally deWnes the referential–attributive distinction in terms of de re and de dicto illocutionary acts (stating, asking, etc.). For example, in a criticism of Donnellan on the semantic-signiWcance thesis, Scott Soames characterizes the distinction by saying that a referential use of a description to refer to an individual o is a use in which the speaker says of o that o is such and such. What, we might ask, is it to say of an individual that it is such and such? The answer, it seems to me, is that to say of an individual that it is such and such is to assert the singular proposition that predicates such and such of that individual . . . In short, referential uses of deWnite descriptions are cases in which the speaker asserts a singular proposition about the individual the description is used to refer to. (Soames 1994: 149–52)16

Soames’s remark alludes to an intimate relationship that obtains, on the directreference theory, between the de-re/de-dicto distinction, on the one hand, and the distinction between singular and general propositions, on the other. To assert (or deny, believe, disbelieve, etc.) that such-and-such is to assert (deny, etc.) a certain proposition, the proposition that such-and-such. So to assert about (or to assert of ) 15 Donnellan has conWrmed in correspondence (see n. 5) that he endorses both of these theses in addition to SS. There are passages in Donnellan (1966) in s. III and again in s. VIII, suggesting, contra DT, that one makes a de re assertion when using a proper description (in the Russellian sense) attributively. Donnellan says that any such suggestion was unintended. 16 Compare also Neale (1990: 85). Though some of the general conclusions reached by Soames are friendly to the views expressed both here and in my previous writings on the subject, most of the arguments he uses to reach those conclusions are not (as will shortly become clearer).

Salmon, The Good, the Bad, and the Ugly

241

someone or something x that he/she/it is thus-and-so is to assert the proposition about x that he/she/it is thus-and-so. The latter is a singular proposition. De re assertion (or denial, etc.) is nothing more nor less than assertion (denial) of a singular proposition.17 Recognizing this relationship, an equivalence between DT and SS can be seen to follow from a general principle governing the separate phenomena that I distinguish under the epithets of ‘speaker assertion’ and ‘semantic content’ (Salmon 1982: 40–1). (AC )

If a speaker utters an English sentence S in an appropriate manner in a context c, then S expresses proposition p as its English semantic content with respect to c iV the speaker, in uttering S in c, asserts p.

This assertion/content principle is plausible. Some might even hold it to be analytic, true solely as a consequence of the meanings of ‘assert’, ‘semantic content’, and ‘utter in an appropriate manner’.18 And especially those under the spell of the speech-act centered conception of semantics tend to embrace the principle as trivial. Soames, on the other hand, must deny AC. For it is logically true that if AC, then (DT iV SS). Donnellan and his followers may have arrived at SS precisely via the assertion/content principle AC in combination with an implicit deWnition or characterization of the referential/attributive distinction in terms of de re and de dicto assertion. But taking DT to be true, let alone analytic, is a mistake. In fact, I contend that both theses DTa and DTr have straightforward counter-examples. Indeed, I believe all are false: AC, DT, and SS. The case against DTand AC is probably best seen in the light of a phenomenon that Kaplan (1989a) has called the pseudo de re: A typical example is ‘John says that the lying S.O.B. who took my car is honest’. It is clear that John does not say, ‘The lying S.O.B. who took your car is honest’. Does John say d d is honeste for some directly referential term d which the reporter believes to refer to the lying S.O.B. who took the car? Not necessarily. John may say something as simple as, ‘The man I sent to you yesterday is honest’. The reporter has simply substituted his description for John’s. What justiWes this shocking falsiWcation of John’s speech? Nothing! But we do it, and often recognize—or don’t care—when it is being done. The form lends itself to strikingly distorted reports. As Church has shown, in his Introduction to Mathematical Logic (Princeton: Princeton University Press, 1956), on page 25, when John says ‘Sir Walter Scott is the author of Waverley’ use of the pseudo de re form (plus a quite plausible synonymy transformation) allows the report, ‘John says that there are twenty-nine counties in Utah’! I do not see that the existence of the pseudo de re form of report poses any issues of suYcient theoretical interest to make it worth pursuing. (Kaplan 1989a: 555– 6 n. 71)

17 I provide an argument for this in Salmon (1986: 2–6) and a more detailed treatment in Salmon (1990). 18 The word ‘assert’ is used throughout the present chapter in its usual sense, which is strict enough to allow for the distinction—highlighted by Grice—between what a person asserts (or ‘says’) and what he/she means (‘implicates’) by what was actually asserted. Not everything that a speaker means is asserted, nor vice versa. Cf. Salmon (1991: 88).

242

II. Referential–Attributive Distinction

The Church argument mentioned by Kaplan is principally concerned with the question of whether sentences should be said to refer (‘denote’). The argument shows under relatively minimal assumptions that ‘Scott is the author of Waverley’ and ‘There are 29 counties in Utah’ refer to the same thing, thus supporting Frege’s doctrine that sentences refer to their truth-values. Adapting Church’s original argument to embeddings of such sentences within the non-extensional phrase ‘John says that’, Kaplan tacitly considers the following chain of assertion attributions (where we may let the Wrst be true by hypothesis): (i) John says that Scott ¼ the author of Waverley (ii) John says that Scott ¼ the man who wrote 29 Waverley Novels altogether (iii) John says that the number n such that Scott ¼ the man who wrote n Waverley Novels altogether ¼ 29 (iv) John says that the number of counties in Utah ¼ 29. Here both (ii) and (iv) are obtained from their immediate predecessors by the liberal sort of substitution characteristic of the so-called pseudo de re, whereas (iii) is obtained from (ii) by the mentioned ‘plausible synonymy transformation’—in this case the assumption that ‘Scott ¼ the man who wrote 29 Waverley Novels altogether’ and ‘The number n such that Scott ¼ the man who wrote n Waverley Novels altogether ¼ 29’ are synonymous (or at least suYciently close in meaning that if John asserted the content of the Wrst, then he may be accurately reported as having asserted the content of the second). The Wnal attribution (iv) intuitively does not follow from (i). Following Church, Kaplan considers a further attribution obtained from (iv) by a second application of the synonymy transformation, though this is strictly unnecessary to the argument. Indeed, even (iv) is unnecessary, since (iii) already goes well beyond what may be validly inferred from (i). Kaplan’s apparent conclusion—based at least in part on his liberal adaptation of Church’s argument—is that pseudo de re substitutions are not in general truth-preserving. It is presumably for that very reason that they are supposed to pose no interesting theoretical issues. Wettstein (1986) has argued that the phenomenon in question leads, on the contrary, to a highly signiWcant conclusion: In many, many contexts of reporting what other people say, think, believe, and so on, substitutions of embedded singular terms preserve truth, and so do substitutions of names for other names, even names for deWnite descriptions, deWnite descriptions for names, or deWnite descriptions for deWnite descriptions, as the following examples illustrate. . . . Tom, a new faculty member, is told about all the new funding that the dean has arranged for faculty research. He says, not having any idea of who the dean is, ‘The dean is obviously very smart’. I report to Barbara that Tom believes that Mike is very smart or that Jonathan’s soccer coach is very smart (in case Barbara, say, characteristically refers to the relevant individual as ‘Mike’ or is most familiar with him in his role as Jonathan’s soccer coach).

Salmon, The Good, the Bad, and the Ugly

243

Such substitutions, at least in the sorts of contexts indicated [like Kaplan’s], are perfectly acceptable. Nor do we, in making such substitutions, have to worry about preserving or reporting the Fregean sense of the original remarks. In such contexts at least, the truth or falsity of the report depends not upon accurately capturing the Fregean thought believed, but simply upon correctly formulating who it is the believer has a belief about and what the believer believes about him. . . . . . . Belief reports are extremely resistant to neat theoretical treatment—and this is so on either the Fregean or the anti-Fregean orientation. Perhaps a neat treatment is not even possible. (Wettstein 1986: 205–8)

Wettstein adds in a footnote that truth-preserving substitutions of nonsynonymous expressions in assertion or belief reports, etc., ‘are particularly interesting, since not only Fregeans but just about everyone has assumed that such substitutions ought not to preserve truth. This shows, I think, that we’ve virtually all had the wrong idea about the semantics of attitude reports.’ In sharp contrast to Kaplan’s summary dismissal of the so-called pseudo de re as theoretically uninteresting, Wettstein says that the phenomenon, in combination with other data, ‘suggests that what is reported is not (at least not exclusively) propositional content believed’. Wettstein adduces this data to motivate an unorthodox and highly controversial theory of the meanings of attitude reports. (Since he does not address Kaplan’s discussion, Wettstein does not respond to Kaplan’s adaptation of Church’s argument.) Wettstein’s example involving Tom and the dean is suYciently similar to Kaplan’s involving John and the lying S.O.B. that both obviously qualify as instances of the general phenomenon that Kaplan means by his phrase ‘the pseudo de re’. One signiWcant diVerence between the two cases, however, or at least a potentially signiWcant diVerence, is that John presumably uses the description ‘the man I sent to you yesterday’ referentially for the liar in question, whereas Tom uses ‘the dean’ attributively. This is supported by (and may be the main point of ) Wettstein’s remark that Tom has no idea who the dean is. But this diVerence does not matter as regards the overall pattern. In fact, the attributive use in Wettstein’s example makes it a purer example, in some sense, of the phenomenon that Kaplan has in mind. The main point is: it seems we readily accept the reporter’s substitution of his own description in either case. The position I take with regard to the pseudo de re steers a middle course between the diametrically opposed conclusions of Kaplan and Wettstein, and has more to recommend it than either of these other treatments. I agree with Wettstein, as against Kaplan, that such substitutions are truth-preserving. But I do not agree that they call for an unconventional account of attitude attributions. The central characteristic of the pseudo de re—and the precise reason for the presence of the words ‘de re’ in the phrase—is that, as Wettstein puts it, ‘the truth or falsity of the report depends not upon accurately capturing the Fregean thought believed, but simply upon correctly formulating who it is the believer has a belief about and what the believer believes about him’. One can put this without begging the crucial question by talking about the acceptability or unacceptability of the report instead of its truth or falsity. The deWning

244

II. Referential–Attributive Distinction

feature of the pseudo de re is that such reports behave in ordinary discourse as if they were de re. They do this, I contend, for a very simple reason: they are de re. I mean that they are not ‘pseudo’ at all; they are genuine, ordinary, conventional, authentic, bonaWde, run-of-the-mill, barnyard-variety, par-for-the-course de re, nothing more and nothing less. In Russell’s terminology, the relevant description occurrence is a primary occurrence rather than a secondary occurrence, the description has wide scope rather than narrow scope. Kaplan’s term ‘pseudo de re’ is a seriously misleading misnomer. In Wettstein’s example, Tom believes the proposition that whoever is dean is very smart. This is de dicto rather than de re, general rather than singular. But, and in part in virtue of this general belief, Tom also believes of the dean, that is, of Jonathan’s soccer coach Mike, that he is very smart. It is this latter de re belief that I contend Wettstein is reporting when he says, ‘Tom believes that Jonathan’s soccer coach is very smart’. In any event, that report must be interpreted de re rather than de dicto—with the description ‘Jonathan’s soccer coach’ taking wide rather than narrow scope—if it is to have any hope of being a sincere attempt at accurate reporting (assuming that Wettstein does not believe that Tom has independently formed the opinion, after watching Jonathan play, that he has been cleverly coached). It would also seem that the report, so interpreted, is indeed true. Tom’s having no idea who the dean is does not prevent him from forming a favorable opinion about the dean’s intelligence. Likewise in Kaplan’s example, the report ‘John says that the lying S.O.B. who took my car is honest’ must be interpreted de re rather than de dicto if it is to have any hope of being accurate—the de dicto reading being ruled out of court as a gratuitous attribution of inconsistency. And interpreted de re, the report is quite accurate, even if not completely faithful. (Indeed, the truth of the de re report is perhaps even more evident in Kaplan’s example, despite his misgivings, due to John’s referential use of the relevant description.) One may object that in uttering the words ‘The man I sent to you yesterday is honest’, what John literally says is that whichever man he had sent the day before is honest. This observation is indeed correct. But it should not be lodged as a protest. For again I say that it is precisely by literally saying that whichever man he had sent the day before is honest that John says of the liar in question, de re, that he is honest. (A faithful report would presumably report what John literally asserts, rather than what he indirectly asserts by virtue of his literal assertion.) This is by no means a singular or unusual case. Nor does it take essential advantage of the fact that the original speaker used a description referentially. There is also someone whom Tom says is very smart (namely of course, the dean/coach), even when using the description ‘the dean’ only attributively. Virtually whenever one asserts that the such-and-such is thus-and-so, one thereby asserts of the such-and-such, if there is exactly one, that he/she/it is thus-and-so. Whether for good or bad, this evidently is how our notion of de re assertion works.19 19 Cf. Salmon (1982: 1–2; 1991: 88). It is not implausible that exceptions arise in cases where d such-andsuche trivially entails d thus-and-soe , as e.g. in ‘The shortest spy is a spy’. See Searle (1979b: 207); and Salmon (1982: 45 n. 7). I am strongly inclined to believe, however, that these are not genuine exceptions to a fully encompassing latitudinarianism with respect to assertion.

Salmon, The Good, the Bad, and the Ugly

245

This position (unlike Wettstein’s) easily blocks Kaplan’s adaptation of Church’s argument. In the succession of assertion reports (i)–(iv), each must be read either de dicto or de re. If the transition from (i) to (ii) is to be an instance of the same general phenomenon as we Wnd in the lying S.O.B. case and the dean/coach case, then (i) is to be interpreted de re, reporting that John says of the author of Waverley that Scott is him. Interpreted de re, (ii) then straightforwardly follows (assuming that Scott wrote exactly twenty-nine Waverley Novels). Similarly, if the transition from (iii) to (iv) is to be an instance of the same phenomenon, then both are to be interpreted de re, as reporting that John says of a certain number that it is twenty-nine. So interpreted, however, the transition from (ii) to (iii) cannot be justiWed as a ‘plausible synonymy transformation’. Indeed, so interpreted, (iii) does not even appear to follow from (ii); if John is suYciently taciturn, (iii) may be false, interpreted de re, even when (ii) is true (interpreted either way). The transition from (ii) to (iii) can be justiWed as a mere synonymy transformation only if both are interpreted de dicto rather than de re. It thus emerges that Kaplan’s adaptation of Church’s argument turns on the fallacy of equivocation— where the crucial ambiguity is not lexical but an ambiguity of scope.20 Invoking the connection between the de re and singular propositions, the position I am defending is tantamount to the claim that in asserting a general proposition to the eVect that the such-and-such is thus-and-so, one typically also asserts the corresponding singular proposition. In a single utterance John asserts at least two diVerent things: that the man sent the day before is honest, and the singular proposition about the liar in question that he is honest. More generally, in uttering a sentence d c(the w)e , one thereby typically asserts two propositions: the general proposition which is the semantic content of the sentence (this is one’s literal assertion); and indirectly (and nonliterally), in virtue of the Wrst assertion, also the corresponding singular proposition about the person or object that uniquely satisWes w, if there is one. The speaker buys two propositions for the price of one. This is so, I contend, whether the deWnite description d the we was used referentially or attributively, and even if the description was used Badly, that is, even if it was used referentially for someone who does not satisfy w. In this special case, the speaker may have asserted no less than three propositions— all for the price of one. It is precisely this possibility of multiple assertion by a single utterance that deWes principle AC and both DTa and DTr . The availability of this simple, straightforward account of the so-called pseudo de re has not been widely recognized.21 There are a number of sources for this oversight. First, there is the familiar (if still somewhat controversial) observation that de re belief does not follow from de dicto, even assuming the relevant person or object exists. In the now hackneyed example, Kevin may believe, solely on the basis of reXection on the concepts, that whoever is shortest among spies is a spy, without thereby suspecting 20 Church’s original argument commits no fallacy; in my judgment, its soundness is unimpeachable. It is Kaplan’s adaptation of Church’s argument that commits the fallacy of equivocation. 21 One possible exception is Sosa (1970: 890). See also Searle (1979b).

246

II. Referential–Attributive Distinction

anyone in particular of being a spy. That is to say, latitudinarianism—the doctrine that the inference from the de dicto, together with an existence premise, to the de re is valid—is mistaken. In order to graduate from de dicto belief to de re, one must bear some epistemically substantial connection to the person or object in question. Notoriously, there is no consensus concerning the precise nature of de re connectedness, but there is widespread (even if not unanimous) agreement that it is cognitively more ‘real’ than the mere coincidence that obtains between Kevin’s apprehension of the concept he/she who is shortest among spies and the person whom that concept happens to Wt. To use Kaplan’s (1969) phrase, one must be en rapport with the entity in question. Russell held that one must be directly acquainted with the entity, in his peculiar sense. This requirement is easily seen to be excessive, and more recent philosophers have substituted various weaker acquaintance relations for Russell’s. Many embrace the view that one must merely know who the person is, or know what object it is, in an ordinary sense. Some say instead—or in addition—that one must have the person or object ‘in mind’ suYciently to be able to use a term referentially for him/her/it.22 It is assumed furthermore that de re assertion has an analogous prerequisite, one appropriate to assertion in lieu of belief. For example, it is held that the subject must possess and use a special sort of singular term—‘a vivid name’ perhaps, or a directly referential, logically proper (Millian) name, or at least a term used referentially rather than attributively.23 In addition to all of the above, there is a general pre-evidential bias in favor of the tenet that a speaker is allowed only one assertion per utterance of an unambiguous sentence (perhaps as a consequence of the assertion/content principle AC ). These are myths. The example of the shortest spy does indeed show that latitudinarianism with regard to belief is mistaken. But de re belief does not require anything as stringent as knowing who the person is or having the object ‘in mind’, in a Donnellanian sense. An eyewitness distinguishing the culprit from the decoys in a police lineup has a de re belief, but may not ‘know who’ that person is, in the usual sense. And when the investigating homicide detective utters ‘Smith’s murderer is insane’ using the description attributively, there is indeed someone of whom the detective suspects insanity, though not someone he has in mind (in the relevant sense). The detective need not even be ‘acquainted’ with the murderer, in any ordinary sense; his knowledge of the murderer is by description, in Russell’s phrase. Never mind; he still manages to pull oV a de re belief. It is enough that the believer is appropriately cognitively connected to the person or object. The de re connection need not be direct and intimate; it may be

22 For an example of the Wrst view, see Quine (1981: 272–3); Soames (1994: 159–62) is similarly Xawed by presupposition of an admixture of both the Wrst and second view. 23 Kaplan (1969) seems to have required the use of a vivid name (among other things). Later in Kaplan (1979) and also in his brief discussion (quoted above) of the so-called pseudo de re in Kaplan (1989a), he evidently presupposes instead that use of a directly referential term is both necessary and suYcient. Still later, in Kaplan (1989b: 583 n. 36), he appears to move toward the less restrictive view that a referential use is suYcient. As should already be clear, I believe this permissive trend in Kaplan’s thought is entirely positive and ought to be followed to its natural, and plausible, conclusion.

Salmon, The Good, the Bad, and the Ugly

247

remote and indirect, perhaps consisting of a network of causal intermediaries interposed between the cognizer and the object.24 Donnellan (1979: 58) suggested that in order to assert something de re about a person or object, there is no requirement, of the sort Kaplan (1969) laid down, that the speaker use a vivid term or even that the speaker use a term that denotes the entity in question. Interestingly, Donnellan suggested instead that Kaplan’s third and Wnal condition is both necessary and, by itself, suYcient: that one use a term that is a name of the entity for the speaker—analogous to the sense in which a bad photograph may be a picture of an object that it does not resemble, and fail to be a picture of another object to which it bears an uncanny resemblance (Kaplan 1969: 227–9). That is, Donnellan suggested that it is necessary and suYcient that the entity enter properly into the ‘genetic’ account of how the speaker came to learn the term he/she uses to refer to it. I am suggesting that some such condition (perhaps one involving ‘mental names’?) may be operative in the formation of de re beliefs, thus blocking Kevin from suspecting anyone in particular of espionage while allowing the homicide detective to form his de re diagnosis of insanity. But pace Donnellan, any such condition seems to me overkill for the making of a de re assertion. Kaplan’s well-documented change of attitude toward the de re is also accompanied by a shift in which attitude is alleged to be de re (see n. 10 above). Kaplan (1969: 228–9) says, ‘I am unwilling to adopt any theory of proper names which permits me to perform a dubbing in absentia, as by solemnly declaring ‘‘I hereby dub the Wrst child to be born in the twenty-second century ‘Newman 1’ ’’, and thus grant myself standing to have beliefs about that as yet unborn child.’ Kaplan (1979: 397) recants: ‘All this familiarity with demonstratives has led me to believe that I was mistaken in ‘‘Quantifying In’’ in thinking that the most fundamental cases of what I might now describe as a person having a propositional attitude (believing, asserting, etc.) toward a singular proposition required that the person be en rapport with the subject of the proposition. It is now clear that I can assert of the Wrst child to be born in the twenty-Wrst century that he will be bald, simply by assertively uttering ‘Dthat (‘‘the Wrst child to be born in the twenty-Wrst century’’) will be bald’.

I say that Kaplan was right on both counts. Where he goes wrong is in thinking that his second observation shows that his Wrst was mistaken. Saying something about Newman 1 is a piece of cake. Forming a belief about him/her, by contrast, requires some degree of cognitive connection, however sparing. De re connectedness is required for de re belief, not for de re assertion. We must guard against deciding at the outset, before considering the evidence, that all of the propositional attitudes behave as one—especially if something that makes as little cognitive demand on the subject as mere assertion is counted as one of the attitudes. Perhaps one must apprehend propositions in order to believe them. And perhaps one must apprehend propositions in order to make assertions. But it is doubtful that one must apprehend what one is asserting in order to assert it. Whereas 24 Cf. Sosa (1970). See also Salmon (1986: 179–80 n. 19; 1987/1988: 199 n. 8, 204 n. 11, 213 n. 17).

248

II. Referential–Attributive Distinction

latitudinarianism fails in the case of belief, some form seems to govern assertion. This conclusion does not reXect an idiosyncratic theoretical bias. Regardless of one’s views on the controversial issues, there seems every reason to admit that, intuitively, when Tom says ‘The dean is very smart’ he thereby says of the administrator in question that he is very smart, and when John says ‘The man I sent to you yesterday is honest’ he thereby says of the liar in question that he is honest. There is even some intuition that when I say ‘Newman 1 is unconnected to us’, I thereby assert something of a particular future individual. Ironically, evidence in favor of my proposal comes indirectly from Donnellan (1979). The burden of that article is to challenge Kripke’s famous examples of allegedly contingent a priori statements. Kripke’s examples are trivial consequences of stipulations that Wx the reference of a new name or other term by means of a deWnite description—sentences like ‘Assuming Newman 1 will exist, he or she will be born in the twenty-second century’. The overall structure of Donnellan’s argument is that the sort of knowledge contained in such sentences is de re. Yet one typically cannot gain such de re knowledge merely on the basis of the reference-Wxing introduction of the name, without further experience of the object. In the course of the argument, Donnellan applies a pair of general principles to show that one has not gained the relevant de re knowledge. The principles—let us call them ‘K 1’ and ‘K 2’—are stated by Donnellan as follows: (K 1)

(K 2)

If one has a name for a person, say ‘N’, and there is a bit of knowledge that one would express by saying ‘N is w’ then if one subsequently meets the person it will be true to say of him, using the second-person pronoun, ‘I knew that you were w’, If an object is called by one name, say ‘N’, by one group of people and by another name by a second group, say ‘M’, and if, in the language of the Wrst group ‘N is w’ expresses a bit of knowledge of theirs and if ‘is c’ is a translation of ‘is w’ into the language of the second group then if the relevant facts are known to the second group, they can say truly that the Wrst group ‘knew that M is c’. (Donnellan 1979: 55)25

Here the ‘you’ and the ‘M’ are to be taken as occurring within the scope of the nonextensional operator ‘knew that’. Donnellan adds that essentially the same considerations that were adduced for denying that there was knowledge of an entity just in virtue of the sort of stipulation that introduces a rigid designator by means of a description can be applied to the other propositional attitudes. It would, for example, seem to me just as incorrect to say to John who turns out to be the Wrst child born in the [twenty-second 25 In an endnote Donnellan recognizes that for the purposes of his argument, he does not need to defend these principles even for cases in which the user of the name ‘N’ also uses a second name for the same person or object, believing the two names to refer to diVerent entities. But Donnellan also there expresses a temptation to extend the principles to some of these cases as well. (My own view is that they should be extended across the board to all such cases.)

Salmon, The Good, the Bad, and the Ugly

249

century], ‘I believed about you some [one hundred and] twenty-Wve years before your birth . . . ’ (Donnellan 1979: 56–7).

Donnellan evidently endorses the following analogues of his stated principles: (A1)

If one has a name for a person, say ‘N’, and one makes an assertion (in the ordinary way) by uttering ‘N is w’ then if one subsequently meets the person it will be true to say of him, using the second-person pronoun, ‘I said that you were w’, (B1) If one has a name for a person, say ‘N’, and one believes what one would express by saying ‘N is w’ then if one subsequently meets the person it will be true to say of him, using the second-person pronoun, ‘I believed that you were w’; etc.

These various principles, in eVect, licence the substitution, under appropriate circumstances, of a name or of a simple indexical (‘you’) for a co-referential name in an attribution of an assertion or other propositional attitude. The basis for these principles is the fact that such sentences as ‘You were w’, ‘N is w’, and ‘M is c’, with ‘N’ and ‘M’ being names, semantically contain singular propositions (in the relevant languages), so that one who utters them assertively makes a literal de re assertion about the referent of the name or indexical, and any knowledge or belief of the propositions they contain is de re knowledge or belief.26 This suggests certain more fundamental principles: (A10 )

If an expression, say ‘w’, expresses a property (state, condition) F in one’s language, then one asserts about a person, de re, that he/she is F iV, if one were subsequently to meet the person, it would be true to say to him, using the second-person pronoun, ‘I said that you were w’. 0 (B1 ) If an expression, say ‘w’, expresses a property (state, condition) F in one’s language, then one believes about a person, de re, that he/she is F iV, if one were subsequently to meet the person, it would be true to say to him, using the second-person pronoun, ‘I believed that you were w’; etc.

Indeed, it is diYcult to imagine a justiWcation for A1 that does not go by way of the notion of de re assertion, or some closely related notion. In any event, I think it is clear that Donnellan bases his principles on more fundamental principles like these.27 26 Donnellan (1974) more or less endorses the idea that ‘predicative statements’ (those predicating properties other than existence, or its cognates) involving names semantically express singular propositions. 27 Strictly speaking, only one of the two conditionals making up the biconditional in B10 is needed for Donnellan’s argument, though I believe it is clear that he would endorse the full biconditional. Just before formulating his general principles, Donnellan (1979: 54) says, ‘I am assuming, to use the jargon, that if we now have any knowledge (other than about linguistic matters) just as a result of the stipulation concerning the sentence, ‘‘Newman 1 will be the Wrst child born in the [22nd] century’’ it would have to be knowledge de re. That is, it would have to be knowledge about an individual in the sense that there is (or will be) an individual about whom we now know something and if that individual turns out to be John we now know something about John.’

250

II. Referential–Attributive Distinction

Earlier, Ernest Sosa also implicitly relied on principles like these in making a determination concerning whether one has a given de re belief about someone or something, or has made a de re assertion, etc.28 I shall call this ‘the Donnellan–Sosa test’. Illustrating how the test applies to the case of Newman 1, Donnellan observes, If the Wrst child born in the [twenty-second century] comes to be named ‘John’ it would not be correct to say then that although we had a diVerent name for him we knew [one hundred and] twenty-Wve years beforehand that John would be the Wrst child born in the [twenty-second century] . . . I suggest that the reason is that the stipulations have not given rise to any knowledge (other than of linguistic matters). And so not to any knowledge a priori. (1979: 55)

My principal concern here is not with the thorny question of whether Kripke’s alleged examples of the contingent a priori hold up under such careful scrutiny.29 Our concern is instead with the more immediate matter of whether there is de re belief or assertion present in the sort of examples that Kaplan has labeled the pseudo de re. And here the more fundamental principles A10 and B10 are the ones to employ. By contrast to the Newman 1 case, John in Kaplan’s example, having sincerely uttered the sentence ‘The man I sent to you yesterday is honest’, surely could truthfully address the man in question with the words ‘I said that you were honest’, and even with ‘I believed that you were honest’. Hence, applying A10 and B10 , John did assert and believe of the man, de re, that he was honest. And just as certainly, Tom in Wettstein’s example could address the dean truthfully (if shamelessly) with the words, ‘I told Wettstein that you were very smart’, and with ‘Even before I learnt who you were, I had already formed the opinion that you were very smart, based on the wonderful things you have done for the faculty’. Consequently, Tom made the relevant de re assertion, and had the corresponding de re belief. The contrast between Kaplan’s example and Wettstein’s now looms large. Recall that Tom, unlike John, uses his description attributively. Since Donnellan endorses DTr as well as the principles A1, etc., in order to avoid inconsistency he must deny that Tom can truthfully say to the dean ‘I said, and believed, that you were very smart’. It is not surprising, therefore, to Wnd that he says of the Newman 1 case, ‘It would . . . seem to me just as incorrect to say to John who turns out to be the Wrst child born in the [twenty-second century], ‘‘I believed about you some [one hundred and] twentyWve years before your birth . . . ’’, ‘‘I asserted about you some [one hundred and] 28 See Sosa (1970: 890–1). Others besides may have also used the test. I believe Sosa may be the Wrst to have done so, and to have concluded, correctly, that in some cases, one may have a de re belief (or other attitude) about someone without knowing who it is. 29 I discuss this issue at length in Salmon (1987). In a lecture presented in 1984 at a conference at Stanford University, Kripke responded to Donnellan’s criticism, arguing that what I am calling ‘the Donnellan–Sosa test’ does not support Donnellan’s contentions in the case of Leverrier’s introduction of ‘Neptune’ as a name for whatever planet it is causing particular perturbations in the orbit of Uranus. Cf. Kripke (1980: 79 n.). Notice that ‘Neptune’ was already a name of the planet (in Kaplan’s sense) for Leverrier when he so named it, even before it was located in the sky. The case is quite diVerent with ‘Newman 1’, where Donnellan’s criticism of Kripke seems signiWcantly stronger. In the Stanford lecture, Kripke developed and modiWed his position on the contingent a priori. See Salmon (1987: 203 n.).

Salmon, The Good, the Bad, and the Ugly

251

twenty-Wve years before your birth . . . ’’, etc.’30 In fact, however, these two sentences seem signiWcantly diVerent. It is indeed dubious that Newman 1’s future contemporaries could truthfully utter ‘Some philosophers of the late twentieth century believed that you would not be born until the twenty-second century’. For despite Kaplan’s heroic eVorts, we simply are not suYciently en rapport to have de re beliefs about Newman 1. The de re connection is lacking. By contrast, there is no reason why Newman 1’s contemporaries could not truthfully utter ‘Some philosophers of the late twentieth century had a name for you, and using that name, they said about you that you were not knowable by them (that you would be born in the twenty-second century, etc.)’. They might add, ‘Of course, they did not know (or even believe) that they were talking about you—how could they?—but you are the one they were talking about’. The case of Tom and the dean is clearer. The analogue of Donnellan’s remark is plainly incorrect, for both belief and assertion. Worse, since Donnellan also endorses DTa , he must also deny that John said that the man he had sent the day before was honest. But surely that is exactly what John did say. The Donnellan–Sosa test does not conclusively settle all such questions. There are some very hard cases. Inevitably individual intuitions in particular applications of the test will sometimes clash. But they often converge, or tend to converge, as in the case of Kevin and the shortest spy. It is fair to say that intuition in applying the Donnellan– Sosa test is not squarely on Donnellan’s side. In many cases—especially cases of attributive use where there is also an epistemically ‘real’ (e.g. causal) connection—it seems clear that intuition is squarely on the other side. (See again n. 29.) Thus, in an article coincidentally published in the same year as Donnellan’s article in which he proVers the Donnellan–Sosa test, Searle says: if I know the sheriV said ‘attributively,’ ‘Smith’s murderer is insane’ and I know Jones is Smith’s murderer I might indeed tell Jones, ‘Jones, the sheriV believes you are insane’, or even report, ‘About Jones, the sheriV believes he is insane’. Furthermore even where I know that Jones is not Smith’s murderer and I know that Ralph said ‘referentially’ ‘Smith’s murderer is insane’, and I know he had Jones in mind, I can still report his speech act by saying, ‘Ralph said that Smith’s murderer is insane’, for he did indeed say just that. (Searle 1979: 207)

V. A correct characterization of Donnellan’s distinction remains neutral with regard to the theses AC , DT , and SS. And this requires that the distinction be given in overtly pragmatic terms. As noted, Kripke has made the tantalizing claim that one distinction characterized in just this way covers the referential/attributive distinction and applies more generally to proper names and other non-descriptive terms. We saw in section II 30 Similar remarks, explicitly echoing Donnellan, occur in Soames (1994: 165).

252

II. Referential–Attributive Distinction

above that one of Kripke’s arguments in this connection is Xawed in that he did not provide an example of a genuinely attributive use of a proper name, and that he misclassiWes our routine, everyday uses of names as attributive rather than as referential. But we also saw that attributive uses of names do genuinely exist. Let us consider Kripke’s more general distinction in greater detail. He writes: In a given idiolect, the semantic referent of a designator (without indexicals) is given by a general intention of the speaker to refer to a certain object whenever the designator is used. The speaker’s referent is given by a speciWc intention, on a given occasion, to refer to a certain object. If the speaker believes that the object he wants to talk about, on a given occasion, fulWlls the conditions for being the semantic referent, then he believes that there is no clash between his general intentions and his speciWc intentions. My hypothesis is that Donnellan’s referential/attributive distinction should be generalized in this light. For the speaker, on a given occasion, may believe that his speciWc intention coincides with his general intention for one of two reasons. In one case (the ‘simple’ case), his speciWc intention is simply to refer to the semantic referent: that is, his speciWc intention is simply his general semantic intention. (For example, he uses ‘Jones’ as a name of Jones—elaborate this according to your favorite theory of proper names—and, on this occasion, simply wishes to use ‘Jones’ to refer to Jones.) Alternatively—the ‘complex’ case—he has a speciWc intention, which is distinct from his general intention, but which he believes, as a matter of fact, to determine the same object as the one determined by his general intention. (For example, he wishes to refer to the man ‘over there’ but believes that he is Jones.) In the ‘simple’ case, the speaker’s referent is, by deWnition, the semantic referent. In the ‘complex’ case, they may coincide, if the speaker’s belief is correct, but they need not. (The man ‘over there’ may be Smith and not Jones.) To anticipate, my hypothesis will be that Donnellan’s ‘attributive’ use is nothing but the ‘simple’ case, specialized to deWnite descriptions, and that the ‘referential’ use is, similarly, the ‘complex’ case. (Kripke 1977: 15)

One discerns in this passage the source (or at least a source) of Kripke’s generalizing Donnellan’s distinction into a conceptually separate distinction. He explicitly catalogues what seems a perfectly ordinary use of the name ‘Jones’ as a ‘simple’ case, and hence, on his proposal for generalizing Donnellan’s distinction, as a generalized attributive, rather than a generalized referential, use. But it is not clear from Kripke’s wording that such uses really exemplify the simple rather than the complex case. I have a ‘general’ intention to use the name ‘Donnellan’ generally as a name for Keith Donnellan. On a particular occasion when I use the name, I also have a ‘speciWc’ intention to refer to Donnellan by my use of the name (as opposed to a speciWc intention to refer to Kripke, or to the man ‘over there’, etc.). Are these intentions of mine the same intention, or are they diVerent? How is one supposed to tell? Must I be conceiving of Donnellan as he whom I generally mean by the name in my speciWc intention in order for it to be the same as my general intention? If so, then my general intention is an intention generally to mean by the name he whom I generally mean by the name. How can such an intention succeed in determining a semantic referent for the name, as Kripke claims? Kripke evidently presupposes that the intentions are one and the same. But how can a standing intention generally to do such-and-such be strictly the very same intention as

Salmon, The Good, the Bad, and the Ugly

253

an occurrent intention on a particular occasion to do such-and-such on that occasion? Or is the relevant intention supposed to be not an intention to refer to Donnellan generally by one’s use of the name, nor to do so now, nor to do so at time t, nor to do so sometime or other, but simply to do so (period)? Do we have temporally nonspeciWc intentions? Can intentions even be temporally nonspeciWc in this way? One ought to feel uneasy, maybe even annoyed, with these questions, at least in the present context—much as we do when the philosophically uneducated ask for the sound of one hand clapping, or when the philosophically miseducated make equally ridiculous demands. It is preferable to minimize the extent to which the legitimacy and intelligibility of Donnellan’s distinction is made to depend on the identiWcation and diVerentiation of intentions in such contexts. More importantly, if commonplace uses of ordinary names fall under Kripke’s notion of a simple case (as he evidently believes), then as noted above, his distinction between simple and complex cases fails to generalize Donnellan’s distinction for deWnite descriptions in the most natural and plausible manner. Notice, by contrast, that typical uses of names whose reference was Wxed by an attributive use of a deWnite description seem more clearly to fall squarely within the parameters of what Kripke means by ‘the simple case’. We have the general intention to refer by ‘Newman 1’ to the Wrst child to be born in the twenty-second century. Kripke would say that our speciWc intention in connection with our use of the name on a given occasion is this very same intention. Either that, or else barnyard-variety attributive uses of deWnite descriptions will exemplify the complex case, contrary to Kripke’s intent. One may raise skeptical questions in this connection—like those I have already posed, and more—but it is clear, or at least relatively clear, that in this case, there is no potentially conXicting, non-semantic intention to refer speciWcally to this person or that, in addition to our pre-set semantic intention to refer to whoever is born Wrst on New Year’s Day, ad 2100 (or is it ad 2101?) The problem with Kripke’s characterization is not that it is oV the wall. It is simply oV target. It is at least arguable, as we have seen above, that there is (and there will be) no one for whom we now have a de re intention to use ‘Newman 1’ as a name. This observation—and not the diVerentiation of intentions—provides the key to Donnellan’s distinction and its most natural generalization to other sorts of terms. It will not do, however, to say that a use of a term is referential whenever there is an occurrent speciWc de re intention concerning some particular person or thing, to refer to him/her/it by the term. As we have seen, the homicide detective investigating Smith’s murder may be said not only to make de re assertions but also to have de re beliefs concerning the murderer (e.g. that he is insane), even at the earliest stage of the investigation before the detective is in a position to use the description referentially. (Recall the Donnellan– Sosa test and the quote from Searle in section IV above.) Intention is suYciently like belief that it would seem correct to say that, even at that stage, the detective also has the speciWc de re intention, concerning Smith’s murderer, to refer to him/her by the phrase ‘Smith’s murderer’. The required de re connection has already been established.

254

II. Referential–Attributive Distinction

The same phenomenon can arise with proper names. A case in point may be the name ‘Deep Throat’, coined by Washington Post reporters Bob Woodward and Carl Bernstein for a highly placed, conWdential source in their famous investigation of the Watergate scandal. (This example is mentioned by Donnellan 1978: 38–9.) It seems correct to say of typical readers or viewers of the work, All the President’s Men, that there is indeed someone whom they mean, de re, by ‘Deep Throat’, and about whom they are expressing their views when discussing the Post’s coverage of Watergate—even though the readers do not know, and may even have no guess concerning, that person’s actual identity. Here again, the Donnellan–Sosa test seems to bear this out. On a given occasion of use, the typical reader has a speciWc intention to refer to this same person. In this respect, ‘Deep Throat’ is more like ‘Jack the Ripper’ (or ‘Smith’s murderer’) than ‘Newman 1’. A typical reader who does not have even a hunch as to Deep Throat’s identity, however, does not have anyone in mind, in Donnellan’s sense, when reading and using the name. That is, the reader does not have anyone in mind suYciently to be able to use a deWnite description referentially. A similar situation arises when one encounters someone’s name for the Wrst time without being adequately introduced (even if only in absentia) to the person so named—say by looking at a new classenrollment list or a luggage identiWcation tag. If one then uses the name to state something about the person so named, while still having no idea whose name it is, one makes a de re assertion and expresses a de re belief. (‘This belongs to one Byron Mallone. Mr Mallone, whoever he is, has traveled to Israel and was very recently around someone who smokes cigars.’) Such uses are not referential. Donnellan’s notion of having an individual in mind—the notion of having-in-mind that is a requirement for using a deWnite description referentially—seems to fall somewhere closer to the notion of knowing who someone is, or perhaps to that of having an opinion as to who someone is (the doxastic analogue of the epistemic notion of knowing-who), than to the distinct notion of having de re beliefs, intentions, or other cognitive attitudes concerning the person in question. To be sure, the notions of having someone in mind and of knowing-who are not the same. One can use a description referentially for someone while having no opinion, let alone knowledge, concerning that person’s identity, in the usual sense. But the notions seem connected or similar, more so than the notion of having-in-mind is similar to that of having de re attitudes. I believe that the relevant notion of having-in-mind, like the notions of knowing-who and having-an-opinion-as-to-who, is best thought of as a cognitive relation between a cognizer and a content appropriate to a singular term.31 In nearly all of the cases 31 Cf. Salmon (1987: 213 n. 17, 214 n. 19). I now think that, in so far as there is a diVerence (however subtle), I should have used the phrase ‘knows what F ’ instead of ‘knows which F’ for the generic notion of which knowing-who is a species. It seems likely that the notion of knowing-who is, at least to some extent, interest relative; someone whose epistemic or cognitive situation remains unchanged might be correctly described in one context (setting one range of interests) as knowing who someone else is, and in another context as not knowing who the other person is. The notion of having someone in mind does not seem interest relative—at least not in exactly the same way and to exactly the same extent as is the notion of knowing-who. It is possible that Donnellan’s notion of having an individual in mind is logically related to the semantic content of the phrase

Salmon, The Good, the Bad, and the Ugly

255

discussed by Donnellan and others in connection with referential use, what one has in mind is an individual person or object that the description is supposed to Wt. Donnellan (1966: 290–1) provides an example in which a speaker uses the description ‘the king’ referentially for someone whom he believes to be a usurper, but whose claim to the throne is known to be unquestioned by the people with whom he is conversing. In such a case, the speaker does not believe the description used actually Wts the object he/ she has in mind, but is adopting the pretense that it does. In still other cases, one might instead have in mind not a particular individual but what Church (following Carnap) calls an individual concept, that is, a descriptive content appropriate to a deWnite description. The latter type of situation is characteristic of an attributive use. When the homicide detective says ‘Smith’s murderer is insane’, he has no one in particular in mind; what he has in mind is the individual concept that person, whoever he or she is, who single-handedly murdered Smith. In pointing out the existence of the referential use of a deWnite description, Donnellan highlighted the possibility of using a description while having a particular person or object in mind that the description may not denote, in Russell’s sense. There is an analogous possibility, which has not been generally recognized, of using a deWnite description while having a particular individual concept in mind that the description does not semantically express. It very often happens that what a speaker has primarily in mind in using a deWnite description is neither the concept conventionally contained in the description used nor an individual that the description is supposed to Wt, but a diVerent description (or at least a diVerent descriptive content), one that the speaker can only use attributively. To modify Donnellan’s king example slightly, consider that the speaker does not have any idea who the usurper is, believing only that the rightful king has been wrongly deposed by someone or other. What he primarily thinks when he says ‘the king’ is: whoever it is that is taken to be king.32 Or again, suppose that the investigating detective is completely convinced that Johnson was murdered by the same culprit, so far still unidentiWed, who committed the recent, very similar murder of Smith. The homicide department has no suspects, no witnesses, and no leads in either case; the detective’s Wrm belief is based entirely on the common MO. When the detective uses the phrase ‘Smith’s murderer’ at the scene of the later crime, he primarily means: the guy, whoever he is, who murdered Johnson. The detective does not actually have the murderer in mind, in the relevant sense; otherwise, he could use the phrase referentially. Instead the detective thinks of Johnson’s murderer by description. Such uses as these are a kind of pseudo or mock referential use. In a sense, the mock referential use is what you get when you cross referential with attributive. In many such uses, there is even someone or something that the speaker intends (de re) to refer to by the description. The only thing preventing the use from being bona-Wde referential is ‘knows what F ’—or more likely, ‘knows which F ’—with respect to one sort of context or range of interests. (I do not feel conWdent about how to resolve these issues.) 32 This modiWcation of Donnellan’s original example is due to Ilhan Inan.

256

II. Referential–Attributive Distinction

the exact nature of the user’s cognitive access to the individual. In this respect, mock referential uses are more attributive than referential. But in other respects, they are so much like genuine referential uses that they ought to have been included in previous discussions of the referential use, and ought to be included in subsequent discussions. A referential use of a deWnite description is Good or Bad, according as the individual that the speaker has in mind is, or is not, denoted (in Russell’s sense) by the description. A mock referential use of a deWnite description is either Good or Bad, depending on whether the individual concept that the speaker has in mind is, or is not, coextensive with the concept conventionally expressed by the description. Let us say that a Good mock referential use is a Pretty use, and that a Bad mock referential use is Ugly. Recall in this connection that Donnellan allows that Russell’s theory of descriptions may give the correct analysis for attributive uses of deWnite descriptions, though not for referential uses. Whatever reasons Donnellan may have for withholding Russell’s analysis from referential uses seem to extend straightforwardly to mock referential uses. When the detective says ‘Smith’s murderer left a smudge print here’ at the scene of Johnson’s murder, someone of Donnellan’s ilk might argue that, in some sense, the detective will have stated something true as long as the smudge was made by Johnson’s murderer, whether or not he also murdered Smith. More revealing, such a philosopher would argue further that, in some sense, the detective will have stated something false as long as Johnson’s murderer did not make the relevant smudge print—even if the detective’s belief that Smith and Johnson were murdered by the same person is incorrect and, purely by happenstance, Smith’s murderer coincidentally left the smudge there sometime prior to Johnson’s murder. An Ugly use of a deWnite description is a very close facsimile of a Bad referential use. We have seen that Kripke’s attempt at generalizing Donnellan’s distinction casts the ordinary use of proper names on the attributive side when they are more at home on the referential side. There is a further diYculty, but opposite in kind. Other uses of terms are miscast as referential when they are attributive, or at least more attributive than referential. This is due to the fact that Kripke’s notion of the complex case does not include as a necessary condition that the speaker have a particular someone in mind, in the relevant sense. It is suYcient that the speaker have an occurrent speciWc intention distinct from, and in addition to, his/her standing general intention. Mock referential uses satisfy this condition. On a particular occasion when the homicide detective uses the phrase ‘Johnson’s murderer’, he may have the occurrent speciWc intention to refer to the repeat murderer responsible for the deaths of both Smith and Johnson. This is clearly diVerent from his background semantic intention always to use the phrase with its usual English meaning. His use therefore exempliWes the complex case. The two intentions may even conXict. If Johnson’s murderer is a copycat, the detective’s use of the phrase ‘Johnson’s murderer’ is Ugly. It does not Wt the paradigm for a referential use, since there is still no one whom the detective has in mind. Otherwise, he should also be able to use ‘Smith’s murderer’ referentially. But he cannot (even though he has various de re beliefs concerning Smith’s murderer, e.g. that he murdered Johnson).

Salmon, The Good, the Bad, and the Ugly

257

VI. Kripke distinguishes between standing (‘general’) intentions always to use a term in such-and-such a way and occurrent (‘speciWc’) intentions to use the term in such-and-such a way on a particular occasion, saying that semantic reference is given by the Wrst kind of intention and speaker reference by the second. This distinction among intentions seems an excessively delicate basis for the comparatively Wrm distinctions between semantic reference and speaker reference, and between referential and attributive use. If I have the occurrent intention to use ‘Smith’s murderer’ to mean the man ‘over there’ on this occasion, because I genuinely believe that man murdered Smith, do I not also form a standing intention always to use the phrase for that man? Conversely, it seems rather likely that standing intentions in connection with our use of language typically (if not invariably) give rise to occurrent intentions on particular occasions. This may even be built into the notion of a standing state. I believe it may be more helpful to replace Kripke’s distinction between standing and occurrent linguistic intentions with a diVerent one: the distinction between linguistic intentions that are purely semantic in nature and those that are not. We who speak English intend to use our words generally with their conventional meanings. Our knowledge of what those words mean allows us to form more speciWc semantic intentions. Thus I intend to use the word ‘guitarist’ in whatever is its usual English sense. Given my knowledge of what the word means in English, I form the additional intention to use the word speciWcally as a term for one who plays the guitar. The Wrst is a general background intention, one that speciWes the intended meaning only as whatever the term means in English; the second identiWes a particular meaning for the term. Both are purely semantic intentions, in that they are meant to govern not merely which individuals the word ‘guitarist’ happens to apply to, but what the word applies to as a matter of the semantics of my idiolect. Indeed, they are also meant to govern what the word is to mean. The second intention may be termed an identifying semantic intention. If I believe that all and only guitarists keep their Wngernails short on one hand and longer on the other, I may form the additional linguistic intention to use the word to apply to individuals of exactly that class. Such an intention would also be an identifying semantic intention, in the sense I intend, since it identiWes a particular extension for ‘guitarist’ (in contrast with an intention to use the word to apply to exactly the things to which it correctly applies in English). But it would not be a purely semantic intention; it depends on an extra-linguistic belief of mine, one which is (and which I recognize to be) non-semantic in nature. In using a singular term the speaker typically has a purely semantic, identifying intention of the form d By my use of this term, I intend to refer to ae . And it is arguably this intention, rather than some non-semantic standing linguistic intention, that governs semantic reference for the term in the speaker’s idiolect. The meta-linguistic variable ‘a’ here may be a stand-in for a deWnite description—or if the quasi-quotation

258

II. Referential–Attributive Distinction

marks are interpreted as content-quotation marks, rather than as syntactic quotation marks, the ‘a’ may be a stand-in for an individual concept. For example, one may have the purely semantic intention expressed by ‘By my use of the phrase ‘‘Smith’s murderer’’, I intend to refer to whoever single-handedly murdered Smith.’ In this case, the intention is a product of more fundamental identifying semantic intentions: to use ‘Smith’ to mean Smith, to use ‘murderer’ as a term for murderers, etc. But the a might instead stand in for a proper name of a person or object. In the case of a typical proper name, the relevant purely semantic, identifying intention is ‘singular’ or de re: ‘By my use of ‘‘Smith’’, I intend to refer to Smith [that very guy].’ By contrast, in the case of a name whose reference is Wxed by an attributive use of a deWnite description, the purely semantic, identifying intention is ‘general’ or de dicto: ‘By my use of ‘‘Newman 1’’, I intend to refer to whoever is born Wrst in the twenty-second century.’ This is not the same as an intention to use ‘Newman 1’ speciWcally as a name for that particular future individual, Newman 1. The Donnellan–Sosa test would seem to indicate that we do not have the latter intention. No de re connection has been established.33 Before attempting to extend Donnellan’s distinction to proper names and other terms in a natural and plausible way, a further point must be made. Our use of a particular term is often accompanied by a plurality of identifying semantic intentions, each of the form d By my use of this term, I intend to refer to ae . It may happen that the speaker regards one or more of these as essential to what he or she means, and the rest as so much window dressing, mere accoutrement. Suppose the speaker is asked, d Consider a hypothetical scenario in which your intention to refer to a by the term and your separate intention to refer to b conXict, because these are diVerent individuals. In such a case, which do you mean by your use of the term?e In reply the speaker may cite one intention as the superseding, decisive intention. We may call this the speaker’s primary linguistic intention. One may come close to generalizing Donnellan’s distinction, then, by invoking the various notions of purely semantic intentions, primary linguistic intentions, and identifying semantic intentions. Let us distinguish between generalized referential and generalized attributive uses as follows. In a g-attributive use of a singular term, the speaker has a primary, identifying, purely semantic intention of the form d By my use of this term, I intend to refer to ae , where a is a deWnite description. This intention is general, as opposed to singular; it is a de dicto intention.34 Further, the speaker does not have in addition a supplementary primary linguistic intention of the form d By my use of this term, I intend to refer to be that is not purely semantic in nature, or where b is a directly referential, Millian term (e.g. a name) for an individual person or object that the speaker ‘has in mind’, in the relevant sense. Here there is no potential conXict 33 Cf. also Salmon (1993a; 1993b: 99–100 n. 27). 34 I do not mean by this to rule out the possibility that the intention is also de re or singular. Some intentions, beliefs, etc. are both de dicto and de re. The detective’s belief that whoever single-handedly murdered Smith is insane, e.g. is de re (singular) with regard to Smith and de dicto (general) with regard to his murderer. Cf. Salmon (1982: 44–5 n. 3).

Salmon, The Good, the Bad, and the Ugly

259

(from the point of view of pure semantics) with the primary de dicto linguistic intention, and speaker reference is therefore governed by that purely semantic intention. In a g-referential use of a term, by contrast, the speaker has a primary linguistic intention (either purely semantic or not) of the form d By my use of this term, I intend to refer to ae , where this time a is a directly referential term for an individual person or object (rather than a deWnite description), one whom the speaker ‘has in mind’ in forming this intention. The speaker’s primary linguistic intention is a de re intention concerning the person or object for which the speaker is using the term g-referentially. In this case, speaker reference may be governed by both this primary linguistic intention and a separate purely semantic intention. It may even happen that the speaker inadvertently refers simultaneously to two (or more) entities by a single use of a term. This distinction aims at capturing conceptually critical elements of the referential and the attributive use of a deWnite description. The generalized distinction has several noteworthy features. First, it seems clear that a use of a deWnite description is referential if it is g-referential, and attributive if g-attributive. The distinction between g-referential and g-attributive use is mutually exclusive, or nearly enough so; no use of a term can be both g-referential and g-attributive except (perhaps) where a speaker has linguistic meta-intentions that create a duality of use by giving equal weight to conXicting linguistic intentions, neither one of which supersedes the other. Donnellan seems to have originally intended his more restrictive distinction also to be mutually exclusive (at least to this same extent). Kripke (1977: 8) says that he does not regard Donnellan’s distinction as exclusive. In his example of a use that might be regarded as simultaneously partially referential and partially attributive (1977: 25–6 n. 28), a speaker utters ‘Smith’s murderer is insane’ based both on the grisly condition of Smith’s body and also on the peculiar behavior of the person whom the speaker has in mind and is observing (believing him to be Smith’s murderer), where ‘neither consideration would have suYced by itself, but they suYce jointly’. I believe Donnellan would probably say that this is simply a referential use and not attributive, and I do not see a compelling reason to dispute this verdict. Indeed, if Kripke’s case is to be regarded as somehow involving an attributive use in combination with a referential use, it raises the specter that most (or at least a great many) uses that are generally taken to be referential and not attributive will turn out to be combined referential–attributive. This would run counter to how we ordinarily conceive of the distinction.35 While the generalized distinction is exclusive, it is not 35 In so far as it is desirable to allow for such a thing as a combined referential–attributive use, the generalization proposed here might be adjusted to accommodate Kripke’s intuitions about the case (perhaps by deleting the condition on g-attributive use that there be no primary purely semantic intention concerning someone or something that the speaker has in mind and/or the condition that there not be a second, non-semantic intention). Kripke’s attempt at generalizing Donnellan’s distinction—in terms of the identiWcation or diVerentiation of ‘general’ and ‘speciWc’ linguistic intentions and the distinction between ‘simple’ and ‘complex’ cases—does not straightforwardly allow for combined uses. Furthermore, since strict identity is a clear-cut, all-or-nothing aVair, it is by no means obvious how to adjust Kripke’s distinction to accommodate combined (partially simple, partially complex) uses—and especially how to do so without making most referential uses into combined uses. Perhaps this can be done.

260

II. Referential–Attributive Distinction

exhaustive. Many uses of deWnite descriptions are neither g-referential nor g-attributive. Mock referential uses, for example—which are common in ordinary speech— have elements of both referential use and attributive use. For that very reason they do not Wt the paradigm for either use, as Donnellan set out the original distinction.36 More interestingly, uses of proper names whose reference has been Wxed by an attributive use of a deWnite description are typically g-attributive. Some other unusual uses of names are also g-attributive, as in the ‘Deep Throat’ example. But commonplace uses of ordinary names are typically g-referential. Perhaps most importantly, one of Kripke’s principal criticisms of Donnellan is upheld. Especially telling, and to the point, is Kripke’s observation—made forcefully with the aid of a variety of postulated languages (1977: 15–17)—that the pragmatic phenomena involving speaker reference, speaker assertion, and the like adduced by Donnellan in connection with the referential use are no evidence for the thesis of semantic signiWcance. The generalized distinction is neutral regarding controversial theses like SS and SS 0 . Using that characterization of the referential–attributive distinction, it is possible to provide an account of the referential use, and more generally of the g-referential use, that attributes to it no special semantic signiWcance, while accommodating, and even predicting, the circumstances and frequency of its occurrence in everyday speech. The latter therefore has no bearing on the question of semantic signiWcance. 36 Neale (1990: 202–3) describes a case similar to the detective’s use of ‘Johnson’s murderer’, as set out above, declaring that it is attributive, by Donnellan’s criterion, and not referential. Inan independently gave a similar example, judging it to be attributive, and concluding, more carefully, that Donnellan’s explicit criterion for attributive use fails to capture the intent. An informal survey of many of the cognoscenti showed that, aside from one explicit abstention, all nine remaining respondents unanimously regard the detective’s use of ‘Johnson’s murderer’ as: (a) attributive; and (b) not referential. I continue to believe that the use is neither referential nor attributive, as Donnellan intended these terms. It is a mock referential use in which the detective primarily intends whoever single-handedly murdered both Smith and Johnson, rather than whoever single-handedly murdered Johnson (whether or not he also single-handedly murdered Smith). I believe Donnellan would therefore withhold Russell’s analysis from it, by contrast with a genuine attributive use. (Two respondents, Anthony Brueckner and Genoveva Marti, supplemented their vote with unsolicited remarks showing a sensitivity to the sort of considerations raised in the text in connection with Ugly uses, like the detective’s use of ‘Smith’s murderer’ for whoever murdered Johnson, where Johnson’s murderer turns out to be a copycat.) Attributive or not, as already noted, the fact that the use is clearly not referential even though it exempliWes the ‘complex case’ shows that Kripke’s attempt to generalize Donnellan’s distinction for deWnite descriptions does not get matters exactly right.

6 Descriptive Indexicals and Indexical Descriptions GeoVrey Nunberg

1. Explaining the Referential–Attributive Distinction As Donnellan (1966) and many others have pointed out, a sentence like (1) has two readings: (1)

The person who’s parked in front of the restaurant is in a hurry.

On the attributive reading, the description the person who’s parked in front of the restaurant is interpreted as a quantiWer: it says that the unique person who is parked in front of the restaurant is in a hurry, with no implication that the speaker has a particular person in mind—maybe she simply inferred his haste from his choice of a parking spot. On the referential reading, the description picks out a particular person—say Jones, who is waiting for his car at the valet parking station with visible impatience. The question is, is (1) semantically ambiguous? Not if you accept the Russellian view, which holds that the description in (1) is always interpreted as a quantiWer, with the referential interpretation arising via a conversational implicature that arises from the hearer’s assumption that the speaker is trying to say something relevant and has a particular person in mind. In that case the utterance does not literally say anything about Jones, though the speaker may succeed in identifying him by uttering it.1 For referentialists, on the other hand, (1) is ambiguous. On the referential reading, the person who’s parked in front of the restaurant functions basically the way an indexical does. This is the eVect achieved by Kaplan’s ‘Dthat’ (see Kaplan 1979), an operator that turns descriptive content into a function from contexts to individuals. In that case For comments on this chapter and on earlier versions of this work, I am grateful to Franc¸ois Recanati, Cleo Condoravdi, and the two anonymous reviewers of this chapter. 1 This is by no means the only possible view of the relation between the readings. Bezuidenhout (1997) and Recanati (1993) have suggested that both referential and attributive readings are generated pragmatically from an under-speciWed semantic representation. For my purposes, though, the only important distinction is between those who argue that referential uses contain something like a hidden indexical in their semantic representations and those who do not—on this cut of things, both Bezuidenhout and Recanati would Wgure in the latter group.

262

II. Referential–Attributive Distinction

an utterance of (1) with the description used referentially is semantically identical to an utterance of a sentence like He is in a hurry—both utterances literally express a singular proposition, in which the content of the referring expression plays no role.2

2. The Argument from Misdirection Is this an empirical dispute? It feels as if it should be, but there does not seem to be a lot of hard empirical evidence that bears on it, and the arguments that people have oVered have not been conclusive, to put it mildly. In introducing the distinction, Donnellan leaned heavily on a line of argument that Neale (1990) calls the argument from misdescription and Recanati (1993) calls the argument from improper use: the observation that a speaker can sometimes succeed in identifying the individual she has in mind with a description like the person who is parked in front of the restaurant even if it should turn out that the description does not actually apply to him. So suppose that Jones is the visibly impatient person who is holding the parking stub for the car that is parked in front of the restaurant, but that Jones is in fact merely getting the car for his friend Smith, who is inside paying the bill in an unhurried way. Even so, Donnellan asserts, we might use the description to pick out Jones. In this sense the description is regarded as analogous to a demonstrative accompanied by an inaccurate gesture, which might nonetheless enable someone to pick out the object that the speaker has in mind. Whereas if the sentence merely says that there is a unique person who is parked in front of the restaurant and who is in a hurry, it is held to be mysterious as to how we might use it to get to someone who does not satisfy that description. The argument is not very persuasive. For one thing, it rests on people’s intuitions about what is actually said by an utterance of (1), which is never a very sound basis for analysis. (The more you listen to people’s intuitive judgments about the ‘literal meanings’ of sentences, the more you come to think that the notion is more jurisdprudential than linguistic.) And in any event, people do not always operate the way Donnellan assumes they do when it comes to interpreting these utterances. True, we often allow a speaker a certain descriptive latitude in order to get on with the conversation, but we are not obliged to do so, particularly if the content of the description is controversial. Suppose that a White House press oYcer says, ‘The man who won the majority of votes in Florida will be visiting Tallahassee on Sunday’, with the intention of informing 2 For the present purposes, I shall assume that direct referentiality (or as I will be calling it, direct interpretation) is a semantic property of indexical expressions as such. In fact Roberts (2002) argues that the direct referentiality of demonstratives, at least, follows from their dependence on a completing demonstration (where ‘demonstration’ is construed broadly). She notes that a Kaplan-type (1989a) analysis of demonstratives cannot deal with various problems raised by Heim (1985), or with the use of demonstratives that refer to discourse antecedents, such as in She ignored his warning, and that got her in trouble. It is not clear, though, how or whether this analysis could be generalized to ‘pure’ indexicals like I or today.

Nunberg, Descriptive Indexicals

263

the public about President Bush’s schedule. My own guess would be that a good number of Americans would respond to that by saying, ‘No, he’ll be spending the weekend in New York with Tipper.’ And from the point of view of a referentialist, it is hard to explain why anyone would have this reaction, since on his view the speaker has not really said that Bush won Florida.3 Then, too, as both Kripke (1977) and Neale (1990) observe, these cases of misdescription are not limited to the uses of deWnite descriptions. Kripke gives the example of someone who points at Jones under the misapprehension that he is Smith, and it is easy enough to construct analogous examples using quantiWers. Imagine an out-of-it dad who says at his 11-year-old daughter’s birthday party ‘Every girl who is wearing a button with a picture of Christina Aguilera is attending my daughter’s party’ when in fact the invitees are all wearing badges with pictures of Britney Spears. Even so, a hearer could doubtless Wgure out who’s being talked about, though clearly there is no question of saying that the expression Every girl who is wearing a button with a picture of Christina Aguilera contains a concealed demonstrative. In fact, as Wettstein (1981), Recanati (1993), and others have pointed out, the argument from misdescription is not really germane to the larger issue of whether sentences like (1) are semantically ambiguous. Wettstein maintains that it is not essential to the referentialist view to accept that referentially used descriptions can function semantically to pick out an individual who does not satisfy them. If (1) is used in a context in which Jones is holding Smith’s parking stub, you could say that the description literally picks out Smith, while noting that the hearer can nonetheless recover the speaker’s intended reference, and use that as a basis for further action if it is appropriate. In short, the argument from misdescription hardly gives us any grounds for resolving the controversy one way or the other.

3. The Argument from Incompleteness This takes me to the second argument that has been appealed to here, which has its origin in Strawson’s observation that we can use a noun-phrase like the table to identify a particular table even though it is not the only table in the world. On a Russellian analysis, the argument goes, this is puzzling unless we can provide a completing sense to distinguish the table from all others, and that creates a number of problems—how do we know which completing sense is pertinent, for example? Whereas if we treat the deWnite description the table on the model of a demonstrative description like that table, we can understand how it can be used to identify a particular salient object. 3 Or sometimes we can nod to the intended reference without endorsing the content of the description. In the movie Quo Vadis, there’s a scene where a Roman general (Robert Walker) takes a Christian hostage (Deborah Kerr) to a banquet at Nero’s palace. ‘Have you ever seen your god up close?’ he asks her? ‘No,’ she answers, ‘I have never seen Nero up close’, deftly demurring from the content of the description.

264

II. Referential–Attributive Distinction

I confess that I have never been able to make much sense of this argument. It seems plain that when someone in an oYce says The table is covered with books, we interpret the description the table as applying to the unique table in the domain of things that the speaker is talking about. I do not mean to suggest that we have a wholly adequate account of domains of discourse, but this ground has been covered in an extensive literature, particularly as regards the interpretations of deWnite noun-phrases.4 And whatever form that account takes, it will have to explain the use of ‘incomplete’ quantiWers, as well, as Neale (1990), Salmon (1991), and others have pointed out. For example, we often say things like Everyone left, where there is no question of proposing a demonstrative reading.5 What is more, it is hard to see how the hidden indexical analysis of referential descriptions is going to let us oV the hook in cases like this. It may be that the people who have taken this view have been misled by the familiar examples of referential descriptions—phrases like Donnellan’s the man with the martini or Strawson’s the table. These typically involve situations where the intended referent is salient in the immediate context, so that you could imagine referring to it with a demonstrative phrase like that man or that table. But ‘incomplete’ referential descriptions can be used to pick out remote things as well, where no demonstration would be possible, and in those cases it is hard to see how they could succeed in referring unless some uniqueness condition was satisWed. Say one of the Duke of Plaza-Toro’s footmen says to another The Duke is visiting Venice, using the deWnite description in a referential way. If it was not clear that there was a unique duke in the domain of discourse, how would it help to assume that the description contained a hidden indexical—how would you determine which duke the speaker had in mind without requiring a completing description or a suitable restriction on the domain of discourse?6 In short, the argument from incomplete descriptions simply has no bearing on the status of these expressions. 4 In this connection, see among many others Heim (1982), Kempson (1986), Lo¨bner (1987), Kadmon (1987), Neale (1990), Birner and Ward (1994), and Roberts (1995, 2002). Among other things, these sources all discuss the possibility that a proper analysis of deWnites requires criteria other than or in addition to uniqueness, such as familiarity. For our purposes, though, it does not matter which analysis of deWnites we adopt, so long as deWnite descriptions come out as quantiWers. 5 Ostertag (1998: 26) argues that this approach does not explain why descriptions like the record do not seem to be equivalent to descriptions like every record when used in incomplete contexts: ‘[A]n utterance of ‘‘the dog is barking’’ will generally be interpreted relative to a very small domain, while an utterance of ‘‘every dog’’ is barking will be assigned to a larger domain. . . . If the equivalence cannot be assumed, then the theory that treats ‘‘the F ’’ as a quantiWer . . . is not Russell’s theory of descriptions’. I have no brief to make for the classical Russellian theory of descriptions, but I do not see the justiWcation for Ostertag’s conclusion, since the presuppositional diVerence between ‘The dog is barking’ and ‘Every dog is barking’ seems to have a straightforward Gricean explanation. In general, we are obliged to couch our utterances in such a way as to accommodate as best we can the actual presuppositions of the context. For example, if it is common knowledge that Cleo is not an American citizen, I could felicitously say (i) but not (ii): (i) If Cleo were an American, she wouldn’t need a Green Card; (ii) If Cleo is an American, she doesn’t need a Green Card. It is not that (ii) would not be truthful, but in the circumstances it would be misleading, since it ignores what is commonly known. The diVerence between ‘The dog is barking’ and ‘Every dog is barking’ is perfectly analogous to this: while the version with every is consistent with the existence of a single dog, it is not a felicitous thing to say when it is common knowledge that there’s only one dog around. 6 See Roberts (2002) for reasons for assuming that a uniqueness presupposition is semantically associated with the uses of demonstratives.

Nunberg, Descriptive Indexicals

265

4. The Argument from Methodological Preliminaries Given the inconclusiveness of these arguments, it might seem that the status of referentially used descriptions is not really empirical at all.7 That seems to be the position of Kripke (1977), who argues that the issue should be thought of in methodological terms. He asks us to imagine a stipulated variety of English in which deWnite descriptions could have only quantiWer interpretations. Even in that case, he argues, we can see how referential readings could arise on Gricean principles. And for that reason, according to Grice’s ‘modiWed Occam’s razor’ principle (‘senses are not to be multiplied beyond necessity’), there is no ground for introducing a semantic ambiguity.

5. The Descriptive Uses of Indexicals I have no quarrel with the form of Kripke’s argument, but it is always more satisfying to have data to fall back on. In this section I shall discuss some linguistic observations which seem to favor the Russellian analysis of referential descriptions, but which up to now have not played any role in these discussions. The phenomenon I have in mind involves what I shall call the descriptive uses of indexicals (the term was suggested by Recanati (1993) to describe some examples I gave in Nunberg (1993) in the service of another point): (2) (3) (4)

Today is always the biggest party day of the year. (Pointing at a TV that is showing a lopsided Stanford–Berkeley basketball game) This is usually an exciting game, but not tonight. Condemned prisoner: I am traditionally allowed to order whatever I want for my last meal.

Unlike indexicals of the more frequently picked-over variety, the ones in (2)–(4) do not refer to individuals, but rather seem to contribute properties to the utterance interpretations. In (2), we take the interpretation of today as some property that is instantiated by the day of utterance—of being November 1, the day that fall midterms are over, the day following the homecoming game, or whatever. (Note though that it is not necessary that the hearer be able to identify the particular property the speaker has in mind—for his purposes, it might be suYcient to know that it is not a good day to try to go for a quiet drink at a bar near campus.) 7 I am ignoring here another set of arguments that people have oVered in favor of the thesis that sentences like (1) are ambiguous. These have to do with the interpretations of pronouns in examples like (i): The horse which I bet on won. Hans had foreseen it. As the argument goes, only if we treat the subject of the Wrst sentence in (i) as having a reading as a singular term can we explain the reading of the utterance in which Hans had foreseen that Secretariat would win without knowing that the speaker had bet on that horse. For an account of the deWciencies of this argument, see Heim (1991).

266

II. Referential–Attributive Distinction

Example (3) works in the same way. The demonstrative this cannot refer to the very contest the speaker is pointing at, which is plainly boring; instead it contributes the property of being a Stanford–Berkeley game. And in (4), the pronoun I doesn’t refer to the speaker, since obviously there couldn’t be any traditions that deal speciWcally with his last meal; instead it refers to the role he exempliWes. In calling these uses of indexicals ‘descriptive’, I do not want to give the impression that their interpretations are equivalent to nonindexical descriptions like the day of our conversation or the game on TV on their normal Russellian readings. To see the diVerence, we have only to recast the sentences so that they contain a modal or counterfactual context: (5) (6) (7) (8)

If we were talking to each other on November 2 instead of now, the day of our conversation would be the biggest party day of the year. If we were talking to each other on November 2 instead of now, today would be the biggest party day of the year. If you had turned to Channel Four as I asked, the game on the TV would be exciting. If you had turned to Channel Four as I asked, this game would be exciting.

The descriptions in (5) and (7) can be re-evaluated relative to the contexts established by the conditional, but the indexicals in (6) and (8) cannot: today is still anchored in the actual day of speaking, and this game is the one whose current episode is on the channel we are watching now, not the one playing on Channel Four. In this regard, the indexicals here are no diVerent from indexicals that pick out speciWc days or individuals, which have to be evaluated relative to the context of utterance. So whether or not these indexicals here are ‘directly referential’ in the strict sense, they are nonetheless what we can think of as directly interpreted, and for this reason they cannot be scoped by other operators.8 For the present purposes, we could describe the interpretations of the indexicals in (2)–(4) in any of several ways—as properties, as higher-order entities of some kind, or as under-speciWed discourse referents or constituents of abstract situations. I am willing to be agnostic about that—my chief interest is in the way these utterances contrast with utterances that contain descriptions. For example, suppose that tomorrow is also my daughter’s eleventh birthday. Then I could use the description Sophie’s eleventh birthday to refer to that day, in what would clearly be a referential use of the expression: (9)

Sophie’s eleventh birthday falls on a Tuesday.

8 The fact that indexicals are directly interpreted even when they contribute properties seems to pose a problem for the analysis of Bezuidenhout (1997), who oVers a symmetrical treatment of descriptions and indexicals: both, she says, begin with under-determined representations from which either ‘attributive’ or ‘referential’ readings can be derived. In that case, though, I am not sure how the diVerences illustrated in (5)–(8) could be explained.

Nunberg, Descriptive Indexicals

267

But I cannot substitute that description for the indexical in (2): (10)

?Sophie’s eleventh birthday is always the biggest party day of the year.

That is, the descriptor Sophie’s eleventh birthday can only apply to a unique date, which is inconsistent with the use of always. And by the same token, suppose that the prisoner who utters (4) has already had his head shaved in preparation for his execution. Then his guard could say (11) but not (12): (11) (12)

The prisoner with the shaved head gets the spaghetti alla Carbonara. ?The prisoner with the shaved head is traditionally allowed to order whatever he wants for his last meal. (6¼ 11)

That is, the description cannot contribute a property of the person it applies to—other than the very property that is explicit in its descriptive content, of course. Suppose that we accepted the referentialist line and said that the deWnite descriptions in these utterances contain Kaplan’s Dthat operator or something along those lines. Then why do they not have the same range of readings that true indexicals like I and that do? In each case you would have an expression containing a (real or crypto-) indexical that picks a certain individual out of the context and creates a singular proposition. But when you use a description like the prisoner with the shaved head, you are required to stop with the individual that the description applies to—you cannot go on to take one of his salient properties and make it the interpretation of the expression. On the other hand, if we assume a Russellian analysis of the deWnite description in a sentence like (11), then it is not odd that the mechanism that produces a descriptive reading does not come into play. The individual that the description applies to is not a constituent of the proposition, so you do not have access to any of his properties, apart from those that are made explicit by the description itself. On the face of things, this seems a strong empirical argument for saying that referentially used descriptions have a Russellian interpretation, and that the sentences that contain them have the same truth conditions as sentences containing attributively used descriptions.

6. A Pragmatic Explanation? Let me deal now with some potential objections to this argument. In the Wrst place, someone might say that the descriptive reading available for sentences like Tomorrow is always the biggest party day of the year is somehow a marginal or secondary phenomenon. This is the line taken by Recanati (1993), who suggests that the readings of indexicals in sentences like Today is always the biggest party day of the year are derived via a conversational implicature from literal readings in which they have their ‘normal’ references, referring to days, persons, or whatever.

268

II. Referential–Attributive Distinction

That approach runs into a number of diYculties, though. For one thing, it is hard to see how a sentence like Today is always the biggest party day of the year could have any kind of literal meaning in which today picks out a unique day, say November 1, 2000, and says of it that it is always the biggest party day of the year—there simply is not any coherent proposition that we could associate with such an interpretation.9 And for that reason, it is not possible to suspend the descriptive reading of indexicals in sentences like these, which is usually taken to be a requirement for postulating a conversational implicature in the Wrst place: (13) (14) (15) (16)

Some of the boys left—in fact, all of them did. My love is a red, red rose, and I mean that literally. Jack took oV his trousers and got into bed, but not in that order. ?Today is always the biggest party day of the year, but only if it’s November 1, 2000.

In any event, all of this is beside the point. Even if we did say that the reading in a sentence like (2) arose via a conversational implicature, we would be left with having to explain why a similar implicature does not arise when we replace the indexicals like I and today with referentially used descriptions like the prisoner with the shaved head and Sophie’s eleventh birthday. On any version of the referentialist hypothesis, after all, the indexicals and descriptions would have exactly the same literal or primary interpretations: each would contribute the individual it applies to.

7. The Maxim of Manner to the Rescue? At this point, a referentialist might suggest that the diVerence in the implicatures available for indexicals and referentially used descriptions follows somehow from the intervention of Grice’s maxim of manner. That is, while the two types of expressions have identical semantic import, there is something about the fact of choosing one or the other that either makes the descriptive reading available only for indexicals, or blocks it for referential descriptions. It is true that conversational implicatures can attach to the choice of a description rather than an indexical as a means of identifying something, as in (17)–(20): (17) (18)

(Spoken on April 9) The doctor can see you tomorrow. (Spoken on April 9) ?The doctor can see you on April 10.

9 It may be that we can interpret Recanati’s analysis as primarily a thesis on the way these utterances are processed. That might be true, but there is no reason to assume that the semantics of the sentence should recapitulate the psycholinguistic mechanisms that speakers bring into play to Wgure out what it means.

Nunberg, Descriptive Indexicals (19) (20)

269

(Spoken by an American) When we Americans travel, we like to have all the comforts of home. (Spoken by an American) ?When Americans travel, they like to have all the comforts of home.

Both pairs demonstrate a general conversational principle to the eVect that when an indexical can be used to identify something, we expect that it will be used. It is generally odd for someone speaking on April 9 to refer to the following day as ‘April 10’, when the use of tomorrow would pick out the same day, though certainly we would not claim that the sentence was false if she did. And similarly, it is a reasonable inference that the speaker of (20) is not an American, since if she were she would have been expected to use a noun-phrase containing we. But even so, the truth-value of the sentence does not stand or fall on the nationality of its speaker. And when the speaker Xouts this rule, that choice usually occasions a conversational implicature. When somebody refers to himself by his name or with a description like ‘your little brother’, we usually assume that he has some ulterior motive for doing so.10 We will see in the Wnal section that this principle does in fact have some bearing on explaining the descriptive interpretations of indexicals. By itself, though, it will not explain the diVerence we are interested in—not if indexicals and referential descriptions have the same type of interpretations, as referentialists claim. To invoke the maxim of manner, after all, we have to be able to assume that the speaker has deliberately chosen one form of expression over another that would have the same truthconditional meaning. But most referential uses of descriptions cannot be paraphrased by indexicals, particularly if the intended referent is not actually present or salient in the context. The footman’s announcement ‘The Duke of Plaza-Toro will arrive at noon’ has no equivalent using an indexical, so there could be no implicature associated with his choosing a description to refer to his master. (There might be an implicature associated with using one particular description rather than another, of course, but that would not be relevant to explaining the behavior of descriptions as a class.) What is more, when we look at cases where a referential description and an indexical phrase can both be used to pick out the same entity, it is hard to see how we could appeal to any diVerence in mode of expression to explain the diVerences in available interpretations. Suppose you have been visiting me at my house and need to pick up some cash on the way home. You could ask me either (21) or (22), to pretty much the same eVect: (21) (22)

Is there an ATM around here? Is there an ATM in the neighborhood?

But now suppose we are trying to get some cash in an unfamiliar neighborhood, and we notice an oV-track betting parlor. I could then say (23): 10 I say ‘we usually assume’ because there are a few situations where the rule that requires the use of indexicals is suspended—e.g. in news stories in which reporters are expected to refer to themselves as ‘this correspondent’.

270

II. Referential–Attributive Distinction (23)

There is usually a cash machine around here.

In this context, (23) means roughly the same as ‘there is usually a cash machine in the vicinity of an oV-track betting parlor’. But note that this interpretation would not be available if I said (24) instead: (24)

There is usually a cash machine in the neighborhood.

That is, the description the neighborhood cannot contribute a property of the place we are in, but only the place itself. But the only diVerence between the two utterances that could explain this fact is that in one case the place is identiWed descriptively, while in the other it is identiWed indexically (see Appendix). And if there is no semantic diVerence between these two modes of referring, what could give rise to a conversational implicature via the maxim of manner? In the light of examples like these, it is hard to imagine how pragmatics could come to the rescue of a hidden indexical analysis. At this point, then, the burden of proof is clearly on the referentialist: as Grice points out, the deWnitive test for a conversational implicature is a demonstration that it be capable of being worked out. If the diVerence between the available readings of indexicals and referentially used descriptions cannot be explained pragmatically, it must rest on a semantic diVerence between the two types of expressions. But where could this come from? I suppose a referentialist could still try to claim that the hidden indexical that Wgures in a referential use of the description the man with the martini was linguistically marked with a feature that arbitrarily blocks the availability of a descriptive reading but otherwise left the interpretation untouched. (We don’t know a lot about hidden indexicals, after all—nobody has ever seen one.) But this is a desperate expedient. It amounts to saying that there is no good explanation for the diVerence between the available readings of indexicals and referentially used descriptions—that is just how nature made them.

8. Where Do Descriptive Readings Come From? The observations I have made up to here are suYcient to show that deWnite descriptions simply do not behave like indexicals and demonstratives, and that the hiddenindexical analysis cannot be sustained. But this leaves us with an unanswered question: why do only demonstratives and indexicals permit descriptive readings, and permit them, it seems, in every language we look at? Or conversely, why can descriptions not have readings like these? To answer these questions I should say something about how demonstratives and indexicals acquire these readings. In Nunberg (1979, 1993), I described examples like these as instances of ‘deferred indexical reference’, a notion based on Quine’s ‘deferred ostension’. The idea is that reference in these cases proceeds via a two-stage process. We

Nunberg, Descriptive Indexicals

271

Wrst identify the index of a term—that is, either the demonstratum of a demonstrative or the contextual element that an indexical picks out in virtue of its linguistic meaning—and then proceed to identify the referent of the term, which is something that stands in a salient functional relation to the index. So if you point at a copy of a newspaper and say ‘That was bought by Rupert Murdoch for Wve hundred million dollars’, you can succeed in referring to a publisher in virtue of the functional relation between newspaper companies and their publications.11 I still think this is the right structural approach to explaining how references are identiWed in these cases of demonstrative metonymy—that is, cases where you point at one thing to identify something that stands in a relation of contiguity to it. But I have come to think that it is misleading to use the term ‘deferred reference’ to describe these examples. It seems just as straightforward to assume that the remote referents picked out by the demonstratives here are themselves made present in the context by a demonstration that accompanies the use of a demonstrative. (I am taking ‘demonstration’ in a broad way, as something implicit or explicit in the context or in the mention of the demonstrative term itself.) When you call to someone’s attention a copy of a newspaper, that is, you are also calling to his or her attention a newspaper publisher. And when you point at a picture of Carnap (to take an example used by Kaplan), you are also pointing at Carnap himself.12 In this sense there is nothing ‘deferred’ about these uses; they are more on the order of other cases of ambiguous or under-determined demonstrations. One important reason for preferring this way of accounting for demonstrative metonymy is that the phenomenon does not extend to the uses of demonstratives whose antecedents are elements introduced into the discourse by earlier referring expressions, rather than things present in the context of utterance itself. For example, I can point at Tiger Woods and say (25): (25)

That’s what I want to take lessons in.

But this use of the demonstrative does not have a parallel in (26): (26)

?Whenever Mary sees Tiger Woods on TV, she wants to take lessons in that.

When we refer to Tiger Woods by name, that is, there is no demonstration, explicit or implicit, that can pick out the virtual golf bag that he is carrying. And for the same reason, these metonymic references are not available for pure indexicals, either. Tiger Woods cannot use the pronoun I , for example, in a way that parallels the use of that in (26): (27)

?Mary wants to take lessons in me.

11 There are further conditions on this kind of reference that I will not go into here; see Nunberg (1993) for a discussion of these. 12 It is notable how often philosophers use examples of people pointing at photographs to identify the people they picture or of maps to identify places, almost always without seeming to notice the apparent indirection of reference in these cases.

272

II. Referential–Attributive Distinction

Or suppose two teachers at a nursery school are talking about how accomplished the mothers of their charges are. One of them can get away with pointing at a little boy and saying, ‘She’s a famous athlete.’ But when the mother visits the school and wants to brag about her son, she cannot say, ‘I am one of your brightest students.’13 Here again, then, we see that metonymic references require a demonstration, whereas I can only pick out something in virtue of the role it plays in the utterance. So it would seem as if the notion of deferred reference is a mistake, if by that we mean that indexicals or demonstratives can sometimes refer to something other than what we can get to directly via their linguistic meanings or the demonstrations associated with them. But in that case, where do the descriptive readings of indexicals like I and today come from? The only remaining possibility is that the interpretations of these uses of indexicals are the very things that their linguistic meanings pick out of the context. When someone says ‘Today is always the biggest party day of the year’, that is, we have to assume that the interpretation of today—a day-type or day-property, say— is directly picked out by the linguistic meaning of the expression.

9. Granularization of the Context In order to make sense of this conclusion, I should say something about the way we talk about contexts and the meanings of indexical expressions. There is a more-or-less standard view of this which cuts across theoretical diVerences, and which goes something like this. Every utterance is associated with a certain set of elements or coordinates that bear a certain relation to it—its speaker, addressee, time, place, world, salient objects, and so on. And the linguistic meanings of indexical expressions like I , here, and today provide descriptions that pick out those elements in virtue of their relation to the utterance. The content of those meanings may not Wgure as part of the content of the utterances that contain the expressions, but we evaluate it relative to a particular context in the same way we might evaluate the use of any other description. In this sense, at least, the context of utterance is a domain like any other, and the question of determining what entity or element is picked out by the linguistic meaning of an indexical should be subject to the same considerations that we apply in determining the references of other descriptions. That is, there is no appreciable diVerence between evaluating the meaning of today in an utterance of (28) and evaluating the meaning of ‘the day of the last Giants game’ in an utterance of (29): 13 Just to Wll in the pattern, note that pure indexicals can also have descriptive readings when they have discourse antecedents. It is not easy to construct examples that make this point unequivocally, but it can be seen in (i): Whenever Prisoner 28528 and I have tried to order the last meal that tradition speciWcally allows us, we have been rebuVed, and so have all the other prisoners on death row. The token of us in (i) does not pick up its reference directly from the context (among other things, that hypothesis would leave us unable to how to explain the sloppy reading of the VP ellipsis, which would require that us be treated as a variable). But the descriptive reading is available nonetheless.

Nunberg, Descriptive Indexicals (28) (29)

273

Today is cloudy. The day of the last Giants game was cloudy.

In particular, the evaluation in both cases requires that we restrict the domain appropriately. One important consideration, obviously, is that the domain is usually restricted to a set of relevant entities—with (29), for example, we take the domain to be the set of all Giants games played in the contextually relevant year, not all the Giants games that will ever be played. But we may also want to restrict the domain in other ways, to include only the relevant properties of the individuals in question. Reimer (1998a) makes this point by way of explaining the interpretation of a sentence like (30), as uttered by a newscaster describing a local Daughter’s Day event: (30)

Every daughter present was accompanied by a parent.

Clearly, Reimer says, an assertion of (30) can be true even if some of the parents are females, and hence daughters themselves. As she puts it (p. 103): After all, if the context can restrict the domain so as to include only the ‘relevant’ individuals, then why can’t it similarly restrict the properties of those individuals, so as to include only their ‘relevant’ properties? . . . [T]o say that certain individuals in the domain are assigned some of their properties, but not others, is not to say they don’t have those other properties. Rather, it is to say that those other properties are contextually irrelevant, and therefore are not assigned to the individuals in question.

Once we acknowledge that only certain conversationally relevant properties of individuals Wgure in the domain, it follows that the conversational purposes can determine what counts as an individual, as well. That is, there can be only as many individuals in the domain as are individuated by the conversationally relevant properties—the context is subject to what we can think of as a contextual granularization. This sort of restriction of domains is very familiar, even if it is described in various ways. Take a sentence like (31)

The burglars came in through the window.

At Wrst blush, (31) might seem analogous to Strawson’s The table is covered with books, where the ‘incomplete’ description applies to the unique table in the domain. But a speaker can utter (31) without being in a position to provide an identifying description of the particular window that the burglars entered through, nor does anyone care which one it was. In fact, (31) would make a true assertion even if it should turn out that there were two burglars who entered through diVerent windows. Some people have taken sentences like (31) as exceptions to the uniqueness condition for deWnites, but I think this misses the mark.14 Rather, we would say that for conversational 14 Kadmon (1990) argues that the referent of a deWnite description must be ‘not relevantly diVerentiable’ in context, and Birner and Ward (1994) oVer a similar analysis of the italicized NPs in (i) and (ii): (i) As soon as my cousin arrived in Santiago, she broke her foot and had to spend a week in the hospital; (ii) Your ten o’clock

274

II. Referential–Attributive Distinction

purposes, the properties that distinguish one window from another in the house in question are simply irrelevant—the only important thing is that the burglars chose this mode of entry rather than the door or the chimney. Examples like (31) have been widely discussed, though often with a sense that the phenomenon is restricted to a relatively small class of words and that it may involve some idiomaticity (see e.g. Abbott 1999). But in fact the phenomenon is a great deal more general than that. Take the uses of the descriptions in (32) and (33), which are suggested by some of the examples of descriptive indexicals that I gave earlier: (32) (33)

The Stanford–Berkeley game is usually exciting. The biggest party day usually comes after midterms are over.

It is odd to suppose that (32) and (33) involve exceptions to the uniqueness criterion for deWnites, as if the words game and day are invariably restricted to referring to particular contests and particular days. Instead, we would say that the properties that distinguish one particular game or day from another simply are not represented in the discourse model—that is, that there is no way in terms of the contextually relevant properties to distinguish the Stanford–Berkeley game that occurred on January 10, 2001 from Stanford–Berkeley games that occurred on other dates. In the discourse model determined by the context, there is only a single thing to which the description the Stanford–Berkeley game can apply. As I suggested earlier, it does not matter for the present purposes whether we describe this interpretation as a set of properties or generalized quantiWer, or as the constituent of an abstract situation. The important thing is that these elements are simply available in the discourse models for certain contexts in the same way that particular games and days are available in the models relevant to other contexts. In this regard, the context is no diVerent from other domains: we recognize only as many distinct contextual elements as are individuated by the conversationally relevant properties. This point may be clearest when we consider the way we interpret indexicals like here and now. Suppose you are at a reception after a conference talk and someone asks you: (34)

Is John here? I could use a ride back to my hotel.

On the natural interpretation of (34), the reference of here is the room or immediate area where the speaker and hearer are located—a place small enough so that it would not be too inconvenient to get hold of John before he drives away. That is, the conversational purposes impose a certain spatial granularity on the domain of discourse, such appointment said he’d be late because he had to stop Wrst at the bank. Both of these analyses assume that these examples do involve some weakening of the condition of uniqueness. Other analyses follow lines more like the one I am taking here; see e.g. Epstein (2000), who analyzes sentences like these by reference to ‘roles’, and in particular Lo¨bner (1987), who takes these NPs as referring to unique constituents of abstract situations, an analysis that is equivalent to the one I am assuming here. In any event, none of these points is crucial to the argument I am making in this chapter, since the crucial point for me is only that all deWnites be analyzed as quantiWers of one sort or another.

Nunberg, Descriptive Indexicals

275

that two places are distinct only when the diVerence between being in one or the other of them has signiWcant consequences for getting a ride home from somebody. But relative to other conversational purposes, the granularity would be diVerent. Suppose that instead of (34), the speaker says: (35)

I need a ride home; the bus service here is terrible.

In this case, the properties relevant to individuating distinct places are those that might entail diVerences in bus services—being in diVerent towns or neighborhoods, for example. And in other cases, here can refer to an area as large as a nation or even a planet, in the same way that now can refer either to an instant or an era. This is not to say that the interpretation of here in (34) makes any mention of neighborhoods—the neighborhood just happens to be the smallest relevant chunk of the context that is available for reference. Of course (34) and (35) do not involve distinct uses or meanings of here. In both cases, here picks out its referent in virtue of the same linguistic meaning, ‘the place of utterance’. The only diVerence between the utterances is in that we appeal to diVerent criteria to determine what counts as a distinct place. Examples like these show that we have to evaluate the linguistic meanings of indexicals relative to a certain granularization of the context. In the case of (34) and (35), this is a question of individuating distinct locales in a continuous region.15 But we can invoke the same principle to explain the descriptive readings of indexicals. Take Today is always the biggest party day of the year. Today picks out the day of utterance, just as it always does. But in the relevant discourse model, there are only as many distinct days as are individuated by the properties relevant to the conversational purposes, just as there are in a sentence like The biggest party day of the year is usually the day after midterms. That is, there is no way to individuate the day of utterance from others that have the same relevant properties. In this sense, the reference here is not ‘deferred’: the linguistic meaning ‘the (calendar) day on which the utterance takes place’ is satisWed by the only day in the domain of discourse that corresponds to the time of utterance—but relative to the conversational purposes, that day simply does not have the properties that diVerentiate the actual day of the utterance from others that fall on the same date. We will tell the same sort of story about an utterance like (4), I am traditionally allowed to order whatever I want for my last meal. Here, the domain of discourse from which the reference of I has to be drawn can contain only the sorts of entities that could be the subjects of traditional dicta—sets of personal properties, for example. And yet the analysis I am suggesting may seem more strained here than it does for the example involving today that I just mentioned. The reason is that the linguistic meaning of today includes the descriptor ‘day’, which applies as well to day properties or types of days as it does to particular days. That is, we could paraphrase the meaning of (2) as (36), provided we give wide scope to the deWnite description: 15 Note that the same inferences that determine the interpretation of here in examples like these are what determine the interpretation of a given use of a description like ‘the economy’ or ‘the weather’.

276

II. Referential–Attributive Distinction (36)

The day of this utterance is always the biggest party day of the year.

Whereas however we describe the linguistic meaning of I —as ‘the speaker of the utterance’, ‘the agent of the context’, or whatever—it is not easy to think of this descriptor as applying naturally to the entity that is the subject of the traditional dictum about last meals. That is, the linguistic meaning of I picks out its contextual index qua its role as speaker, but that entity is being asked to Wgure in the content of the propositional content qua its status as a condemned prisoner. So (4) does not have a natural paraphrase as (37), even when we try to give wide scope to the description ‘the speaker of this utterance’: (4) (37)

I am traditionally allowed to order whatever I want for my last meal. The speaker of this utterance is traditionally allowed to order whatever he wants for his last meal.

But this takes me back to the second question I began this section with: why can’t the referential uses of deWnite descriptions have these descriptive interpretations? If the description ‘the speaker of this utterance’ can take us to the speaker, why can’t we then throw out the content of that description and take that person himself as the interpretation? And if the properties relevant to individuating persons in the contextual domain do not permit us to diVerentiate the speaker from other persons who are also condemned prisoners, then why can’t the subject of (37) have a descriptive interpretation? At this point I want to revisit a familiar assumption about the way the content of a description functions in these referential uses. As Wettstein (1981: 36) puts it: Consider the referential use [of deWnite descriptions]; there are contexts in which a speaker wants to draw his audience’s attention to an entity, perhaps one visually present to both speaker and audience, in order to go on and, for example, predicate something of it. It is irrelevant to the purposes of the speaker, in many such cases, how the attention of the audience is directed to the referent. Pointing with one’s Wnger or uttering a proper name would do as well as some elaborate description.

In other words—and this view is generally shared by both Russellians and referentialists—the choice of descriptive content in a referentially used description has no conversational relevance over and above its utility in picking out the intended referent. And that is what permits the referentialist to further assume that we can discard the content when its identiWcational work is done. But this is not quite true. The most obvious counter-examples are sentences like (38), say as the opening sentence of a newspaper story: (38)

The contractor who contributed $150,000 to the governor’s re-election campaign was awarded a $10 million state contract yesterday.

The description in the subject NP of (38) is clearly used referentially—the speaker has a particular person in mind, and the story will most likely go on to identify him by name. But the content of the description is not functioning simply to identify the

Nunberg, Descriptive Indexicals

277

individual, but to attribute some property to him that is relevant to the predication— the sentence implies that the contract was awarded because the contractor made a large contribution (or at least, it suggests that some people might infer that some funny business was afoot). And clearly the choice of another description that identiWed the same person would change what was said. Suppose, for example, that the contractor who contributed $150,000 to the governor’s campaign was also his brother-in-law. Then on the referentialist view, there would be no semantic diVerence between saying (38) and (39): (39)

The governor’s brother-in-law was awarded a $10 million state contract yesterday.

But obviously (39) has a very diVerent conversational signiWcance. I am not saying that sentences like these entail a causal relation between the content of the description and what is predicated of its referent, but they strongly invite that inference. And they do so on the assumption that the content of a referentially used description can have a conversational relevance over and above its usefulness in identifying the referent that the speaker has in mind. In fact, these are only the extreme cases of a more general principle: following the maxim of relation, we tend to construe the content of any description as being relevant to the conversational purposes at hand, even when it seems to be functioning in a purely identiWcational way. Consider, for example, the diVerence between saying ‘please pass the potatoes’ and ‘please pass the red plate’. It may very well be that either description would be a perfectly reasonable way of identifying the thing that the speaker wants the hearer to pass him, but the second utterance is clearly an odd way to go about things, and it would probably trigger some sort of conversational implicature—maybe the speaker is interested in the plate qua plate (is he a china specialist?). And by the same token, we invite diVerent inferences when we say ‘The man with the martini is Jones’, ‘The man in the ill-Wtting suit is Jones’, and ‘The man who just yelled ‘‘blimey!’’ is Jones’, even if each of those descriptions is an equivalently eVective way of drawing the hearer’s attention to the person that the speaker has in mind. Since the choice of descriptive content is always under-determined, that is, we assume that the speaker generally has some additional motive for choosing one description rather than another as a way of identifying someone—though it is true that sometimes not much attaches to the diVerence. These observations sit very easily with the Russellian analysis of referential descriptions, which assumes that the content of the description is always in fact a part of what is said. By themselves, though, they may not rule out a referentialist account, since a referentialist could always argue that the inferences associated with utterances like these arise out of conversational implicatures via the maxim of manner. That is, the writer of (37) may not actually have asserted that the man who was awarded the state contract was a contributor to the governor’s campaign, but the fact that she chose that means of identiWcation rather than some other might nonetheless be signiWcant.

278

II. Referential–Attributive Distinction

But at this point let us return to the observation that these referential uses of descriptions cannot have descriptive readings. Why should this be? After all, if the relevance of descriptive content is merely a conversational inference, rather than implicit in the semantics of the utterance itself, then we ought to be able to ignore that content when the conversational purposes require us to. Take an example like ‘Sophie’s eleventh birthday is always the biggest party day of the year.’ Suppose that the description ‘Sophie’s eleventh birthday’ takes us to November 1, 2001, and that the property of falling in one or another particular year is not conversationally relevant—in the domain, that is, there are only as many distinct days as are individuated by the date of the year they fall on. Then once we have identiWed the referent of the description, we should be able to toss its content aside and take the day-type as its interpretation, so that the utterance comes out as meaning something like ‘November 1 is always the biggest party day of the year.’ But that is not possible here. We are required to interpret the utterance as if the property of being Sophie’s eleventh birthday is relevant to what is being said, even if that leads us to an incoherent interpretation of the whole. As much as we’d like to, that is, we simply cannot throw out the content of the description—it turns up like a bad penny. And that is a fact that is simply incompatible with the referentialist account of these sentences. This same observation explains why an utterance like (37) sounds odd to us: in choosing to refer to himself as ‘the speaker of this utterance’ rather than as I , the speaker conversationally implicates that he has some relevant reason for choosing that particular descriptive content—we would infer that he means us to believe that the tradition about last meals applies to people in virtue of their roles as speakers, even if that does not make a lot of sense. But there is no analogous implicature when a speaker refers to himself as I . Inasmuch as the linguistic meaning of the expression does not Wgure as part of the utterance content, we are not obliged to construe it as conversationally relevant. The fact that we have picked out an element of the context in virtue of its role as the speaker does not mean that that property is pertinent to its individuation.16

Appendix I should say something about the diVerence between deWnite descriptions and demonstrative descriptions (or ‘complex demonstratives’) like ‘that guy’. It was suggested by 16 In fact there can almost never be a manner implicature associated with the use of an indexical, since as we saw that use is obligatory whenever it is felicitous. Since you are ordinarily obliged to refer to today as ‘today’, there is no signiWcance attached to that choice. (There are some exceptions to this generalization: the choice of a demonstrative description can be signiWcant when a deWnite description would do as well, as in ‘You’re not going to tell that story about your accident again, are you?’ But these are invariably tinged with an aVective import, and do not have any bearing on the cases we are talking about here.)

Nunberg, Descriptive Indexicals

279

one of the reviewers of this chapter that such expressions should be regarded as equivalent to deWnite descriptions. But there are good reasons for distinguishing between the two. For one thing, demonstrative descriptions cannot be re-evaluated relative to another hypothetical context: (i) If you had turned to Channel Four as I asked, the show would be more interesting; (ii) If you had turned to Channel Four as I asked, this show would be more interesting. In (i), the description the show could refer to the show on Channel Four, but this show in (ii) can only be evaluated relative to the actual context—that is, it refers to the show that is currently visible, not the one on Channel Four. In this regard, this show behaves like a pure demonstrative, not like a description. It should not be surprising, then, that demonstrative descriptions permit descriptive readings that are not easily available for deWnite descriptions. For example, you might ask either (iii) or (iv) to someone watching the Stanford–Cal game on TV: (iii) Are you enjoying the game? (iv) Are you enjoying this game? And if the respondent is in fact enjoying the game that is on TV, he might answer with either (v) or (vi): (v) Yes, I’m enjoying the game a lot. (vi) Yes, I’m enjoying this game a lot. But if the respondent wants to invoke a general disposition to enjoy Stanford–Cal games, he cannot felicitously answer with (vii), but must use (viii): (vii) ?Yes, I always enjoy the game. (viii) Yes, I always enjoy this game. That is, only the version with the demonstrative description this game permits a reading where the description contributes a property of the game that is currently on view.

7 The Case for Referential Descriptions Michael Devitt

1. Introduction DeWnite descriptions (‘deWnites’) typically have the form, ‘the F ’, but may also have forms like ‘his F ’ (equivalent to ‘the F of him’). IndeWnite descriptions (‘indeWnites’) have the form, ‘a/an F ’. According to Russell’s theory of descriptions (1905), ‘the F is G ’ is equivalent to ‘there is something that is alone in being an F and it is G ’; and ‘an F is G ’ is equivalent to ‘there is something that is an F and it is G ’. So descriptions are to be understood in terms of quantiWers and the nominal ‘F ’. Under the inXuence particularly of Keith Donnellan (1966, 1968), many now think that deWnites are ‘ambiguous’, having not only the ‘attributive’ meaning captured by Russell1 but also a ‘referential’ meaning like that of a name or demonstrative. Under the inXuence particularly of Charles Chastain (1975),2 some now think the same of indeWnites. It is generally agreed that descriptions have a referential use as well as an attributive use. Used attributively, ‘the F ’ conveys a thought about whatever is alone in being F and ‘an F ’, a thought about some F or other; they convey ‘general’ thoughts. Used referentially, each description conveys a thought about a particular F that the speaker has in mind, about a certain F; each conveys a ‘singular’thought.3 Despite agreement that descriptions have these two uses, there is no agreement that they have two meanings. The quantiWcational attributive meaning described by Russell is uncontroversial, but many have appealed to ideas prominent in the work of Paul The chapter has beneWted greatly from the comments of Emma Borg, Ernie Lepore, Gary Ostertag, Marga Reimer, Gabriel Segal, Rob Stainton, and, particularly, Kent Bach and Stephen Neale (whose current position is closer to RD than to that in his 1990). It has also beneWted from discussions in various courses on reference that I have given at the University of Maryland and the Graduate Center, and from presentations at Syracuse University in Nov. 2001 and the Graduate Center in Apr. 2002. 1 Any diVerences that there may be between what Donnellan says of ‘attributive’ descriptions and Russell’s theory of descriptions are beside the point of this chapter. 2 See also Strawson (1950); G. Wilson (1978). 3 Some philosophers call these thoughts ‘de re’, others, ‘object-dependent’. I think that there are reasons against both usages (Devitt 1996: 144 n.; 1985: s. 3).

Devitt, Case for Referential Descriptions 281 Grice (1989) to deny that descriptions also have a referential meaning. They argue that the referential use of descriptions does not aVect truth conditions and is not semantically signiWcant; it is merely pragmatic. Saul Kripke (1977) is a famous example of someone who has argued along these lines. Stephen Neale (1990) has presented the most thorough and persuasive case against the referential meaning of deWnites, and Peter Ludlow and Neale (1991), against that of indeWnites.4 I think that the case for the thesis that descriptions have referential meanings has been greatly underestimated. I shall defend the thesis that the referential uses of both deWnites and indeWnites are semantically signiWcant: there are referential descriptions not just referential uses of descriptions. Let us call this thesis ‘RD’. I shall present six arguments for referential deWnites and Wve for referential indeWnites. My aim is to develop and distinguish arguments for RD already in the literature, drawing attention to those that have not yet been criticized; to respond to some criticisms; and to propose some novel arguments. I shall presuppose some familiarity with the literature and so my discussion will often be light on examples. What are referential meanings? For the purposes of this chapter, the important part of the answer is as follows: the core of the referential meaning of a description token is its reference-determining relation to the particular object that the speaker has in mind in using the description.5 I will go further into the answer in section 5.

2. Argument I: The Regular Expression of Singular Thoughts DeWnites Consider this well-known example: ‘Someone sees a woman with a man. Taking the man to be her husband, and observing his attitude to her, he says, ‘‘Her husband is kind to her’’ . . . Suppose the man is not her husband. Suppose he is her lover, to whom she has been driven precisely by her husband’s cruelty’ (Kripke 1977: 7). The speaker has a thought that is not a general one about whoever happens to be her unique husband. It is a singular thought about a certain person, the lover, that the speaker has in mind. Donnellan’s discussion of such examples clearly shows that the speaker’s utterance 4 See also King (1988). 5 I use ‘semantic’ and ‘meaning’ more broadly than some. On my usage, it is a semantic fact, not a pragmatic one, that a certain deictic token of ‘he’ refers to the person the speaker has in mind and that the containing utterance depends for its (literal) truth on that reference. And it is a semantic fact about the pronoun type ‘he’ that it has that deictic use and also has an anaphoric use. According to RD, it is similarly a semantic not pragmatic fact that a certain description token refers to the person the speaker has in mind and that the containing utterance depends for its (literal) truth on that reference. And it is a semantic fact about the description type that it has such a use as well as its use as a quantiWer. In contrast, Kent Bach (1987/1994) restricts semantics to the study of types, and places reference in pragmatics. For him a token has only such meaning as it inherits from its type and so the reference of our token ‘he’ is not part of its meaning. Bach’s motivation is that semantics concerns only what the competent speaker knows. I have a very diVerent view (Devitt 1996).

282

II. Referential–Attributive Distinction

conveys this singular thought even though ‘her husband’ misdescribes the lover. This use of the deWnite, ‘her husband’, is referential. So far, Russell could go along with Donnellan. However, Donnellan seems to be suggesting much more:6 that in this referential use the lover not the husband is the referent of ‘her husband’. So the referential use aVects truth conditions: it is semantically signiWcant. In particular, (a) The truth of the remark depends on whether the lover is kind to her, despite the fact that the remark misdescribes him. (b) Whether or not the husband is kind to her is irrelevant to the truth of the remark. Russell, in contrast, holds that the truth depends on the husband not the lover. A great deal of the debate has been over (a) and the issue of misdescription. To defend (a) one has to claim that the nominal ‘F ’ in ‘the F ’does not play a semantic role in sentences containing the referential ‘the F ’ but only a pragmatic role. This is prima facie implausible given the semantic role nominals generally play. Later (section 5), I shall oVer further reasons for thinking that ‘F ’ does indeed play a semantic role. Meanwhile, I shall simply assume that it does with the result that (a) is false: for the remark to be true the person in mind, the lover, must be one husband of the woman even if not her unique husband. In any case, it is (b) not (a) that is essential to RD. I think that (b) is intuitively appealing. Kripke and Neale do not. Clearly we need to go beyond intuitions to settle the matter. Kripke and Neale rightly point out that the fact that a sentence conveys a singular thought does not alone show that the sentence literally and conventionally7 has this singular meaning; it does not show that the referential use is, in this sense, semantically signiWcant. For Grice has drawn our attention to the way sentences can convey meanings that they do not literally have; their ‘speaker meaning’ can be diVerent from their conventional meaning; we may not ‘say what we mean’. So Grice has opened up the possibility that the referential use of a deWnite is to be explained pragmatically not semantically. Neale has argued persuasively that a pragmatic explanation is the right one by making a comparison with other quantiWers (1990: 87–91).8 Thus, consider ‘every’: Suppose it is common knowledge that Smith is the only person taking Jones’ seminar. One evening, Jones throws a party and Smith is the only person who turns up. A despondent Jones, when asked the next morning whether his party was well attended, says, 6 The caution is necessary because Donnellan’s discussion is rather equivocal, as Kripke (1977) points out. See also Bertolet (1980). 7 Mostly we can identify the literal meaning of an utterance with its conventional meaning but Donald Davidson (1986) has shown with his discussion of Mrs Malaprop’s ‘nice derangement of epitaphs’ that this is not always so. But this is irrelevant to our present concerns. 8 Kripke’s comparison of the case of the lover, involving the deWnite ‘her husband’, with the case of Smith raking the leaves, involving the name ‘Jones’ (Kripke 1977: 15–18) is much less persuasive: the two cases are crucially diVerent (Devitt 1981b: 512–16).

Devitt, Case for Referential Descriptions 283 (7)

Well, everyone taking my seminar turned up

fully intending to inform me that only Smith attended. The possibility of such a scenario would not lead us to complicate the semantics of ‘every’ with an ambiguity; i.e., it would not lead us to posit semantically distinct quantiWcational and referential interpretations of ‘everyone taking my seminar’. (87–8)

Similarly, the possibility of Donnellan’s scenarios should not lead us to complicate the semantics of ‘the F ’. Neale goes on to argue that Grice’s pragmatic theory of ‘conversational implicature’ (1989) explains the mechanism by which, in all these scenarios, the speaker conveys a meaning that his words do not literally have. Thus, the theory explains how Neale, by assuming that Jones is acting in accordance with ‘the Cooperative Principle’ and its maxims, derives the implicature (speaker meaning), ‘Only Smith turned up’, from what Jones literally said (conventional meaning). The point is well taken. Still, I think that there is an eVective response to it (Devitt 1997a: 125–8; 1997b: 388; Reimer 1998c). The basis for RD is not simply that we can use a deWnite referentially, it is that we regularly do so. When a person has a thought with a particular F object in mind, there is a regularity of her using ‘the F ’ to express that thought. And there need be no special stage setting enabling her to conversationally imply what she has not literally said, nor any sign that her audience needs to use a Gricean derivation to understand what she means. This regularity is strong evidence that there is a convention of using ‘the F ’ to express a thought about a particular F, that this is a standard use. This convention is semantic, as semantic as the one for an attributive use. In each case, there is a convention of using ‘the F ’ to express a thought with a certain sort of meaning/content. ‘Every’ and other quantiWers are diVerent. There is no convention of using them to convey a thought about a particular object in mind. With special stage setting they certainly can be used for that purpose, as Neale illustrates. But then Grice shows us that with enough stage setting almost any expression can be used to convey almost any thought. If there is indeed a semantic referential convention for deWnites, then the referent of ‘the F ’ when used referentially must be the object the speaker has in mind. So, (b) is correct: the husband is irrelevant to the truth of ‘her husband is kind to her’ in the above example. Critics of RD have not yet replied to this argument. How might they do so? (1) They might deny that deWnites are regularly used referentially. But the range of examples produced by Donnellan and others make this denial implausible: day in and day out we do seem to use deWnites to express singular thoughts. (2) A more likely reply accepts the regularity but attempts to explain it in a Gricean way. The claim would be that a Gricean derivation could yield the appropriate conversational implicature in each referential use. Each use can be seen as exemplifying the pragmatic conversational convention of observing the Cooperative Principle and its maxims. So if the regularity of referential uses is to be seen as a convention at all it is to be seen as simply a derived

284

II. Referential–Attributive Distinction

pragmatic one. There is no need to see it as a semantic convention, no need to see deWnites as literally having referential meanings. There is something deeply wrong with (2). We can get an indication of this by going further down its path. (2) rests on the following view: where an utterance has a conventional meaning and we can derive a diVerent speaker meaning (implicature) from this conventional meaning with the help of appropriate assumptions about the context and mind of the speaker, there is no need to suppose that this derived meaning is another conventional meaning. Now someone over-inspired by Grice—‘a fundamentalist Gricean’—might go further, arguing that there are no conventional meanings at all: it is pragmatics all the way down. Consider the time before there were languages. Then there were clearly no conventional meanings to feature in derivations of speaker meanings. Yet, presumably, there came a time when people spoke and others derived the speaker meanings of these utterances simply from appropriate assumptions about the context and the mind of the speaker (just as we still often have to in a foreign country). There came to be regularities in the associations of speaker meanings with sounds and, we are inclined to say, over time some of these ‘caught on’ and became the Wrst conventional meanings of a language (Devitt and Sterelny 1999: 7.6). But our fundamentalist is not inclined to say this at all: ‘There is no need to take these Wrst meanings to be conventional because they can all be derived from assumptions about the context and other minds. We must always remember the word of Grice. His ‘‘ModiWed Occam’s Razor’’ tells us ‘‘Senses are not to be multiplied beyond necessity’’ (1989: 47). I say unto you that no senses are necessary.’ How will a defender of (2) resist this reductio of her position? She is likely to object that although we could in principle recreate the derivations of these Wrst meanings, contemporary speakers cannot as a matter of fact do so. In contrast speakers can as a matter of fact derive referential meanings.9 But why is this contrast between fundamentalism and (2) signiWcant? It rests on a contingent and seemingly irrelevant matter of ignorance. Suppose that there was a massive breakthrough in the study of the origins of English so that we could now all recreate the original derivations. That would surely not eliminate conventional meanings! A consideration of metaphors drives the point home. A metaphor is a Gricean paradigm: a derivation from the conventional meaning yields an implicature that is the metaphorical speaker meaning. Now, in time, a metaphor often ‘dies’: an expression comes to mean literally what it once meant metaphorically. Yet an argument along the lines of (2) seems to show that this death is impossible: there would be no moment at which we would cease to be able to give a Gricean derivation of the formerly metaphorical meaning and hence no moment at which we would need to conclude that this meaning had become a new convention. The defender of (2) might respond: ‘There is indeed no determinate moment at which a metaphor becomes dead, but this 9 Bach (1987/1994) treats the referential use of incomplete deWnites as a case of ‘standardized non-literality’. Bach (1995: 683) thinks that in such cases, but not in cases of semantic conventions, speakers have the information to go through a Gricean derivation of the meanings conveyed.

Devitt, Case for Referential Descriptions 285 is just another example of vagueness. In the end, people do cease to be able to derive the expression’s formerly metaphorical meaning and then the metaphor is dead.’ So, once again we have a defense of (2) that rests on the ignorance of contemporary speakers. And this time the defense clearly fails because speakers are often not ignorant. The expression often still has its original meaning as well as its formerly metaphorical meaning and so speakers could still give a Gricean derivation of the latter meaning. Consider Marga Reimer’s nice example of the verb ‘incense’ (1998c: 97–8). It still means ‘make fragrant with incense’ although it is now more commonly used to mean ‘make very angry’. That use was once metaphorical, to be explained by a Gricean derivation, and it still could be, as Reimer demonstrates. So, according to (2), there is no need to see ‘incense’ as literally meaning ‘make very angry’. Similarly for thousands of other dead metaphors: these expressions do not literally mean what dictionaries say they mean. What has gone wrong? First, (2) is mistaken in suggesting that the ability of speakers to give a Gricean derivation of the meaning conveyed by an expression makes that meaning a matter of pragmatics not semantics. Indeed, if (2) were right about this, we could make semantic meanings pragmatic simply by removing ignorance of their original derivations: we could eliminate semantic meanings by bringing people up on a close study of the Oxford English Dictionary; Gricean fundamentalism run amok!10 We can agree that if speakers not only could but do use a Gricean derivation to grasp the meaning conveyed then the meaning is a pragmatic not semantic matter. Thus it is plausible that Neale grasped ‘Only Smith turned up’ by a Gricean derivation, and that audiences grasp metaphors likewise. But people do not now grasp what speakers commonly mean by the verb ‘incense’ in that Gricean way. That is why the metaphor really is dead. And people do not grasp what a speaker means by a referentially used deWnite in that way either. Rather, they grasp the meaning immediately and directly because that is the meaning it conventionally has. Second, I suspect that the mistake arises out of inattention to what makes a convention semantic. Gricean discussions have much to say about pragmatic conventions but little about what makes other conventions semantic and not pragmatic. This is probably because it is very diYcult to say what makes a convention semantic. And I do not claim to be able to say.11 But such grip as we do have on the matter makes it hard to see what basis there could be for denying that the referential use of deWnites is a semantic convention. To undermine the view that the referential use of deWnites is semantic it is not enough to show that the use could be explained pragmatically. We need an argument 10 And being able to explain why a quantiWer phrase can readily be used to refer does not show that when it is so used the referential meaning conveyed is a matter of pragmatics not semantics. Indeed, diachronic linguistics seeks laws of meaning change that explain the development of one meaning from another. The crux of Bach’s response to me in point 13 (this volume) is that the distinctive quantiWcational character of deWnites helps explain how they can readily be used to refer and plays a role when they are so used. The editors of the volume encouraged us to take account of each other’s chapters. However, someone had to have the last word here and the agreement was that I would not respond in my chapter to his point 13 comments. I plan a response elsewhere. 11 My best attempt is Devitt (1997a: 127–8).

286

II. Referential–Attributive Distinction

to show that it should in fact be explained pragmatically. Neale’s comparison of this use with the referential use of other quantiWers seemed to be a persuasive argument to this eVect. I claim to have rebutted that argument. Perhaps another argument can be mounted comparing the referential use of deWnites persuasively to other phenomena that seem to be pragmatic; for example, to ‘generalized’ conversational implicatures (Grice 1989: 37–8) apparently exempliWed by some uses of ‘and’ and ‘some’; or to ‘standardized non-literality’ illustrated by ‘Have you eaten?’ and ‘I have nothing to wear’ (Bach 1987/1994: 79–81).12 That remains to be seen. So far as I know, no such argument has been oVered.13 And, in light of dead metaphors and the threat of a fundamentalist reductio, the prospects of an argument do not seem promising. The regular use of deWnites to express singular thoughts looks like a semantic convention.

IndeWnites Much the same story goes for indeWnites. Consider the following example: Several of us see a strange man in a red baseball cap lurking about the philosophy oYce. Later we discover that the Encyclopedia is missing. We suspect that man of stealing it. I go home and report our suspicions to my wife: ‘A man in a red baseball cap stole the Encyclopedia.’ Suppose that our suspicions of the man we saw are wrong but, ‘by chance’, another man in a red baseball cap, never spotted by any of us, stole the Encyclopedia. I wish to convey a singular thought about the particular person we saw in the oYce not a general thought about anyone in a red baseball cap. I do so by using an indeWnite referentially. The literature contains many similar examples of indeWnites being used to convey singular thoughts. This much is relatively uncontroversial (but see section 5, pp. 293–4). Controversy begins with the further claim that, contrary to Russell, my remark is not true: the fact that some man in a red baseball cap who was not my suspect stole the Encyclopedia is irrelevant to the truth-value of my remark. This claim is analogous to (b) above for deWnites. I think that the claim is intuitively appealing but Ludlow and Neale (1991) do not. Once again, we need to go beyond intuitions. The familiar Gricean point prevents any simple inference from the referential use of indeWnites to their having a referential meaning: indeWnites, like any other expression, can speaker mean what they do not conventionally mean. What supports the inference 12 Bach claims, plausibly, that these sentences have a standard use that is non-literal and hence not a semantic convention. But the standard use of sentences containing referential deWnites is explained, as the use of Bach’s sentences is not, by the standard use of a contained expression: sentences containing ‘the F ’ are standardly used referentially because ‘the F ’ has a standard referential use. Bach’s examples give no reason to think that this use of ‘the F ’ is not a semantic convention. Still, standardized nonliterality, and generalized implicatures, show that the absence of a Gricean derivation is, sadly, not suYcient to make a use semantic. 13 At times Neale seems to suggest that referential uses are generalized implicatures (1990: 81, 90) but his argument is the one I have discussed, concerned with the ‘particularized’ implicatures involved in referential uses of quantiWers.

Devitt, Case for Referential Descriptions 287 is that we regularly use indeWnites referentially. When a person has a thought with a particular F object in mind, there is a regularity of her using, without any special Gricean stage setting, ‘an F ’ to express that thought. This is strong evidence that there is a convention of using ‘an F ’ to express a thought about a particular F, that this is a standard use. This convention is semantic, as semantic as the one for an attributive use. Do the conventions for the referential uses of deWnites and indeWnites diVer and, if so, how? I shall brieXy address this question in section 5.

3. Argument II: ‘Donnellan English’ and ‘Chastain English’ DeWnites Kripke has an argument along the following lines in defense of Russell’s view. Let us stipulate that a group of people speaks a language based on English in which deWnites are only attributive. Call this language ‘Russell English’. Kripke claims that the phenomena generated by speakers of this language would not diVer from those generated by speakers of English. So these phenomena do not falsify Russell’s view that English simply is Russell English (1977: 16–17). I reject Kripke’s claim: the phenomena would diVer because there is no convention in Russell English of using deWnites to express singular thoughts (Devitt 1981b: 520). With careful stage setting there would be the occasional use for this purpose, just as there is with any quantiWer in English, but there would not be the regular use. When speakers of Russell English wanted to express a singular thought they would typically rely not on deWnites but on other devices in the language that are conventionally for that purpose. In the next section I shall argue that demonstratives and pronouns are such devices in English. If so they would also be in Russell English. Speakers of Russell English would behave diVerently from speakers of English because in the vast majority of situations where speakers of English use a deWnite referentially, speakers of Russell English would use a demonstrative or pronoun. So Russell’s view is falsiWed after all. We can go further. Let us stipulate that a group of people speaks a language based on English in which there is not only the Russellian attributive convention but also a convention of using descriptions to express singular thoughts. Call this language ‘Donnellan English’.14 The phenomena generated by speakers of this language would not diVer from those generated by speakers of English; there would be just the same regularities. So, these phenomena conWrm that English simply is Donnellan English; they conWrm RD. 14 According to Richard Larson and Gabriel Segal there are languages that directly display an ambiguity in (their translation of ) ‘the’: ‘German, modern Greek, and spoken colloquial Spanish’ (1995: 334); think e.g. of ‘Der Hans’.

288

II. Referential–Attributive Distinction

I do not claim that this point about Donnellan English adds much to the case for referential deWnites: it is unlikely to seem persuasive to someone who has not already accepted the case. But, similarly, Kripke’s claim about Russell English is unlikely to seem persuasive to someone who has not already accepted the case against referential deWnites.

IndeWnites Inspired by Kripke someone might stipulate another feature of Russell English: in it indeWnites (like deWnites) are only attributive (Ludlow and Neale 1991: 179–80). The argument in defense of Russell’s view that English is Russell English would then proceed as before. And so would my response: Russell English would seem diVerent from English. Among speakers of Russell English, with careful stage setting, there would be the occasional use of indeWnites to express singular thoughts, just as there is with any quantiWer in English, but there would not be the regular use. Russell’s view would be falsiWed. Again we can go further. Let us stipulate a language, ‘Chastain English’, in which there is a convention of using indeWnites (as well as deWnites) to express singular thoughts. The phenomena generated by speakers of this language would not diVer from those generated by speakers of English; there would be the same regularities. So, these phenomena conWrm that English simply is Chastain English; they conWrm RD.

4. Argument III: Comparison with Deictic Demonstratives DeWnites It is agreed that there are singular thoughts, thoughts about a particular object in mind. We should then expect there to be conventional ways of expressing them. The most obvious and uncontroversial way is by the simple demonstratives, ‘this’ and ‘that’, and the pronouns ‘he’, ‘she’, and ‘it’. Indeed, this seems to be the sole role of these terms in their deictic (non-anaphoric) use. And it seems to be one role, at least, of the complex demonstratives, ‘this F ’ and ‘that F ’, in their deictic use. I shall focus on ‘that F ’. The argument, which builds on argument I, is that the deWnite, ‘the F ’, is like the deictic ‘that F ’ in two ways (Devitt 1981a, b; Wettstein 1981). First, the expressions have a very similar conventional role. We could usually change one for the other without apparent cost to our goal of communicating a singular thought. When a demonstration (pointing gesture) is called for, ‘that F ’ may seem more appropriate, when it is not, ‘the F ’. However, in each case, the other expression would usually do Wne. Little if any change in stage setting is required to change from one expression to

Devitt, Case for Referential Descriptions 289 the other.15 Neale himself talks of referential deWnites ‘functioning like a name or like a demonstrative’ (1990: 85–6). This function of deWnites is as conventional as that of demonstratives. The convention for demonstratives is semantic not pragmatic, if anything is, and the convention for deWnites is too. So deWnites have a referential meaning. Second, in that meaning, deWnites are like demonstratives in depending for their reference partly on whatever mechanism determines which object is in mind. (I shall discuss this mechanism in section 5.) Although all deWnites have a function like a demonstrative,16 argument III is most persuasively run using ‘incomplete’ deWnites (like ‘the book’) as examples. Incomplete deWnites are the concern of argument V below. But it is important to see that argument III is quite diVerent from argument V. III is a direct argument for the semantic signiWcance of all referentially used deWnites—for RD—by comparing them with demonstratives. It does not rest on the claim that the Russellian theory fails to handle incomplete deWnites. V, in contrast, is primarily an argument that the Russellian theory does fail at this. Hence V is indirectly an argument for RD that can handle incomplete deWnites. V does not rest on any comparison with demonstratives. V has been much discussed and criticized. III has not. Why not? One reason is that the two arguments are not sharply distinguished in the literature.17 Another, it seems to me, is a conservative climate of opinion that takes the task to be simply to defend the Russellian status quo from criticisms rather than to examine the independent merits of Donnellan-inspired alternatives.

IndeWnites With one interesting diVerence, the argument for referential indeWnites is the same as for deWnites. Aside from that diVerence, ‘an F ’ has a conventional function very like the deictic ‘that F ’ (and hence the referential ‘the F ’). The diVerence dictates a diVerence in stage setting but, apart from that, a speaker could change from the demonstrative to the indeWnite, or vice versa, without cost to her goal of communicating a singular thought. The convention for demonstratives is semantic and so the slightly diVerent one for indeWnites must surely be so too. So indeWnites have a referential meaning. And in that meaning, indeWnites like demonstratives depend for their reference partly on whatever mechanism determines which object is in mind. 15 Bach claims: ‘Suppose you want to refer to some thing (or someone). Suppose it is not perceptually present, has not just come up in the conversation, and is not otherwise salient. Suppose that it does not have a name or that you are unaware of its name or think your audience is unaware. Then you cannot use an indexical, a demonstrative (pronoun or phrase), or a proper name to refer to it . . . Your only recourse is to use a description’ (This volume, point 3). Bach oVers no examples to support his claim that there are situations where a description (‘the F ’) could be used to refer but a complex demonstrative (‘that F ’) could not. I want to insist that if there are such situations, they are relatively rare. 16 Pace Bach (1987/1994), who thinks that ‘no one would suggest’ this of complete deWnites. 17 The focus of Wettstein (1981) is much more on V than III. The focus of Devitt (1981a, b) is much more on III than V.

290

II. Referential–Attributive Distinction

The diVerence in the conventions for demonstratives and indeWnites lies in the intentions of the speaker. I shall discuss the diVerence at the end of the next section.

5. Explaining Referential Meanings Perhaps those who reject RD are partly inXuenced by doubts about whether the referential meanings of descriptions could be explained. So let us brieXy consider that matter. The argument of the last section suggests that the explanation of these meanings will be much the same as that of the meanings of complex demonstratives. For both explanations we must look to the singular thoughts that demonstratives and referential descriptions express. In virtue of what is a thought a singular thought about a certain object in mind? There are diVerences of opinion about the answer. Thus, I argued that the thought is simply one that is grounded in the object by a certain perceptual causal process (Devitt 1974: 191–2).18 Gareth Evans called this austere view ‘the Photograph Model’ and rejected it on the ground that a singular thought about an object requires more than this sort of perceptual link: the thinker must also ‘have the capacity to distinguish the object . . . from all other things’ (Evans 1982: 89). Martin Davies (1981: 97) has a similar view tentatively endorsed by Neale (1990: 18). I have argued against Evans’s extra requirement (Devitt 1985). However, for present purposes, we need not take a stand on these theoretical diVerences. It is suYcient to go along with the consensus that a singular thought is about a certain object in mind in virtue of a link to that object that is, at least partly, some sort of perceptual causal link. This link between a singular thought and an object provides the core of the meaning of any token complex demonstrative or referential description produced by the thought. For the link is central in explaining the fact that the token designates that object.19 This claim captures the appealing idea that a demonstrative designates the object that the speaker intends to refer to (Kaplan 1989b: 582–4). But it goes beyond that idea: it explains in virtue of what that particular object is the intended one. The claim is at odds with two other ideas about demonstratives. The Wrst is also appealing: that a demonstrative token designates the object demonstrated (Kaplan 1989a: 489–91).20 This idea has three problems. (i) A demonstration is often so vague that it alone would not distinguish one object from many others in the environment. (ii) If an object is suYciently salient in an environment, a demonstrative that refers to it may well not be accompanied by a demonstration. (iii) Reference is often to an object 18 The explanation is developed in Devitt (1981a: 37–40; 1981b: 515–16). It is far from complete because of the ‘qua-problem’: Devitt and Sterelny (1999: ss. 4.5, 5.3; 7.7). 19 A claim along these lines about demonstratives is to be found in Husserl (1900/01: 1, s. 26; vi, ss. 1–5). 20 Stanley and Szabo´ Wnd this ‘uncontroversial’ (2000a: 220–1).

Devitt, Case for Referential Descriptions 291 that is not around to be demonstrated; for example, ‘That drunk at the party was boring’. In my view, demonstrations should be treated as referring devices distinct from demonstratives and requiring their own ‘semantics’. The claim is also at odds with an idea that strikes me as unappealing: that, in each use, ‘that F ’ is equivalent to some attributive deWnite, for example, to ‘the F that I am demonstrating’. A theory of this sort is a ‘description theory’ of complex demonstratives. In eVect, it treats ‘that book’ in one of the ways that a Russellian might treat the incomplete deWnite ‘the book’. So the criticisms that I will oVer of these ways in section 7 will apply equally to description theories of complex demonstratives. But these theories have a major problem of their own. They accept the similarity that I have just urged between these demonstratives and referentially used descriptions but they respond in the opposite way. I take these descriptions to be like the demonstratives in that they are conventionally used to express singular thoughts about a particular object in mind. These description theories take the demonstratives to be like the descriptions in being conventionally used to express a general thought about whatever Wts a description. So, ‘that F ’ is not used to express a singular thought about the particular object in mind. What, then are we to say about the simple demonstrative ‘that’? Is it like the complex ‘that F ’ in this respect, or is it used to express such singular thoughts? Either way is bad news for the theory (Borg 2000: 235–6). There must surely be some conventional way of expressing these singular thoughts. Simple demonstratives are the most likely candidates.21 So it is not plausible to treat them, as the theory treats ‘that F ’, as not expressing these thoughts. Yet if ‘that’ is used to express these thoughts, the theory makes it implausibly diVerent from ‘that F ’. I talked of ‘the core’ of a token’s meaning being provided by its link to the object in mind in order to allow for there being more than this to its meaning. In discussing misdescriptions (s. 2), I assumed that ‘F ’ also contributes to the meaning of ‘the F ’, pointing out the prima-facie implausibility of claiming otherwise. The same implausibility attaches to analogous claims about referential indeWnites and complex demonstratives.22 Further reasons have been provided for thinking that ‘F ’ does indeed contribute to the meaning of ‘that F ’.23 These reasons can be carried straight over to referential descriptions. Thus, the following argument forms, which are obviously valid when the description is attributive, seem so also when it is referential: ‘All Fs are G; so, if the/an F exists it is G ’; ‘The/an F is G; so, some F is G ’; ‘The/an F is G; so, something is F and G ’. And statements of the following form seem contradictory: ‘The/an F is not F ’. It is hard to see how this could be so if ‘F ’ were not making a semantic contribution to the referential ‘the/an F ’. 21 Note that not all terms can be covered by description theories of reference. A description theory of a term passes the referential buck from that term to descriptions the speaker associates with it, simply postponing the ultimate explanation of reference. Some terms will have to be explained nondescriptively if language is to be ‘hooked onto the world’ (Devitt and Sterelny 1999: 60). Simple demonstratives are likely hooks. 22 Still, the claim about deWnites and demonstratives has its supporters, e.g. Larson and Segal (1995: 213, 340–1). 23 See e.g. Borg (2000) and Lepore and Ludwig (2000).

292

II. Referential–Attributive Distinction

How might ‘F ’ make this semantic contribution? I have always favored the view that ‘F ’ plays a role in determining the reference of the referential ‘the/an F ’, as also in determining the reference of the demonstrative ‘that F ’.24 So, on my view of singular thoughts (on which, I emphasize, the arguments of this chapter do not depend), ‘the/ an/that F ’ would designate an object that ‘F ’ applies to and that ‘the/an/that F ’ is causally grounded in by perception.25 Other possibilities suggested for ‘that F ’ take ‘F ’ to contribute independently of ‘that’. ‘That F is G ’ is treated as equivalent either to ‘That is F and G ’ or, as Ernest Lepore and Kirk Ludwig (2000) urge, to ‘[The x: x ¼ that and x is an F ](x is G )’.26 The same possibilities are available for the description ‘the/an F ’ in so far as we can treat it as implicitly containing something like the simple demonstrative ‘that’.27 I shall remain neutral between these possibilities.28 My neutrality is more extensive than this. It is common to force discussions about complex demonstratives and referentially used descriptions into a semantic framework, particularly the framework of ‘direct reference’ arising out of David Kaplan’s pioneering work on indexicals and simple demonstratives (1989a, b). Thus, it is assumed that if something is a referential description then its contribution to ‘the proposition expressed’ must be simply the object designated. I appreciate the motivation for this assumption but I am not making it. It is generally appropriate that one’s assessment of the evidence for a view should be theory-laden. However, it seems to me that there is not much theory in semantics that is well-enough established to bear this burden. In particular, I doubt that direct reference can bear it (Devitt 1996: ch. 4). So I aim to produce a case for RD that is fairly neutral about semantic frameworks. I think

24 Although I was a bit tentative about this in early writings (Devitt 1981a: 54; 1981b: 519). Borg (2000) favors this view for demonstratives. 25 In Donnellan’s cases of misdescription, we feel a tension arising from the speaker having done something right and something wrong. Neale claims: ‘The referentialist can say nothing useful here’ (1990: 91; see also Soames 1994: 154). On the contrary, the referentialist can explain the tension as I did: ‘the F ’ is causally grounded in an object that ‘F ’does not apply to. 26 There has been some discussion of an alleged diYculty for the former possibility (Richard 1993; Braun 1994; Borg 2000): sentences of the form ‘Nec(that F exists ! it is F )’, which should be false, would turn out true. If this diYculty were real, it would carry over to Lepore and Ludwig’s proposal. They argue persuasively that attention to scope ambiguities removes the diYculty (2000: 221–5). They also oVer two reasons for preferring their proposal to my favored view that ‘F’ determines reference. The Wrst, which they think decisive (213 n.), is that the view cannot account for the truth of sentences of the form ‘Nec(that F is G ! some F is G )’. But this is not so. If ‘F ’ must apply to what ‘that F ’designates then of course such a sentence will be true. Where a description plays a role in determining a word’s reference it will always yield necessities; consider e.g. ‘Nec(Jack the Ripper is G ! some murderer is G ’; ‘Nec(some bachelor is G ! some unmarried man is G )’. Second, they suggest that the view cannot accommodate the fact that ‘one can quantify into the nominal, and terms in the sentence can be anaphoric on quantiWer expressions in the nominal’ (213). They do not explain why the view cannot. 27 The soon-to-be-discussed ‘interesting diVerence’ between ‘an F ’ and ‘that F ’ poses a problem for these two possibilities for indeWnites. 28 This neutrality requires a modiWcation of the above criticism of description theories of complex demonstratives since Lepore and Ludwig’s (2000) view is, in eVect, such a theory. However, their view avoids the criticism because it allows that part of what ‘That F is G ’ expresses is the singular thought that that (the object in mind) is G.

Devitt, Case for Referential Descriptions 293 that this case is suYciently strong that any acceptable semantic framework should have to accommodate RD. Finally, although referential deWnites are very like complex demonstratives, referential indeWnites diVer from them in an interesting way. I shall discuss the diVerence brieXy. In a deictic use of ‘that F ’, as indeed of ‘that’, the speaker S wishes to convey S’s intention that the audience A should identify the object S has in mind with an object that A has in mind independently of S. A’s having the object in mind ‘independently’ rules out A’s having it in mind simply as a result of ‘borrowing’ the capacity to do so from S via the utterance. A must have some other link to the object. This independent link might have been established before the utterance or it might be immediately established by the object’s perceptual salience in the context of the utterance; for example, ‘that F ’ said while looking at, perhaps gesturing toward, a particular F. In the latter sort of case, the utterance prompts a link between A and the object that is additional to any that underlie S’s utterance. The conventions for a referential deWnite are very similar to those for the demonstrative but those for an indeWnite diVer interestingly. In using ‘an F ’ referentially, S (usually) does not convey any intention that A should identify the object S has in mind with an object that A can identify independently.29 (a) S may lack this intention because S thinks that A does not already have the object in mind and that the context of the utterance will not immediately prompt A to have it mind. Or, (b), whatever S thinks along these lines, S may not intend A to make the appropriate identiWcation. Perhaps S is concerned that A not make the identiWcation. Some philosophers have a diVerent view of indeWnites, treating a use that accords with what I have just described as the convention for referential indeWnites not as referential at all but rather as ‘speciWc’ (Bach this volume, point 6; see also Ludlow and Neale 1991: 180–3). They urge this treatment because even though S communicates that he has a certain individual in mind, S does not intend A to identify it and so does not convey a singular proposition. I think this misdescribes the missing intention. S lacks the intention that A should identify the object that S has in mind with one that A has in mind independently. But A can borrow the reference of the indeWnite from that very use of the indeWnite thus identifying the object, and S intends that A do so. As a result, S does intend to convey a singular proposition. Similarly, someone with no independent capacities to refer to Catiline or elms can borrow these capacities from a speaker’s use of ‘Catiline’ and ‘elm’. Shocking as this idea may once have seemed, I think (Devitt 2001) that it has been established by such works as Kripke (1980) and Putnam (1975). And the borrowing in all these cases has a similar causal explanation (cf. Ludlow and Neale 1991: 182 n.). If this point is granted, it is simply a verbal issue 29 I am here describing what I take to be the conventional meaning of a referential indeWnite. IndeWnites may of course sometimes be used, like a referential deWnite, with the intention that A should independently identify the object in mind; Ludlow and Neale provide some examples (1991: 176–80). I think that Bach is probably right in claiming that such uses are ‘relatively rare’ (this volume, point 6).

294

II. Referential–Attributive Distinction

whether we call these uses of indeWnites ‘referential’, as I do, because they are like referential uses of deWnites in conveying a singular proposition, or call them ‘speciWc’ (or whatever), as Ludlow, Neale, and Bach do, because they are unlike referential uses of deWnites (although not, note, referential uses of proper names) in not having to be accompanied by an intention that A should independently identify the object in mind. What matters, I claim, is that these uses of indeWnites exemplify a semantic convention other than the Russellian convention. Consider some examples of the referential use of indeWnites. My earlier remark to my wife, ‘A man in a red baseball cap stole the Encyclopedia’, is an example of (a). I use the indeWnite because I suppose that my wife has no way independent of my remark to identify the suspect I have in mind. But suppose I knew that she had been among those who had observed the man in the red baseball cap lurking in the oYce. Then I would very likely have said ‘The man in the red baseball cap . . . ’, or ‘That man in the red baseball cap . . . ’30 The diVerence between using an indeWnite and a deWnite can be subtle. Consider this adaptation of an example of Ludlow and Neale (1991: 177): S says, ‘A man is uprooting your turnips’, while looking at a man out the window. Compare this with S’s saying ‘The man is uprooting your turnips’ in the same circumstances. Suppose that A was not in a position to see the man out the window. Then the statement with the indeWnite but not the deWnite would be appropriate because A cannot make an immediate independent identiWcation of the man. Suppose next that A were beside S at the window. Then the statement with the deWnite is certainly appropriate. The interesting point to note is that the one with the indeWnite might be too, exemplifying (b). S might not care whether A makes the independent identiWcation. Or he might care but not be insisting on it in this utterance. Here is another example of (b). Both S and A found a certain person the life and soul of a party. Next day, S wishes to pass on to A some particularly juicy gossip that she obtained from that person. But she does not want to reveal her source. Saying ‘The man at the party told me’ would reveal the source, but saying ‘A man at the party told me’ would not. For a variety of reasons, people who have a singular thought about a certain F often want to express it without conveying an intention that A identify the object in mind with an object A has independently in mind; they want A to ‘open a singular Wle’, not add to one independently opened. It would be surprising indeed if there were no conventional way for them to do this. Using ‘an F ’ is a brief conventional way to do it. Using ‘a certain F ’ or ‘a particular F ’ seems to be a more lengthy way.31 It is worth noting that ‘this F ’ seems to be another way (in colloquial English at least). Thus, instead of using an indeWnite, I might have said to my wife, ‘This man in the red baseball cap stole the Encyclopedia’, and S might have said to A, ‘This man is 30 So a referential deWnite can initiate an anaphoric chain just as an indeWnite can; cf. Chastain (1975: 206). 31 This account of the diVerence between referential deWnites and indeWnites diVers from those of Chastain (1975: 206–9) and G. Wilson (1978: 67–8).

Devitt, Case for Referential Descriptions 295 uprooting your turnips’. So, ‘this F ’ not only has a referential use like ‘that F ’, which conveys the identifying intention, but one like ‘an F ’, which does not.32 My aim in this section has clearly not been to give a full explanation of the meaning of a referential ‘the/an F ’. Rather, the aim has been to identify three aspects that a full explanation must take into account: Wrst, and most important, the description’s perceptual causal link to the object in mind; second, the contribution of ‘F ’; and, third, the diVering conventions for ‘the F ’ and ‘an F ’ just described. I trust that this is enough to show that the meanings of referential descriptions are no more inexplicable and mysterious than meanings in general. Perhaps I should add that my aim has not been to specify or analyze the meanings where this is taken to involve Wnding other ways of expressing the meanings. Since the meaning of a referential description is largely constituted by its causal links to the world the only way to express its meaning may well be to use that very description (or another that is only trivially diVerent like ‘a certain F ’ or ‘the particular F ’); similarly the only way to express the meaning of a proper name may be to use it. The task should be to explain the meaning not express it (Devitt 1996: 162–3).

6. Argument IV: ‘Weak Rigidity’ DeWnites Consider Donnellan’s classic example of Jones on trial for the murder of Smith. Observing Jones’s strange behavior someone remarks, ‘The murderer is insane’. This is a referential use of a deWnite. Nathan Salmon claims that, according to RD, this sentence, as used on this occasion, is true with respect to any possible world in which Jones is insane, even if Smith is alive and well, Jones is no murderer at all, and in fact, no murders are committed by anyone anywhere. It seems quite clear, however, that the sentence ‘The murderer is insane’ is not true with respect to such a world . . . (1982: 42; see also Neale 1990: 92–3)

In brief, Salmon is arguing (1) (2) (3)

If deWnites had a referential meaning, then deWnites used referentially would be rigid. DeWnites used referentially are not rigid. So, deWnites do not have a referential meaning.

Howard Wettstein, a defender of RD (for deWnites), accepts (1) (1981: 249–50). So does Michael Nelson (1999: 578), a critic of both Wettstein and Salmon. I think that this is a mistake. I share Salmon’s intuitions about (2) and the truth-values of ‘The 32 Larson and Segal note that ‘this F ’ has a similar meaning to ‘an F ’ but take this meaning to be like that of an ‘existentially quantiWed expression’ (1995: 336). I take it to be a referential meaning.

296

II. Referential–Attributive Distinction

murderer is insane’ in the possible worlds he describes. However, far from being at odds with RD, these intuitions are just what RD predicts, given the contribution of ‘murderer’ to the meaning of ‘the murderer’. At least, they are just what RD predicts if we do not force referential descriptions into the framework of direct reference. In the last section, I noted three ways in which ‘murderer’ might contribute. First, it might play a role in determining the reference of ‘the murderer’. Then, in the actual world, ‘the murderer’ will designate Jones, the person the speaker has in mind, only if he is a murderer. And just the same should surely go for any other possible world. The referential ‘the murderer’ can no more refer in any possible world to a person who is not a murderer than can the attributive ‘the worst murderer’. So ‘The murderer is insane’ is indeed not true in the worlds Salmon describes. RD does not require that a referential deWnite be rigid, as (1) claims, but rather ‘weakly rigid’ according to the following deWnition: e is weakly rigid iV it designates the same object in every possible world in which that object exists and any descriptive element of e applies to that object. The other two ways that ‘murderer’ might contribute take ‘the murderer’ to contain an implicit demonstrative that operates independently of ‘murderer’: ‘The murder is insane’ is equivalent either to ‘That is a murderer and insane’ or to Lepore and Ludwig’s ‘[the x: x ¼ that and x is a murderer](x is insane)’. Once again ‘The murderer is insane’ is not true in the worlds described and ‘the murderer’ is not rigid. But it is ‘weakly rigid’ according to the following deWnition: e is weakly rigid iV it contains, implicitly or explicitly, an element that designates the same object in every possible world in which that object exists and any sentence containing e is true only if any descriptive element of e applies to that object in that world. Earlier I have likened the referential deWnite ‘the F ’ to the complex demonstrative ‘that F ’. So what I have just argued for deWnites should apply also to demonstratives. And so it does: ‘that F ’ is not rigid but only weakly rigid. Had the speaker in Donnellan’s example said ‘That murderer is insane’, her remark would have been no more true in the possible worlds Salmon describes than was ‘The murderer is insane’.33 Once one has accepted that the meanings of ‘that F ’ and the referential ‘the F ’ involve not only the link to the object in mind but also the nominal ‘F ’, there is no motivation for resisting the intuition (partly endorsed by Salmon) that ‘that F ’ and the referentially used ‘the F ’ are not rigid.34 This response to Salmon’s criticism of RD provides the basis of an argument for RD. The argument draws its inspiration from Kripke’s ‘lost rigidity’ argument against the description theory of proper names (1980): 33 In ‘Demonstratives’ (1989a), written in 1977, Kaplan argued for the rigidity of simple demonstratives but, so far as I can see, not for that of complex demonstratives. Still, it is natural to suppose that he believed in that rigidity. By the time of ‘Afterthoughts’ his opinion has become much more nuanced (1989b: 580–4). 34 This discussion has the following consequences (taking the deWnites to be referential). (i) If ‘The murderer is Jones’ is true it is not necessarily true in the sense of being true in every possible world where Jones exists. It is true only in worlds where Jones is a murderer; cf. Salmon (1982: 43). (ii) Although ‘The murderer did not kill anyone’ might convey a truth it is not literally true in any possible world. But this is quite compatible with the truth of ‘The murderer might not have killed anyone’ (with the scope of the deWnite wide).

Devitt, Case for Referential Descriptions 297 DeWnites used referentially are weakly rigid. Attributive deWnites are not weakly rigid. So, deWnites used referentially are not attributive.

IndeWnites An analogous argument can be made for indeWnites: IndeWnites used referentially are weakly rigid. Attributive indeWnites are not weakly rigid. So, indeWnites used referentially are not attributive. Consider my remark: ‘A man in a red baseball cap stole the Encyclopedia’. This remark is true in any possible world in which the person we actually sighted lurking about the oYce wears a red baseball cap and steals the Encyclopedia. It is not true in a world where that person is not in a red baseball cap. ‘A man in a red baseball cap’ is weakly rigid whichever deWnition is appropriate. And so is ‘a man’ in S’s ‘A man is uprooting your turnips’. Unfortunately, these ‘lost rigidity’ arguments do little on their own to support RD. For the critic of RD can resist their Wrst premises by appealing once again to Grice: although referentially used descriptions convey weakly rigid thoughts—or express weakly rigid propositions—this is not because they have referential meanings. After all, the critic points out, we might all accept that Jones’s use of ‘everyone taking my seminar’ in Neale’s example ‘Well, everyone taking my seminar turned up’ conveys a weakly rigid thought, but nobody supposes that this expression has a referential meaning. We need arguments I–III to establish that descriptions have referential meanings. Then the lost rigidity arguments bring out a respect in which those meanings diVer from attributive meanings.

7. Argument V: Incomplete Descriptions DeWnites Consider Wettstein’s example (1981: 246), ‘The table is covered with books’, said with a particular table in mind. Its deWnite is ‘incomplete’35 in that it does not uniquely describe any object and nobody would suppose that it did. Yet remarks of this sort often seem to be literally true. RD can readily explain this: ‘the table’ is referential and so it does not depend for its reference on a unique description but rather on a percep35 Donnellan (1968) called these deWnites, ‘indeWnite deWnite descriptions’. Wettstein (1981) followed this infelicitous usage. Kripke (1977) called them ‘improper’. Devitt (1981a, b) called them ‘imperfect’. The received term for them now is ‘incomplete’.

298

II. Referential–Attributive Distinction

tual causal link. Clearly, if the Russellian view cannot explain how such incomplete deWnites yield truths, we have a powerful argument against that view and hence for referential deWnites.36 The obvious Russellian explanation is that an incomplete deWnite is elliptical for a longer description that the speaker could supply. Neale calls this explanation ‘the explicit approach’. Although this approach may handle some examples, we shall soon see that it is generally hopeless. A much more promising Russellian explanation is what Neale calls ‘the implicit approach’: ‘the context of utterance delimits the domain of quantiWcation’ of the deWnite (1990: 95). But this explanation also has problems. It is important to note that any argument against these Russellian approaches faces a signiWcant constraint. The constraint arises from the uncontroversial fact that some incomplete deWnites are attributive and Russellian: they are used without any particular object in mind.37 As Neale points out (1990: 94–5), we need an account of these incomplete deWnites whatever we say about the referential uses of descriptions. Indeed, we need an account of incomplete quantiWers in general; consider, for example, the response ‘Everyone was sick’ to a query about last night’s dinner party. The signiWcant constraint is that an argument against a Russellian treatment of referentially used incomplete deWnites must not be so strong as to count equally against any Russellian treatment of incomplete attributive deWnites. For some Russellian treatment of these deWnites has to be right. In brief, there must be no overkill. I have emphasized the similarity between referentially used deWnites and complex demonstratives. The similarity is implicitly acknowledged by description theories of complex demonstratives although, of course, these theories have a very diVerent view of that similarity. These theories treat the demonstratives the way Russellian theories treat incomplete deWnites (s. 5). Indeed, whenever a Russellian theory of incomplete deWnites is proposed, an analogous theory of complex demonstratives might be proposed; and vice versa. I criticized description theories of complex demonstratives because they cannot be plausibly combined with a plausible theory of simple demonstratives. Aside from that, any description theory of complex demonstratives is as plausible or implausible as the analogous Russellian theory of incomplete indeWnites. And any criticisms of Russellian theories, like those below, apply equally to the analogous description theories of complex demonstratives. Given all this, it is interesting that we do not mostly Wnd that wherever a theory of the one sort is proposed the analogous theory of the other sort is also. For example, so far as I know, nobody has proposed a description theory of complex demonstratives that follows Neale’s implicit approach to incomplete deWnites. In criticizing Russellian theories of referentially used deWnites we can look for inspiration to problems that are, in my view, devastating for description theories of names and are potentially diYcult for description theories of any term (Kripke 1980; 36 Bach (1987/1994: 106, 125) avoids this problem by denying, implausibly, that incomplete deWnites do literally yield truths. 37 I overlooked this in Devitt (1981b).

Devitt, Case for Referential Descriptions 299 Donnellan 1972; Putnam 1975). Thus, the ‘lost rigidity’ argument in the last section was inspired by one such problem. We shall now consider criticisms related to four others: the problems of ‘the principled basis’, ‘unwanted ambiguity’, ‘ignorance and error’, and ‘theoretical redundancy’.38 In furthering the cause of RD I aim to criticize not only Russellian theories that have been proposed but ones that might be. I shall start with theories that adopt the explicit approach. Principled basis From the beginning, the description theory of names was bothered by the problem of choosing which of the many descriptions speakers associate with a name is the one that expresses its meaning and determines its reference. A similar problem faces the explicit approach to incomplete deWnites. As Wettstein (1981) argues, a speaker is likely to have many ways to complete the description and there is often no basis for determining the correct way. Neale (1990: 101) attempts to answer this criticism but, as Reimer demonstrates (1992: 352–4), the attempt surely fails. One might respond that, indeed, the speaker does not say anything determinate (W. Blackburn 1988; SchiVer 1995). But this is ad hoc and unconvincing. Gary Ostertag (1999: 126–32) brings out nicely the seriousness of the principled basis problem for the explicit approach. Unwanted ambiguity Another initial problem for the description theory of names was that any solution to the principled basis problem would be likely to select diVerent descriptions for diVerent speakers. So, implausibly, one speaker would mean one thing by the name, another speaker would mean another. Once again, a similar problem faces the explicit approach to deWnites. Any solution a theory could provide to the principled basis problem would be likely to have the implausible consequence that uses of ‘the book’ by various people in a dialogue, each with the same book in mind, had diVerent meanings because the descriptive completions for the uses would diVer.39 Ignorance and error A more recent and, in my view, much more devastating problem for the description theory of names is that in requiring competent speakers to be able to supply identifying descriptions of the referent the theory puts far too large an epistemic burden on speakers. Here is a recipe for generating examples that pose a similar problem for the explicit approach. Take a referential use of an incomplete ‘the F ’ where the object in mind is indeed F. So RD captures the intuitive view that the deWnite refers to that object. Then attribute to the speaker beliefs about the referent that are too inadequate—ignorance—or too wrong—error—to enable her to supply a completion that 38 These problems, and others, for descriptions theories are discussed in Devitt and Sterelny (1999: chs. 3 and 5). 39 This criticism relates to Wettstein’s (1981) that even if there were a correct completion of a speaker’s incomplete description, a listener would not be in the position to select it.

300

II. Referential–Attributive Distinction

uniquely describes the referent. George Wilson (1991) has generated a case of ignorance and a case of error. It is not hard to generate others. These count equally against the implicit approach and I shall postpone giving details until discussing that approach. Theoretical redundancy Some defenders of description theories of names responded to the ignorance and error problem, and to the new ‘causal theory’, by claiming that a name is equivalent to an associated description along the lines of ‘the cause of this token’. (This view is known as ‘causal descriptivism’.) Similarly, most description theories of complex demonstratives take ‘that F ’ to be equivalent to an associated description along the following lines: ‘the F pointed to by a gesture accompanying this token’ (cf. Reichenbach 1947); or ‘the F I am perceiving’ (cf. SchiVer 1978); or, allowing for the alleged rigidity of demonstratives, ‘the actual F I am demonstrating’ (cf. Neale 1993). What all these description theories have in common is that they are parasitic on plausible nondescription theories. The description theories allege that the reference of a name or demonstrative is what a description associated by the speaker denotes. If so, then the referent would have to stand in the described relation of causing the token, being pointed at, being perceived, etc. But this relation alone would then be suYcient to explain reference. And, of course, plausible nondescription theories claim that such relations are indeed suYcient to do so. Requiring the speaker to associate a description of the relation does no theoretical work. The description theories’ contribution to explaining reference is redundant. Clearly, there could be similarly parasitic theories of the referentially used incomplete deWnite, ‘the F’. They would exemplify the Russellian explicit approach. So far as I know, no such theory has yet been proposed but, given the history of theories of names and demonstratives, it seems likely that one soon will be. It is important to the case for RD to see that such a theory would have the same defect of theoretical redundancy. When one adds parasitic descriptions to the already large set of possible completions, the principled basis and unwanted ambiguity problems for the explicit approach worsen. The cumulative force of these criticisms overwhelms theories of referentially used incomplete deWnites that take the explicit approach.40 And so it would also description theories of complex demonstratives that took that approach.41 40 Consider also this neat point by Larson and Segal: ‘material that is semantically signiWcant but phonetically unpronounced is treated in discourse as if it were nonetheless overt’ but there is no sign of such treatment with deWnites (1995: 331). 41 John Perry (1977: 485–6) in eVect contemplates a nonparasitic explicit theory on behalf of Frege. G. Wilson (1991: 375–6) criticizes such a view. Lepore and Ludwig have some other criticisms of parasitic theories (2000: 207–11). Despite all these criticisms, demonstratives may on occasions be used as abbreviated descriptions. Colin McGinn (1981: 161–2) and Rod Bertolet (1987: 258–9) give nice examples of such uses. Perhaps these uses show that deictic demonstratives can be attributive as well as referential but it seems more likely that the uses are examples of attributive speaker meaning not conventional meaning.

Devitt, Case for Referential Descriptions 301 But have I gone in for overkill? Indeed, the principled basis and unwanted ambiguity problems show that the explicit approach is mostly unsuitable for incomplete attributive descriptions and incomplete quantiWers in general. Still, there is no overkill. The moral to draw is that the explicit approach is not only unable to handle referentially used incomplete deWnites but also most uses of incomplete quantiWers (although it may be able to handle some). So I turn to the implicit approach. I must avoid overkill by presenting an argument that counts against that approach to referentially used incomplete descriptions but not against that approach to incomplete attributive descriptions. First, I need to be clearer about the implicit approach. On this approach, the domain of the quantiWer is restricted by the context. There has been considerable discussion about whether this restriction is on whole sentences, clauses, or particular occurrences of a quantiWer. I am convinced by Ostertag’s argument (1999: 132–7) that the restriction is on the latter. But this still leaves a question: what determines the restriction? There has been surprisingly little attention to this question but I take it that the only plausible answer is that the domain is restricted by the speaker’s intentions:42 the speaker associates with the quantiWer some description of its domain (perhaps a description containing a demonstrative). How then does the implicit approach diVer from the explicit one? It diVers in not requiring the speaker to have settled on one particular description: the speaker may associate several diVerent descriptions provided that they yield the same restriction of the domain.43 Thus, in saying ‘Everyone was sick’, it will not matter if the speaker implicitly restricts ‘everyone’ by ‘I had dinner with last night’ and ‘in my house last night’ provided these descriptions restrict the domain to the same group of people. Should the associated descriptions provide diVerent restrictions of the domain, then there will be some indeterminacy in what is said. But it is plausible to think that there is often some indeterminacy (Ostertag 1999: 141–2). The implicit approach to referentially used deWnites seems to do well enough with the principled basis and unwanted ambiguity problems. Thus, in Wettstein’s example, the diVerent descriptions that his speaker associates with ‘the table’, namely ‘in room 209 of Campden Hall at t1 ’ and ‘at which the author of The Persistence of Objects is sitting at t1 ’, can each be seen as implicitly restricting the domain of the quantiWer to one and the same table. We do not have to choose between the descriptions. And, in Donnellan’s example, the associations ‘of Smith’ and, say, ‘now in Court 5’ might well both restrict ‘the murderer’ to Jones. But the implicit approach does no better than the explicit with the problem of ignorance and error. It is time to give more details of that problem. Consider ignorance Wrst. Each of the following cases of referential uses can be developed into plausible examples of a speaker not knowing enough about the object 42 And see Ostertag (1999: 141). 43 This diVerence undermines Neale’s suggested conversion of the implicit approach into the explicit (2000: 288–9).

302

II. Referential–Attributive Distinction

in mind to supply the needed restriction. (i) Wilson’s case of the aged and forgetful Valjean, who has many old enemies, saying ‘My old enemy is back in town’ (1991: 373). (ii) Suppose that Wettstein’s speaker made his remark about a table in a room with many similar tables. (iii) Suppose that Donnellan’s Smith had two murderers and that the other one is also in Court 5.44 (iv) Suppose that S passes on some juicy gossip by saying ‘The man at the party told me . . . ’ with a particular man in mind whom she can but dimly remember. A defender of the implicit approach may be tempted to respond to this problem by going parasitic: for example, the domain of ‘the table’ is restricted by the associated description ‘that I am perceiving’.45 But this sort of proposal acknowledges, in eVect, that the description has its reference determined in the way that RD claims while adding the theoretically redundant requirement that the speaker associate a description of that way. Next, consider error. Each of the following cases of referential uses can be developed into plausible examples of a speaker providing the wrong restriction: (i) Wilson’s case of a man saying to his wife, ‘The girl ought to be punished’, having in mind a girl he wrongly believes to be their daughter (1991: 374). (ii) Suppose that Wettstein’s speaker has various mistaken beliefs about the location of the table. (iii) Suppose that Jones murdered Robinson not Smith and is in Court 4 not 5. (iv) Suppose S wrongly believes that the source of the gossip was a man from Columbus in a black suit (he was from Cleveland in a blue suit). I have been brief in describing these cases because it is really very easy to develop them into plausible counter-examples to the implicit approach. And yet in each development, despite ignorance and error, the speaker uses an incomplete deWnite to express a thought about a particular object in mind. Is the ignorance and error argument a case of overkill? Does it count equally against the implicit approach to incomplete attributive descriptions and other quantiWers? I think not. With attributive descriptions, unlike the referential ones, any ignorance or error by the speaker really does aVect the truth of what is said.46 Thus suppose someone hears of the ghastly murder of Smith and remarks ‘The murderer is insane’. Now suppose that two people in fact murdered Smith. Any plausible restriction on the quantiWer will not achieve uniqueness and so the remark will be false on the implicit approach. This seems right. That is a case of ignorance. Here is one of error. Suppose 44 Reimer has a nice example of this sort: ‘Suppose that John and Mary have recently hired two men to do some remodelling on their home: Joe, a carpenter and Fred, an electrician. Suppose further that, after watching the two men argue all day long, John says to Mary. . . ‘‘the carpenter is not getting along with the electrician’’ ’ (1998a: 97). Now suppose that, unbeknownst to John and Mary, Joe is also an electrician and Fred is also a carpenter. 45 Ostertag seems close to this (1999: 138). 46 Bach thinks not. He has argued vigorously for a pragmatic approach according to which a speaker’s use of an incomplete quantiWer is accompanied by an implicit domain restriction which aVects what she means but not what she literally says. So the utterances of ‘The murderer is insane’ and ‘Everyone was sick’ are literally false. Whether or not the pragmatic approach to incomplete quantiWers is better than the semantic is a subtle matter; see Bach (2000b), Stanley and Szabo´ (2000a, b), and Neale (2000).

Devitt, Case for Referential Descriptions 303 that the victim is not Smith but Robinson. Any plausible restriction on the quantiWer will involve reference to Smith and so on the implicit approach the remark will be false. Again that seems right. In sum, the truth condition of an utterance involving an incomplete ‘the F ’ used referentially is unaVected by the ignorance or error of the speaker (provided that she is right that the object in mind is an F ); that of one involving an incomplete ‘the F ’ used attributively is aVected by such ignorance and error. Neale claims that the problem of incomplete descriptions is ‘essentially the same’ whether the description is used referentially or attributively (2000: 284). Perhaps, but the solutions are essentially diVerent. We can hope to solve the problem for the attributive ones within the Russellian framework by adopting the implicit approach (occasionally the explicit). There is no hope of solving the problem for the referential ones without abandoning Russell and adopting RD. RD can adopt the implicit (occasionally explicit) approach to incomplete attributive ascriptions and incomplete quantiWers in general. Neither the implicit nor the explicit approach gives a plausible account of referentially used incomplete deWnites. RD does. This is a powerful argument against the Russellian view of deWnites and hence for RD. The strength of this argument has not been appreciated by defenders of Russell47 because they have largely ignored the devastating ignorance and error problem. There can be no analogous argument for indeWnites because no issue of uniqueness arises for them.

8. Argument VI: Exportation from Opaque Contexts DeWnites Quine (1953: 139–59) claimed that the ‘exportation’ of ‘Ortcutt’ in the inference from (1)

Ralph believes that Ortcutt is a spy

(2)

Ortcutt is such that Ralph believes him to be a spy

to

seems to be ‘generally implicative’. But such an exportation is not always in order (Kaplan 1969; Sleigh 1967). Thus, consider the inference from (3)

Ralph believes that the shortest spy is a spy

to 47 Neale thinks that the argument from incompleteness is a ‘dead-end’ and ‘futile’ (2000: 284).

304

II. Referential–Attributive Distinction (4)

The shortest spy is such that Ralph believes him to be a spy.

Exportation seems to be appropriate only from an opaque ascription that requires the believer to be en rapport with the object of belief, to have a singular thought about the object. I have argued (Devitt 1981a: chs. 9 and 10) that the inference is in order provided the exported singular term, whether a name, demonstrative, pronoun, or deWnite, is referential (and nonempty).48 So the inference from (1) to (2) is in order because the name ‘Ortcutt’ is referential and that from (3) to (4) is not because the description ‘the shortest spy’ is attributive. Next, consider (5)

Ralph believes that Smith’s murderer is insane.

If this is uttered simply on the strength of Ralph’s comments at the disgusting scene of the crime, then ‘Smith’s murderer’ cannot be exported because it is attributive. But if the sentence is uttered on the strength of Ralph’s comments on observing the murderer Jones in the courtroom, then the deWnite can be exported because it is referential. The exportation works when the exported term is referential because then the opaque ascription requires that the believer be, like the speaker, en rapport with the object of belief. Rapport with an object is explained by causal–perceptual links to it. So the referential–attributive distinction bears on truth conditions and is semantic. RD for deWnites is conWrmed. So far as I know, this argument has not been addressed in the literature. We can anticipate a Gricean response. The response should not simply follow earlier ones by claiming that when (5) is uttered on the strength of Ralph’s courtroom comments the description is a quantiWer used to convey a singular thought, for it is not so used in this context. The response should be rather that the description in that utterance of (5) is a quantiWer used as a referential term like a name and hence conveying the thought that Ralph has a singular thought about Jones. This response would be plausible if the role of the description in the utterance was a rare one requiring special stage setting. But it is not. The role is a regular established one understood by the audience without any Gricean derivation. Argument VI adds to the case made in arguments I to III that there is a semantic convention of using deWnites as referential terms.

IndeWnites A similar argument can be made for referential indeWnites. Adapt Quine’s famous story about Ortcutt. Imagine Tom remarking ‘Ralph believes that a man at the beach is a spy’ to Dick in two diVerent situations. First situation: Tom is at the beach with Ralph when the man, in fact Ortcutt, is seen by them both. The man’s behavior strikes Ralph as very suspicious. Later Tom comes across Dick in the water and makes the remark. (Note that ‘the man at the beach’ would be inappropriate; s. 5.) In this situation, the 48 Names are normally referential. However some like ‘Jack the Ripper’ are attributive (Devitt 1981a: 40–7).

Devitt, Case for Referential Descriptions 305 ascription requires rapport and ‘a man at the beach’ can be exported yielding: ‘A man at the beach is such that Ralph believes him to be a spy.’ Second situation: Ralph hears that the FBI have received a tip oV that there is a man on the beach who is a spy. He reports this so credulously to Tom that Tom makes the remark to Dick. In this situation, no rapport is required and exportation is not in order. I propose that we treat the exportation of indeWnites just as we treated the exportation of other singular terms: they can be exported if they are referential (and nonempty) and not otherwise. For, where the exported indeWnite is referential, the opaque ascription requires that the believer be, like the speaker, en rapport with the object of belief. So the referential–attributive distinction bears on truth conditions and is semantic. RD for indeWnites is conWrmed.49

9. Conclusion I have considered six arguments for RD, for the view that there are referential deWnites and indeWnites, not just referential uses of them.50 I put most weight on the positive arguments I to III. IV adds little to these. V is primarily an argument against the Russellian view and has been much discussed. In the end, I claim, it does count strongly for RD (for deWnites). VI adds a bit to the case. All in all, the case for RD is very strong.51 Puzzle: given that Donnellan, Chastain, and others have made the frequency of referential uses of descriptions apparent, why are so many philosophers so committed to the view that such uses do not exemplify a convention, or at least not a semantic convention? What more could be said to show these philosophers that the uses are conventional and semantically signiWcant? What would they count as establishing this? 49 I am in general sympathy with the line that the referential–attributive distinction is irrelevant to issues of scope. See Kripke (1977); Neale (1990: ch. 4); Ludlow and Neale (1991). 50 G. Wilson (1991) has proposed an unusual argument for RD, claiming that deWnites have a certain three pronominal roles. The criticisms in Simons (1996) and Neale (this volume) make me doubt that the argument holds up. 51 Kripke made a brief argument against RD based on the anaphoric role of a deWnite in a certain dialogue (1977: 21). I responded tentatively that the deWnite might be a ‘pronoun of laziness’ (Devitt 1981b: 522). Neale (1997) has recently provided some evidence for this response.

This page intentionally left blank

PA RT I I I

Presupposition and Truth-Value Gaps

As is well known, Russell claimed that the surface subject–predicate form of a sentence such as (1) disguises the true logical form of the proposition it expresses. DeWnite descriptions are devices of quantiWcation, not devices for referring. For present purposes we can follow Russell’s own way of informally capturing his alternative analysis (1919: 176) and represent the logical form of (1) by means of the conjunction of quantiWcational sentences given in (10 ): (1) The present king of France is bald. (10 ) There is at least one present king of France, there is at most one present king of France, and whoever is a present king of France is bald. Since (10 ) entails the proposition that there is a present king of France and since this is false (when evaluated in the actual world in the year 2003), Russell would say that (1) itself is false/expresses a false proposition.1 Strawson (1950) argued that Russell’s view rests on a mistaken conXation of linguistic expression types (including sentence types) with the uses of those expression types. This led Russell to a conXation of reference and meaning, and led him to attribute properties—properties such as referring or making true or false claims—to expression types that, strictly speaking, are properties of the uses of these expression types. Strawson considers meaning to be a property of expression types, whereas he thinks that it is people who refer or fail to refer by their uses of expression types, and who make true or false assertions by their uses of sentence types in certain conversational contexts.2 1 Russell’s analysis of ‘The F is G’ is standardly formulated in modern predicate logic as follows: 9x[F (x) ^ 8y(F (y) ! y ¼ x) ^ G(x)]. Kent Bach has pressed on us the point that the fracturing of syntax imposed by Russell’s own formulations of his theory, and by modern reformulations of it in Wrst-order logic, is inessential to the key idea of the theory. This key idea is that ‘the F ’ is a device of quantiWcation, on a par with ‘every F ’, ‘no F ’, etc. Neale (1990) makes this point forcefully by representing ‘The F is G ’ in his restricted quantiWer notation as follows: [Thex : F (x)] G(x). However, Russell’s ‘fractured’ syntax has the virtue of making (more) explicit the entailment relations that Russell thinks hold between (1) and the existential proposition that there is a present king of France. 2 Two points should be noted. First, Strawson’s view is in fact more complicated than this. He distinguishes also between the use of an expression and the utterance of an expression. Uses are more like social practices that can exist for extended periods of time (e.g. there can be a use of ‘the king of France’ to refer to Louis XVI that extends throughout the reign of Louis XVI). On the other hand, utterances of expressions are particular dated

308

III. Presupposition and Truth-Value Gaps

Strawson thinks that for a speaker to be able to use (1) felicitously to make an assertion, certain presuppositions must be satisWed. In particular, the use of the deWnite description in (1) presupposes that there is a present king of France. Since this presupposition fails in the actual world in the year 2003, a speaker would not succeed in making a truth-evaluable claim by uttering (1) in these circumstances. Instead, the speaker’s utterance would lack a truth-value—there would be a truth-value gap. Strawson (1950) does not use the terms ‘presuppose’ and ‘presupposition’. Rather, he says that to utter (1) is to imply that there is a king of France. But this is a very special and odd sense of ‘imply’. ‘Implies’ in this sense is certainly not equivalent to ‘entails’ (or ‘logically implies’). And this comes out from the fact that when, in response to this statement, we say (as we should) ‘There is no king of France’, we should certainly not say we were contradicting the statement that the king of France is [bald]. We are certainly not saying that it is false. We are, rather, giving a reason for saying that the question of whether it is true or false simply does not arise. (1950: 330, emphases in original)

In later writings, Strawson (1952, 1954) does use the term ‘presupposition’ and deWnes it as Frege (1892/1970: 69) does: A presupposes B iV A is neither true nor false unless B is true. This has come to be known as the semantic conception of presupposition, and contrasts with the pragmatic conception of presupposition defended by Stalnaker (1970, 1973, 1974, 1978, 1998, 2002).3 Unfortunately, Strawson (1954, 1964) complicates matters somewhat by admitting that certain uses of nondenoting deWnite descriptions would result in claims that are false, rather than neither true nor false. He oVers examples of the following sort, where speech events. Second, while it is true that Russell equates the meaning of a logically proper name with its referent (a view that Strawson (1950) heaps scorn on), Russell does not habitually talk of sentences making true or false claims. He is much more likely to talk of sentences as expressing true or false propositions or simply to talk of sentences and propositions interchangeably. One might say that it is Strawson who is insensitive to the distinction between making an assertion and expressing a proposition. This is controversial since not all philosophers think it is correct the take a speech-act oriented approach to semantics, as does Strawson. Salmon (this volume) contrasts two conceptions of semantics, namely a speech-act centered conception and an expression centered conception. The latter conception, favored by Salmon and arguably by Russell, is one ‘according to which the semantic attributes of expressions are not conceptually derivative of the speech acts performed by their utterers, and are thought of instead as intrinsic to the expressions themselves’ (this volume: p. 238). 3 The pragmatic conception makes the notion of speaker presupposition primary. A sentence can be said to carry the presupposition P just in case the use of that sentence would normally be inappropriate unless the speaker were presupposing P. A speaker presupposes P in a conversational context just in case she believes P to be a part of the common ground. P belongs to the common ground if all the conversational participants accept that P and it is common belief that everyone accepts that P. To accept that P is not necessarily to believe that P, but only to act for some reason as though P were true. Note also that for something to belong to the common ground it need not be generally accepted prior to the speaker’s utterance. It may only be as a result of the speaker’s utterance that P becomes generally accepted. For example, if a speaker utters the sentence ‘I have to pick up my sister at the airport’, the speaker need not assume that it was part of the common ground prior to her utterance that she has a sister. But she will assume that it will be part of the common ground once she has uttered the sentence, since her hearer is likely to accommodate her by coming to accept this proposition. Two excellent recent discussions of Stalnaker’s views, which do a great deal to clarify the relations between sentence and speaker presupposition and the notion of presupposition accommodation are von Fintel (2001b) and Simons (forthcoming).

III. Presupposition and Truth-Value Gaps

309

we are to understand that there is no question that my friend really exists or that there really was an Exhibition yesterday: (2) (3)

My friend went for a drive with the king of France. The Exhibition was visited yesterday by the king of France.

Two issues are raised here. One concerns the conXicting truth-value intuitions that Russell and Strawson had about (1). Who is right? Kai von Fintel (this volume) argues that this is not an issue that can be solved by a direct appeal to truth-value intuitions. Support for the idea of truth-value gaps will have to be provided indirectly, by appeal to more complicated facts such as presupposition projection facts. The so-called projection problem for presuppositions is to predict and explain the presuppositions of complex sentences on the basis of the presuppositions of their parts. Under what conditions do the presuppositions of sentences survive when those sentences are embedded in larger constructions? It is well known that presuppositions survive under certain sorts of embeddings (these act as ‘holes’), whereas certain other sorts of embeddings block presuppositions (these act as ‘plugs’), and yet others allow only some presuppositions to survive (these act as ‘Wlters’). See Heim (1983b), Karttunen (1973). For example, the existence presupposition associated with (1) persists even when (1) is the sentential complement of an attitude verb (‘I hope the present king of France is bald’) or the antecedent of a conditional (‘If the present king of France is bald, Jim will be upset’). Von Fintel suggests that distinguishing between falsity and truth-valuelessness is supported because it enables us to oVer a theory of presupposition that gives the best explanation for the projection behavior of presuppositions and other pragmatic facts.4 A second issue raised by examples (1)–(3) is a theory-internal problem for those who accept Strawson’s semantics for deWnite descriptions. How is a Strawsonian to account for the fact that (1) gives rise to a truth-value gap, whereas (2) and (3) are judged to be straightforwardly false? This second issue is the central concern of both Atlas and von Fintel in their chapters in this volume. As von Fintel puts it, we need an explanation for why we are squeamish about giving a truth-value judgment in the case of (1), whereas we are conWdent in our judgments that (2) and (3) are false.5 Let us call sentences such as (2) and (3) ‘F-sentences’, and sentences such as (1) that we are squeamish about ‘#-sentences’. One explanation oVered by Strawson himself, as well as by Reinhart (1981), Gundel (1977) and Horn (1989), is that in F-sentences there is no presupposition of existence associated with the use of the deWnite, whereas 4 Kent Bach points out that to assume that description sentences carry presuppositions is already to assume that Russell’s truth-value intuitions are incorrect, if ‘presupposition’ is understood in the semantic sense, according to which A presupposes B iV A is neither true nor false unless B is true. Russell would not agree that (1) presupposes that there is a king of France in the semantic sense, since even when this ‘presupposition’ is false Russell would maintain that (1) has a determinate truth-value. 5 The term ‘squeamish’ that von Fintel uses is suggested by an alternative way that Strawson has of expressing his intuition that (1) lacks a truth-value. Strawson writes: ‘[w]e feel very squeamish indeed about The King of France is bald presented abruptly, out of context’ (1964: 114).

310

III. Presupposition and Truth-Value Gaps

in #-sentences there is a presupposition of existence. The diVerence is said to lie in the topic/focus articulation of these sentences.6 In #-sentences the deWnite is in topic position, whereas in F-sentences it is in focus position. The claim is that all and only topical deWnites carry existence presuppositions. When these existence presuppositions fail, this gives rise to truth-value gaps. Von Fintel oVers a test for speaker presupposition that he calls the ‘Hey wait a minute’ test.7 When submitted to this test, F-sentences are seen to carry presuppositions, just as #-sentences do: (4)

A: My friend went for a drive with the king of France. B: Hey, wait a minute. I had no idea that France is still a monarchy. B: #Hey, wait a minute. I had no idea that she knew him.

Moreover, the same presupposition projection facts that hold for #-sentences hold for F-sentences. An existence presupposition projects out of the embedding constructions in (5) and (6): (5) (6)

I hope that my friend went for a drive with the king of France. If my friend went for a drive with the king of France, then her sister will be envious.

6 The topic–focus distinction (sometimes referred to as the topic–comment distinction) has long been of interest to those who study the pragmatics of natural language. This distinction concerns the way in which the information conveyed by an utterance is structured. Part of this information (the topic) can be relegated to the background, since it is information that is already established in the common ground shared by the conversational participants. But part of this information (the focus) will be new and it will need to be foregrounded. There are many conventional ways of marking the topic–focus distinction. In English, for example, the subject expression is usually, though by no means invariably, used to mark the topic. The predicate expression is then used to make some sort of comment on the topic. It-clefting is another device for indicating topic–focus structure. ‘It was Mary who visited London’ and ‘It was London that Mary visited’ arguably have the same truthconditional content, but they are associated with diVerent information structures. In the former case it is already established that someone went to London, and the new information being asserted is that Mary was that person. In the latter case, it is already established that Mary visited some place, and the new information being asserted is that London is that place. Among other devices for marking the topic–focus distinction are stress and/or intonation (‘Mary went to london’). In some languages, e.g. Japanese, there are topic-markers. These are particles that attach to expressions to indicate the topichood of the referents of those expressions. The notion of topic can be found in ancient Greek rhetorical treatises and in medieval Latin works on logic, rhetoric, and grammar. The notion of focus most likely has its roots in 19th-century experimental psychological treatises on attentional focusing. One of the Wrst to articulate these ideas within the context of modern linguistics is Chafe (1976). 7 Following Stalnaker, von Fintel deWnes a speaker presupposition as something that the speaker takes to be a part of the common ground. (See n. 3 above.) von Fintel adds to Stalnaker’s account by connecting pragmatic and semantic presuppositions. This is done via a bridge principle that connects the partial semantics of sentences to pragmatic facts about presupposition. A description sentence of the form ‘The F is G ’ expresses a partial proposition that is deWned only for worlds in which there is a unique F and that is true in a world w iV the unique F in w is G in w. Thus ‘The F is G ’ semantically presupposes that there is a unique F, since it has no truth-value in those states of aVairs in which there is no unique F. The principle that connects this semantic fact to pragmatic presupposition is one that states that a sentence can only be used in smooth communication if its presuppositions are entailed by the common ground prior to the utterance of that sentence (or perhaps after the utterance event but prior to the acceptance or rejection of that sentence—see von Fintel 2001b). According to this principle, any semantic presupposition becomes a pragmatic presupposition, although not all pragmatic presuppositions need to be grounded in semantic presuppositions.

III. Presupposition and Truth-Value Gaps

311

What about the second strand of the Reinhart/Gundel/Horn view, that presuppositionality is tied to the topic–focus distinction and that all and only topical deWnites carry existence presuppositions? The following appears to be a counter-example: (7)

I had breakfast with the king of France this morning. The king and I both had scrambled eggs.

After the utterance of the Wrst sentence, the king of France will have become topical. So the topic–focus account should say that the second sentence in (7) carries an existence presupposition. Since this presupposition fails, the second sentence should be a #sentence. But it isn’t. It is clearly false. In the course of his chapter von Fintel presents many other examples that are problematic for the topic–focus account. For example, the following sentences have diVerent topic–focus structures, and yet we would judge them both to be false: (8) (9)

Among the bald people in the world is the king of France. The king of France is one of the bald people in this world.

After considering and rejecting an alternative solution to Strawson’s problem suggested by Lasersohn (1993), von Fintel oVers his own solution to the problem. His own view is that (1) and (2) do not diVer in the presuppositions they carry. Thus if we accept Strawson’s semantics for deWnites, we must recognize that both induce truthvalue gaps. Moreover, both are rejected for the same reason, namely that there is no present king of France, and not because of any conXict with entrenched information (as Lasersohn claims). However, perhaps there is some pragmatic explanation for our diVering truth-value intuitions regarding (1) and (2). If a sentence is oVered that has a false presupposition, we must reject that sentence. But, von Fintel notes, there are two ways to do this. We can directly challenge the mistaken supposition. Alternatively, if the sentence has entailments that can be falsiWed independently of discussing its presupposition, we can focus on these mistaken entailments. Where the second option is available, we prefer to take it rather than to engage in a direct challenge. It is these cases (the independently falsiWable ones) that give rise to a judgment of falsity. Because this judgment is the result of following a certain conversational (pragmatic) strategy, von Fintel says that (2) is pragmatically false, even though semantically it induces a truth-value gap. When the second option is not available, this gives rise to a feeling of squeamishness. In other words, although our truth-value intuitions cannot be used directly to settle the dispute between Russell and Strawson, they do reveal something about the conversational strategies we employ. F-sentences then are ones that are independently falsiWable; they provide an independent foothold for rejection. In these cases, von Fintel claims, there is a contextually salient entity whose properties are in principle suYcient to falsify the sentence. He goes on to spell out what sort of thing can provide a foothold for rejection. For instance, it does not have to be an individual, but can be a set of individuals or a particular episode or situation. He points out that it must be possible to falsify the sentence by inspection

312

III. Presupposition and Truth-Value Gaps

of that independent entity. So the mere mention of an independent entity is not suYcient. Moreover, for there to be an independent foothold it must be possible to falsify the sentence without at the same time showing that its presupposition is false. In some cases the independent foothold can be a general fact, so long as this is something we believe independently of any belief in the falsity of the presupposition (e.g. ‘The king of France can jump 100 feet into the air unaided’ is an F-sentence, because it is falsiWed by the general fact that no human being can jump 100 feet into the air unaided). Finally, the independent foothold does not have to be mentioned in the same sentence or even explicitly mentioned at all. One consequence is that even (1) can in certain contexts turn out to be an F-sentence, because some preceding sentence has made an entity salient that can act as an independent foothold for rejection. For example, if someone asks what reigning monarchs are bald, and someone else replies by uttering (1), this would be judged false, because the set of reigning monarchs introduced by the questioner provides an independent foothold for rejection. Atlas (this volume) takes an opposing position. He wishes to defend a topic–focus account of the distinction between F- and #-sentences, although he rejects the strong claim that all and only topical deWnites carry existence presuppositions. He rejects the claim that all topical deWnites carry existential presuppositions, but accepts the claim that only topical deWnites carry such presuppositions. He calls the claim that he accepts the Strawson–Grice (S–G) condition, and formulates it as follows: S–G Condition: The existence of a reference for an NP is presupposed in making a statement only if the NP is a topic NP (where ‘NP’ is a meta-variable ranging over proper names and simplex deWnite descriptions). Thus Atlas would not be embarrassed by counter-examples such as example (7) mentioned by von Fintel. Atlas provides his own counter-example: (10)

There is a queen of England, and the queen of England is gracious.

Atlas notes that this conjunction does not carry an existential presupposition, and yet it has the queen of England as its topic.8 Atlas oVers the following pair of sentences as examples involving nondenoting deWnite descriptions for which we have diVering truth-value intuitions: (11) (12)

The king of France is bald. It is the king of France who is bald.

Atlas argues that (11) is judged to be neither true nor false, whereas (12) is judged false. His explanation is that the cleft sentence (12) has a diVerent topic/focus structure than 8 Some would say that the existence presupposition carried by the deWnite in the second conjunct is blocked in this case. This is because the second conjunct is evaluated in a context that has already been updated with the information that the queen of England exists. Since the queen’s existence has already been asserted there is no need to presuppose it.

III. Presupposition and Truth-Value Gaps

313

(11). In the cleft sentence the king of France is in focus position. Hence by the S–G condition it fails to carry an existence presupposition, and (12) is judged false. Atlas goes on to discuss a class of cases mentioned by Donnellan (1981), which Donnellan claims induce truth-value gaps. These are cases in which there is a comparison of an existent entity with a nonexistent entity, such as: (13)

The queen of England knows more about baseball than does the king of France.

Atlas tries to save Donnellan’s truth-value intuition about this case, but concludes that Donnellan is simply mistaken. (13) is not truth-valueless. Atlas’s own explanation is that in these cases the term referring to the nonexistent entity is not in topic position. So by the S–G condition (13) does not carry an existence presupposition. So the fact that there is no king of France does not induce a truth-value gap. Clearly here Atlas and von Fintel are at odds with one another. We have already seen that von Fintel would say that even F-sentences carry existence presuppositions, and he uses his ‘Hey, wait a minute’ test to show that they carry existence presuppositions. If we apply this test to (13), we get: A:

The queen of England knows more about baseball than does the king of France. B: Hey, wait a minute, I didn’t know that France was still a monarchy. B 0 : # Hey, wait a minute, I had no idea she was so much better than him on this topic. This would seem to indicate that (13) does indeed carry an existence presupposition. However, Atlas mentions his own test for presupposition, and by applying it to (13) concludes that (13) lacks an existence presupposition. Atlas’s test is that if (13) has an existence presupposition it should be preserved under negation. But, Atlas claims, the negation of (13) equally lacks an existence presupposition and that, far from being truth-valueless, (14) may be true: (14)

The queen of England does not know more (knows no more) about baseball than does the king of France.

We leave readers to decide for themselves who is right. We will say only that von Fintel could agree with Atlas’s judgment that (14) is true, but he would have to say that this is merely pragmatic truth. The queen of England is a contextually salient entity whose properties provide an independent foothold, not for rejection in this case, but for acceptance. (14) can be heard as an exaggerated way of saying that the queen of England knows nothing about baseball. And we can establish that the queen knows nothing about baseball without addressing the issue as to whether there is or is not a king of France. Atlas’s main focus in his chapter is with a class of sentences that von Fintel does not discuss at all, namely negative existential claims involving nondenoting terms, such as:

314

III. Presupposition and Truth-Value Gaps (15)

The present king of France does not exist.

Strawson would say that (15) is true rather than truth-valueless. As we have seen, Strawson would say that if there is no truth-value gap this must be because there is no existence presupposition. However, Atlas notes that Strawson never explained why (15) and its ilk lack existence presuppositions. Atlas claims that if we can show that the deWnite description in (15) is not in topic position, then the S–G condition will yield the conclusion that (15) lacks an existence presupposition. He goes on to discuss various linguistic tests for topichood. He concludes that in ‘The king of France exists’ and ‘The king of France does not exist’, the topic concerns what does or does not exist respectively, and hence in both cases the deWnite description is in focus rather than topic position.9 The chapters in this section show that what might have seemed to be rather tired old examples, such as ‘The king of France is bald’ and ‘The king of France doesn’t exist’, are in fact still capable of opening a window onto a range of very interesting pragmatic and semantic issues. 9 For a diVerent approach to the presuppositions of existence attaching to deWnite descriptions and to negative existentials such as (15), see Asher and Lascarides (1998b). They oVer a novel analysis of presuppositions using the framework of segmented discourse representation theory (SDRT) Wrst outlined in Asher (1993). In the case of a sentence such as (1), they argue that the existence presupposition carried by the deWnite description attaches to the discourse representation structure (DRS) for (1) via the rhetorical relation Background. But straightforward accommodation of the presupposition via Background is not possible in the case of (15), as the two states of the king existing and not existing cannot overlap. However, the relation of Background can hold between the presupposed information and the interpreter’s concepts, in this case the concept the King of France. As evidence in support of this analysis Asher and Lascarides note the infelicity of ‘The man does not exist’ when uttered discourse initially. In this case there ‘isn’t enough content associated with the description to provide the necessary thematic continuity and to bind the presupposition to any particular concept via Background ’ (Asher and Lascarides 1998b: 295).

8 Would You Believe It? The King of France is Back! (Presuppositions and Truth-Value Intuitions) Kai von Fintel

This chapter is concerned with the contrast between two kinds of sentences involving deWnite descriptions. When people are asked to assign truth-values, they feel ‘squeamish’ about The king of France is bald, while they conWdently and without hesitation judge My friend went for a drive with the king of France last week to be false. This contrast has often been taken to show that the Wrst sentence has no truth-value because its presupposition (that there is a king of France) is not satisWed, while the second sentence has no presupposition of existence (of a king of France) and is thus simply false. I will argue against this picture. Even the second sentence carries an existence presupposition. Rather than concluding that therefore the truth-value gap analysis of presupposition is incorrect, I will say that the second sentence actually does have no truth-value. The judgment people give is based on a fall-back strategy that assigns truth-values even to sentences that strictly speaking have no truth-values. Much of the chapter is concerned with describing this fall-back strategy.

1. Background: A Plausible Theory of Presupposition Sentences with deWnite descriptions such as those in (1) and (2) have two distinct meaning components, one of which involves a presupposition, at least that is what I will assume, indeed presuppose, without argument. This chapter grew out of an idea that occurred to me in the Fall 1996 semantics introduction at MIT. Since then, I have talked about this work at Indiana University, the Max Planck Institute in Nijmegen, the ZAS in Berlin, the University of Maryland, Yale University, the University of Tu¨bingen, and Stanford University. I thank the audiences at these occasions for their interest, comments, and criticism. I thank Danny Fox, Irene Heim, David Pesetsky, Orin Percus, Mats Rooth, and San Tunstall for helpful discussions.

316

III. Presupposition and Truth-Value Gaps (1) (2)

The king of France is bald. The mathematician who proved Goldbach’s Conjecture is a woman.

The sentence in (1) presupposes that there is a king of France and asserts that he is bald. The sentence in (2) presupposes that there is a mathematician who proved Goldbach’s conjecture and asserts that that mathematician is a woman. Of course, when we talk of a sentence doing things like ‘asserting’ and ‘presupposing’, that can only be some sort of shorthand description. It is really language users who are doing such things. What we should say is that something in the semantics of these sentences constrains the uses that speakers can put these sentences to. A theory of presupposition is concerned with this interplay of semantics and pragmatics. In particular, such a theory speciWes (i) what the pragmatic facts are that need to be explained, (ii) what it is in the semantics of these sentences that constrains language users in such a way as to produce the pragmatic facts about presupposition, and (iii) how the semantics is connected to the pragmatics. I will sketch my favorite theory of presupposition without suggesting that one should not explore alternatives.1 What this chapter is about is what role truth-value intuitions play in the investigation of presuppositions. As far as I can see, this issue arises in the alternatives to my preferred theory just as much and thus even followers of these alternatives should Wnd something useful in my discussion.

The Pragmatic Data about Presupposition The main empirically observable facts that motivate theories of presupposition are two: (i) sentences like (1) and (2) are hard to use felicitously unless the speaker takes it for granted that the ‘presupposed’ component of meaning is already common ground among the participants in the conversation (as Stalnaker puts it: these sentences ‘require a speaker presupposition’);2 (ii) the fact in (i) persists even when such sentences occur embedded in larger constructions (this phenomenon is known as ‘presupposition projection’). Usually when one is asked to appreciate the fact that (1) and (2) are normally taken as signaling certain things about what the speaker is taking for granted, one is supposed to just consult one’s intuitions about felicitous uses of these sentences. But there are at least some suggestive tests. Observe for example:3 1 The theory I sketch is largely based on the writings of Karttunen (1974), Stalnaker (1970, 1973, 1974, 1978, 1998, 2002), and Heim (1982, 1983, 1990b, 1992). The one ingredient that I add is to ground pragmatic presuppositions in semantic/logical presuppositions (at least for deWnite descriptions). For recent surveys of the area see Soames (1989b) and Beaver (1997). 2 I do not wish to get bogged down in debates about the proper characterization of the pragmatic facts about presupposition. Some discussion of related matters, especially those concerned with the phenomenon of presupposition accommodation, can be found in von Fintel (2001b) and Stalnaker (2002). What is important here is that there are such facts and that we recognize them when we see them. 3 The ‘Hey, wait a minute’ test is a variation of a criterion proposed by Shanon (1976).

von Fintel, Would You Believe It? 317 (3)

A: The mathematician who proved Goldbach’s Conjecture is a woman. B: Hey, wait a minute. I had no idea that someone proved Goldbach’s Conjecture. B 0 : #Hey, wait a minute. I had no idea that that was a woman.

Hearer B legitimately complains that A presupposed that someone proved the conjecture, when it was not in fact established prior to A’s utterance. Hearer B 0 illegitimately makes a parallel complaint about an asserted, non-presuppositional component of A’s statement. While asserted, non-presuppositional components of meaning are manipulated at will by the embedding construction, presuppositional components are typically inherited by the whole construction. Two sample environments are the complements of attitude verbs like hope and the antecedents of conditionals: (4)

(5)

a. I hope that the king of France is wise. b. If the king of France is wise, James (the Francophobe) will be quite disappointed. a. I hope that the mathematician who proved Goldbach’s Conjecture is a woman. b. If the mathematician who proved Goldbach’s Conjecture is a woman, James (the misogynist) will be quite disappointed.

In (4), it is not part of what is hoped for that there is a king of France; instead it is assumed background that there is a king of France. In (5), again it is not part of the hope (but a background belief ) that someone proved Goldbach’s Conjecture. Similar observations can be made about the conditional examples. Here, the existence of a king of France or of someone who proved Goldbach’s Conjecture is background and the distinction made is between situations where the king is wise or not and between situations where the person who proved Goldbach’s Conjecture is a woman or a man.4

The Semantic Source of the Pragmatic Facts about Presupposition I will assume that the source (or one source) of the pragmatic facts I just sketched lies in the semantics of sentences. I will assume that sentences express partial propositions that sometimes assign no truth-value to a state of aVairs or possible world.5 One says 4 Again, there is of course much more to be said about presupposition projection, but I trust that my readers are familiar enough with the phenomenon. 5 Other analyses do not start with partial propositions as the semantic values. They assume a Russellian twovalued semantics. Facts about pragmatic presuppositions are encoded at an independent level. This idea was perhaps Wrst suggested by Sellars (1954) in his response to Strawson (Sellars should therefore be cited as an early advocate of a dynamic perspective on communication): ‘Russell’s analysis is correct, but needs to be supplemented by an account of the conventions relating to the dynamics of discussion or argument, the order in which assertions should be made and challenged.’ There are a number of such theories on the market. Perhaps the best known is Karttunen and Peters’ system where a sentence is associated with a pair of propositions as its semantic value: the ordinary content and the presupposed content (Karttunen and Peters 1979).

318

III. Presupposition and Truth-Value Gaps

that a sentence has the semantic presupposition that p iV the proposition it expresses does not assign a truth-value to states of aVairs where p does not hold. I will work with a Frege–Strawson semantics for deWnites: (6)

The P is Q expresses a partial proposition which is deWned only for worlds in which there is a unique P and which is true in a world w iV the unique P in w is Q in w.

I will discuss shortly that this semantics is not one that we can argue for on the basis of raw truth-value intuitions. Rather its advantages lie in how well it can be used to derive the pragmatic facts about presupposition.

The Semantics–Pragmatics Interface The way that we connect the partial semantics of sentences to the pragmatic facts about presupposition is via a bridging principle. This has been hinted at by Stalnaker in a number of places: a sentence with a semantic presupposition can only be used in smooth communication if the truth of the presupposition has been established prior to the utterance of the sentence. According to this principle, any semantic presupposition becomes a pragmatic presupposition.6 The other fact that needs to be explained is the projection behavior of a presupposition. Here there are two ways to go. (i) One can locate the mechanics in the semantics so that projection then is simply the outcome of compositional semantics. Embedding operators in general would ‘pass on up’ the partiality of their complements, with the exceptions of the ones that Karttunen called plugs and Wlters. One thus computes a partial proposition expressed by the larger structure, which gets turned into a pragmatic presupposition via the bridging principle. (ii) The alternative is a dynamicpragmatic account of projection a` la Stalnaker or Heim. Here one thinks of the ingredient clauses in the larger construction as independently accessed by the pragmatics. For example, when a conjoined sentence is asserted, in eVect there is a sequence of two assertions. Since the second assertion occurs after the Wrst one, any presuppositional requirements it may have need to be satisWed by the context as aVected by the Wrst assertion. I will not explore the choice between these options here. Even though my presentation has been very sketchy, I hope that it is not too hard to imagine that a plausible theory of presupposition can be built that is grounded in a partial semantics for sentences like (1) and (2) and that derives the observable pragmatic facts about presupposition in a principled manner from that partial semantics. 6 Soames (1989b) argues that this principle, which is assumed to be quite natural by Stalnaker, is in fact not one that follows with any kind of conceptual necessity. Maybe it is better seen as an irreducible property of the way natural language is used. Note also that not all pragmatic presuppositions need to be grounded in semantic presuppositions. The principle goes one way only: all semantic presuppositions trigger pragmatic presuppositions.

von Fintel, Would You Believe It? 319

2. The Trouble with Truth-Value Intuitions One might have thought that a particularly direct argument for the theory of presupposition that I just sketched could be based on truth-value intuitions. The theory says that when the presupposition of a sentence like (1) or (2) is not satisWed in a given world, the partial proposition expressed by the sentence fails to assign a truth-value to that world. So, we should expect that people judge these sentences to have no truthvalue in the relevant circumstances. (1), for example, should be judged to have no truth-value given that there is currently no king of France. This is in fact what Strawson claimed to be the case in ‘On Referring’, the paper that largely inaugurated the modern study of presuppositions. He wrote the following about an example that is a variant of (1): Now suppose someone were in fact to say to you with a perfectly serious air: The King of France is wise. Would you say, ‘That’s untrue’? I think it is quite certain that you would not. But suppose that he went on to ask you whether you thought that what he had just said was true, or was false; whether you agreed or disagreed with what he had just said. I think you would be inclined, with some hesitation, to say that you did not do either; that the question of whether his statement was true or false simply did not arise, because there was no such person as the King of France. You might, if he were obviously serious (had a dazed astray-in-the-centuries look), say something like: ‘I’m afraid you must be under a misapprehension. France is not a monarchy. There is no King of France.’ (Strawson 1950: 330)

In a later paper, Strawson reported his intuition in less theoretically laden terms: ‘[w]e feel very squeamish indeed about The King of France is bald presented abruptly, out of context’ (1964: 114).7 Russell, however, denied the intuition that the relevant sentence is neither true nor false: Suppose, for example, that in some country there was a law that no person could hold public oYce if he considered it false that the Ruler of the Universe is wise. I think an avowed atheist who took advantage of Mr. Strawson’s doctrine to say that he did not hold this proposition false would be regarded as a somewhat shifty character. (Russell 1959: 243–4)

A natural reaction at this point is to consider people’s intuitions about truth-value gaps to be too unreliable to merit scientiWc attention.8 This does not at all mean that semantic theories that assume the existence of truth-value gaps are thereby discredited. Far from it. But argumentation cannot and should not proceed by simply polling naive speakers about whether sentences are false or neither true nor false. Unsurprisingly then, the attention of research on presupposition has been directed elsewhere, 7 We need to do some philological digging to Wnd why, how, and when the central example changed from attributing wisdom to the king to reporting his loss of hair. 8 ‘Direct reXection is obviously of little use, since Russell and Strawson seem to have honestly had opposite intuitions in this case’ (Thomason 1990: 327).

320

III. Presupposition and Truth-Value Gaps

primarily toward Wnding an adequate analysis of the projection behavior of presuppositions.9 A clear statement of the state of aVairs is found in Soames’s dissertation: We have reliable intuitions distinguishing truth from untruth. If we also had reliable intuitions distinguishing falsity from truth-valuelessness, then we could evaluate the claim that [a relevant sentence] is truth-valueless rather than false. Since I don’t think we have such intuitions, I don’t think that we can directly evaluate this claim. Logical presupposition is not a relation about which we have direct intuitions, but rather is a theoretical construct that can be used to account for certain inferences. (Soames 1976: 169) In considering simple sentences (including simple quantiWcations) one would have no reason to abandon bivalence if doing so did not facilitate the account of larger sentences. If one were only concerned with simple sentences, and nothing more complex, then one could simply write into the truth deWnition whatever is necessary for these sentences to be true, letting all failures of truth be instances of falsity. This is troublesome only when diVerent reasons for a sentence’s being untrue aVect whether or not some more complex sentence is true. Distinguishing between falsity and truth-valuelessness allows us to mark such diVerences. (Soames 1976: 202–3)10

We should thus conclude that attributing partial propositions to sentences as their semantic values is not something that is subject to secure and direct intuition. The move is a theoretical one put forward as a useful and perhaps necessary ingredient of an overall theory of presupposition, which is primarily concerned with Wnding the best available explanation of the projection behavior of presuppositions and the pragmatic facts I alluded to earlier. This is of course a not too unfamiliar picture: a theory whose basic assumptions are not subject to direct intuition but that is globally supported by indirect argumentation based on more complicated facts.

3. Strawson’s Contrast But things do not stay that simple. Consider some additional examples oVered by Strawson in work following up on his 1950 paper: Suppose, for example, that I am trying to sell something and say to a prospective purchaser The lodger next door has oVered me twice that sum, when there is no lodger next door and I know this. It would seem perfectly correct for the prospective purchaser to reply That’s false, and to give as his reason that there was no lodger next door. And it would indeed be a lame defense for me to say, Well, it’s not actually false, because, you see, since there’s no such person, the question of truth and falsity doesn’t arise. (Strawson 1954: 225) 9 This is, I take it, the consensus view of presupposition experts. It is not, however, a point that has yet to be widely recognized outside this circle. Quite often, e.g. when relevant claims in the syntax/semantics literature— that a particular expression or construction carries a presupposition—are debated, there are appeals to intuitions about truth-value gaps rather than a presentation of projection facts. See von Fintel (2001a) for an example. 10 Soames reports that Dummett (1967: 61–3, 1973: 417–29) comes to the same conclusion.

von Fintel, Would You Believe It? 321 Suppose I am ignorantly boasting about my friend’s visit to Rome and mention the King of France as one among the distinguished people he had seen there. I might say He had lunch with the prime minister, had an audience with the pope, and then went for a drive with the King of France. Someone might say, Well, at least it’s false (not true) that he went for a drive with the King of France—for there’s no such person. (Strawson 1954: 226) For example, there is no King of France; and there is, let us say, no swimming-pool locally. But there is, let us say, an Exhibition in town; and there is, let us say, no doubt of Jones’ existence. If we consider the statements (i) Jones spent the morning at the local swimming-pool (ii) The Exhibition was visited yesterday by the King of France it may seem natural enough to say (i) that it is quite untrue, or is false, that Jones spent the morning at the local swimming-pool, since there isn’t one; that, however Jones spent the morning, he did not spend it at the local swimming-pool, since there is no such place; and similarly (ii) that it is quite untrue, or is false, that the Exhibition was visited yesterday by the King of France; that, whoever the Exhibition was visited by yesterday, it was not visited by the King of France, since there is no such person. (Strawson 1964: 112)

To concentrate, consider the following pair: (7) (8)

#The king of France is bald. F Last week, my friend went for a drive with the king of France.

People who are aware that there is no king of France quite reliably and conWdently judge the second sentence to be false, while they display a certain amount of ‘squeamishness’ about the Wrst. I wish to understand this diVerence. From now on, I will annotate examples to indicate the relevant judgment: # means ‘squeamishness’, F means ‘conWdent judgment of falsity’.11 Strawson’s new examples should make us reconsider the view I arrived at about the status of truth-value intuitions. Once one appreciates the robustness of the contrast, the suspicion arises that perhaps I was too quick to dismiss truth-value intuitions as unreliable and useless for the theory of presuppositions. Based on the clear distinction in intuitions, Fodor (1979) in fact argued that the contrast between the new examples and the original cases (The king of France is bald) is a point in favor of a truth-value gap analysis for the latter. Perhaps, this pattern of judgments should be taken as showing that (7) indeed does not have a truth-value while (8) has the truth-value false. One 11 I have found that most of my informants consistently felt there to be a clear diVerence between these sentences, and that once they Wxed on the diVerence they could easily assign other sentences to one of the two cases. There are also speakers that have simpler patterns of judgment. I have encountered people who consistently report Russellian judgments of falsity, even for sentences like (7). There may also people who have ‘squeamish’ judgments throughout, even for sentences like (8). Both these populations are easy to ‘explain’ within my favorite theory (and probably in any theory of presupposition). The Russellians simply map any untruth (false or undeWned) to FALSE, while the squeamish ones have direct intuitive access to the Strawsonian semantics of deWnite descriptions. The real challenge comes from the majority of speakers who make the diVerence between (7) and (8) as described in the text. For these, we need an analysis of the kind I plan to develop in what follows.

322

III. Presupposition and Truth-Value Gaps

might further think that the reason for this diVerence in truth-value status is that, while (7) presupposes that there is a king of France, (8) does not carry that same presupposition. In this chapter, I will argue against this picture. I will argue that even (8) presupposes that there is a king of France and that (at least from the perspective of the theory of presuppositions that I favor) it may well be the case that this is because it has no truthvalue in the technical sense. Truth-value intuitions continue to be untrustworthy. This, however, is not the consensus view in the Weld, to put it mildly.

4. Previous Analyses of Strawson’s Contrast Paradise Found Many people have reacted to Strawson’s contrast with an extremely seductive analysis. The Wrst tenet of this analysis is that truth-value gap judgments do after all correlate neatly with absence/presence of a presupposition. In the F-sentences there is no presupposition that there is a king of France. In the #-sentences there is such a presupposition. The second tenet of the analysis is that the diVerence between the two kinds of examples lies in what the sentences are about, their topic/focus articulation. Basically, the story goes that all and only topical deWnites carry presuppositions. There are two possible ways to go from here. The more straightforward system attributes a semantic presupposition to all deWnite descriptions, but then introduces a mechanism that renders this presupposition harmless whenever the deWnite is nontopical. Only presuppositions in the topic survive this procedure. This may have been roughly what Strawson had in mind when he talked about a referring expression being ‘absorbed’ into the predicate. In most work inspired by Strawson’s 1964 paper, a diVerent line is pursued. DeWnites initially have a non-presuppositional, Russellian semantics; presuppositions are induced if and only if the deWnite is (in) the topic. This idea about the link between topichood and existential presupposition continues to be a very popular assumption and can for example be found in: Reinhart (1981, 1995); Hajicova (1984); Gundel (1977); Horn (1986); Lambrecht (1994); Erteschik-Shir (1997); Zubizaretta (1998). Concrete and explicit formalizations of this idea are hard to Wnd, but see work by Cresti (1995) and Percus (1996, 1998). Let me be blunt: these analyses are fundamentally mistaken.

Paradise Lost The most important point I want to get across in this chapter is that there is no neat correlation between truth-value gap judgments and presence/absence of presuppos-

von Fintel, Would You Believe It? 323 itions. If we take the new Strawson examples that are judged false without hesitation and submit them to the usual tests for the presence of presuppositions, we Wnd that they clearly carry existence presuppositions just like the old examples.12 A speaker who says I had breakfast with the king of France this morning is not asserting that there is a king of France. Rather, the existence of the king of France is still taken for granted. Here’s the ‘Hey, wait a minute’ test for a couple of representative examples: (9)

(10)

A: B: B 0: A: B: B0:

F

The king of France attended the APEC conference this week. Hey, wait a minute—I had no idea that France is still a monarchy. #Hey, wait a minute—I had no idea that he was at that conference. F This year’s Fields Medal was awarded to the mathematician who proved Goldbach’s Conjecture. Hey, wait a minute—I had no idea that someone proved the Conjecture. #Hey, wait a minute—I had no idea that she got the medal.

By this test, the deWnites in these sentences still carry existence presuppositions. This diagnosis is also conWrmed by some projection data: (11) (12)

(13) (14)

I hope that the king of France attended the APEC conference this week. If the king of France attends the APEC conference this week, there is a chance the proposed treaty will take eVect soon. cf. If France has a king and he attends the APEC conference this week, there is a chance the proposed treaty will take eVect soon. I hope that this year’s Fields Medal was awarded to the mathematician who proved Goldbach’s Conjecture. If this year’s Fields Medal is awarded to the mathematician who proved Goldbach’s Conjecture, my friend James (who has hopes on it himself ) will be quite disappointed.

Clearly, all these examples show the presupposition triggered by the deWnite description project out of the embedding constructions. Therefore, these examples have the usual presuppositions, although the local sentences in which the deWnite appears are judged false when asserted on their own. Presuppositionality and truth-value judgments are still not linked as tightly as one might have hoped.13 12 I am disagreeing with Strawson here. He writes about the example My friend went for a drive with the king of France that ‘the existence of a King of France is not, in this example, a presupposition of the whole discussion, as is the existence of the friend whose exploits I am recounting’ (Strawson 1954). Similarly, Horn (1989: 488) writes of his example The king of France is sitting in the chair to your right: ‘if I note that the designated chair is clearly empty, I am far less inclined to grant that you have presupposed the existence of the King of France—and even less likely to grant that what you have said is neither true nor false if, in fact, France is a republic’. By the way, Horn notes that for his example ‘[w]hat is at issue is not the criterion of aboutness, but a distinction in veriWcation procedures’. This remark takes us to the main line of thought to be pursued in this chapter. 13 The only work on presuppositions that also stresses this point is Burton-Roberts (1989: esp. s. 2.2 and ch 9). However, since I do not understand his treatment of presupposition, I will not discuss it.

324

III. Presupposition and Truth-Value Gaps

This is the central negative lesson of this chapter: do not take truth-value gap judgments as reliable tests for presuppositions. The remainder of the chapter is devoted to the search for an empirically adequate analysis of the circumstances under which sentences with failed presuppositions are judged plainly false rather than giving rise to squeamishness.

Not about Topic–Focus Even if one is persuaded that the pattern of truth-value intuitions is not reXective of the presence or absence of presuppositions, one might still be attracted to the other plank of the majority view. Perhaps, it is indeed the topic/focus articulation of a sentence that determines whether a problematic deWnite will give rise to truth-value gap judgments. Perhaps, a deWnite that carries a false presupposition will indeed give rise to a truthvalue gap judgment if and only if it is a topical deWnite. But here as well, I have to raise some doubts. It just does not appear to be true that all topical/non-focal deWnites lead to truth-value gaps: (15)

Let me tell you about my friend, the king ofFrance. the king of France F I had breakfast with this morning. him The second sentence in (15) is on a par with the other new examples, it is plainly false. Nevertheless, it would appear to be about the king of France, since it occurs in a discourse that is meant to be about the king. Or imagine overhearing a dialogue between two people laboring under the misapprehension that there is a king of France: (16)

A: B:

Have you heard anything about the king of France recently? I think he may be getting old and decrepit. Well. F Bill Clinton had breakfast with him last week and he looked just Wne I hear.

We would of course think that these two belong somewhere nice and cozy where they can’t hurt anyone. But still they are talking about the king of France and what B says feels blatantly false. Another example: (17)

F

I had breakfast with the king of France this morning. F He and I both had scrambled eggs.

The second sentence seems clearly false, even though after the Wrst sentence (which is certainly false as well) the king of France will have become topical. There is then some empirical reason to doubt the validity of the generalization that it is topical deWnites that lead to truth-value gap judgments while it is deWnites absorbed into the predicate that contribute to simple judgments of falsity. One may of

von Fintel, Would You Believe It? 325 course still safely hold that it is usually not a good thing to talk about a nonexistent entity as if it existed. So, we need to explore further what the precise circumstances are under which presuppositionally loaded sentences are judged false and precisely when they give rise to squeamishness. This is not an easy task. Donnellan (1981) wrote a whole paper that, apart from introducing a few further examples to worry about, is a meditation on the Wckleness of linguistic intuitions. He writes: I Wnd no easy way to organize the explanation for our intuitions about the sorts of examples philosophers have been primarily concerned with—reference failure. And I have not tried to deal with the much broader application of the notion of presuppositions that linguists have made. Nothing I have said supports or denies that assertions have presuppositions or that, about the sorts of examples philosophers have been interested in, Russell was or was not correct as against Strawson. I do suggest that intuitions are not simple things to account for. (1981: 142)

Indeed.

5. The Mechanics of Rejection I will develop an alternative view of Strawson’s contrast. The basic intuition is that the judgments we have about the relevant sentences are not directly reXective of their semantic status. Rather, once we are faced with presupposition failure with its concomitant truth-value gap, there are fall-back strategies to Wll in the gap. I will speculate at the end why there should be such strategies. Here is the idea of what the relevant strategy is. The later Strawson examples are rejected as false without squeamishness because they misdescribe the world in two ways: their presupposition is false, but in addition there is another untruth, which is independent of the failure of their presupposition. I imagine that someone who rejects them as false reasons that, even if their presupposition is true (which it isn’t), they will still be false because of that other false claim. I start by discussing a previous incarnation of the analysis I will explore.

Lasersohn’s Idea Lasersohn (1993) argues that what is going on is that, whenever presupposition failure looms, we test whether we can assign a truth-value independently of our knowledge about the nonexistence of the referent in question. He writes: [A]n aYrmative statement which might otherwise be judged of indeterminate truth-value (because it contains a term which fails to refer) can instead be judged false, provided the context makes it possible to determine that the statement could not possibly be true regardless of whether the term has reference or not. . . . (1993: 115)

326

III. Presupposition and Truth-Value Gaps

Why is it that someone who points at an empty chair and says The King of France is sitting in that chair seems to be saying something false? I would like to suggest that it is because even if we suspend our knowledge that there is no King of France, there is no way of consistently extending our information to include the proposition that the King of France is sitting in that chair. Such an extension is impossible because we know the chair to be empty. In contrast, if we suspend our knowledge that there is no King of France, our information may then be extended either to include the proposition that the King of France is bald, or to include the proposition that the King of France is not bald. (1993: 116)

The idea then is this, The king of France is sitting in that chair is false for us, since the following is true for us: Suppose I didn’t know that the king of France didn’t exist. Then, if I learnt that the king of France did exist, I would still know that he is not sitting in that chair (because that chair is empty).

Rejection Relative to a Body of Information Lasersohn framed his analysis in terms of ‘data semantics’, based on work by Veltman (1983) and Landman (1986). Instead of explicating his particular formulation, I will work with a perhaps simpler framework. First, I will continue to assume a Frege– Strawson semantics for deWnites, as laid out at the beginning of this chapter. With this decision I am of course at an initial loss to account for Strawson’s contrast. All the sentences referring to the present king of France should equally fail to assign a truthvalue to our world. To make the system Xexible enough, I add to this semantics a pragmatics of assent and rejection. I will in eVect assume two kinds of ‘truth-values’: (i) the semantically assigned values 1 and 0 which are the values a partial proposition can give for a particular evaluation world, and (ii) the pragmatically assigned values TRUE and FALSE which reXect the status of the sentence with respect to a given body of information. The claim I am developing is that naive judgments of truth-values give intuitions about pragmatic TRUTH and FALSITY and are not directly reXecting the semantic status of a sentence. Sentences are assessed with respect to a given body of information D, which I will model as a consistent set of propositions. A sentence is accepted as TRUE iV it is entailed by the given body of information D, or equivalently, iV it has truth-value 1 in all the worlds compatible with D. A simple rule for rejection would say that a sentence is rejected as FALSE iV its negation is entailed by the given body of information, or iV it has truth-value 0 in all the worlds compatible with D. (18)

(19)

Acceptance Accept a sentence f as TRUE with respect to a body of information D iV for all worlds w compatible with D: sft(w) ¼ 1. Rejection ( Wrst attempt) Reject a sentence f as FALSE with respect to a body of information D iV for all worlds w compatible with D: sft(w) ¼ 0.

von Fintel, Would You Believe It? 327 Note quickly that there is a lot of space in between TRUE and FALSE. If there are both worlds in D where f is true and worlds where it is false, then f will be judged neither TRUE nor FALSE. Further, as soon as f is undeWned for any world in D, it will also be neither TRUE nor FALSE. So far, I have not gained anything yet for my problem with sentences with nonreferring deWnites. Rejection must be more complicated. Lasersohn’s idea was that the given body of information which does not admit evaluation of the sentence with its failed presupposition is adjusted or revised to admit the sentence. Let me postulate an operation that does this: (20)

Revision For any body of information D and any proposition p, revp (D), the revision of D so as to entail p, is a body of information that is as much like D as possible but that entails p. If D already entails p, revp (D) ¼ D.

My procedure for pragmatic rejection is then: (21)

Rejection (revised) Reject a sentence f (with presupposition p) as FALSE with respect to a body of information D iV for all worlds w compatible with revp (D): sft(w) ¼ 0.

Before I talk more about the speciWcs of the revision procedure employed here, let me quickly do a couple of examples. (22)

#The king of France is bald.

The sentence will be accepted as TRUE iV in all of the worlds compatible with the given body of information there is a king of France, who is bald. The sentence will be rejected as FALSE if in all of the worlds compatible with the given body of information the king of France has hair. The sentence will also be rejected as FALSE under the following conditions: (i) the given body of information does not entail that there is a king of France, (ii) but once we revise the body of information enough to include the existence of a king of France, the revised body of information entails that the king of France has hair. (23)

F

The king of France is sitting in that chair.

The sentence will be accepted as TRUE iV in all of the worlds compatible with the given body of information there is a king of France, who is sitting in that chair. The sentence will be rejected as FALSE if in all of the worlds compatible with the given body of information the king of France is somewhere other than in that chair. The sentence will also be rejected as FALSE under the following conditions: (i) the given body of information does not entail that there is a king of France, (ii) but once we revise the body of information enough to include the existence of a king of France, the

328

III. Presupposition and Truth-Value Gaps

revised body of information entails that the king of France is somewhere other than in that chair. The intuition about the contrast between the two sentences is that it follows from a diVerence in what kind of information is retained during the revision. Once we revise our information to include the existence of a king of France, we have no way of determining positively whether he has hair or not. Nothing in our prior state makes that determination and the mere addition of a king of France to the furniture of the world does not necessitate anything about his hair. In contrast, even if we revise our information to include the existence of a king of France, this will not get rid of the piece of information that the chair pointed at is blatantly empty. The substance of the analysis then lies in specifying how exactly the revision procedure works. Lasersohn’s account is actually only the shell of an analysis. As Lasersohn (1993: 121 n. 4) admits, ‘a rigorous deWnition of [revision] is not a trivial enterprise’. But from his informal remarks and from a look at contrasts such as the one between (22) and (23), we can draw some preliminary conclusions about the intended analysis.

Epistemic Revision, Not Counterfactual Reasoning One immediate decision I have to make about the nature of revision in pragmatic rejection is whether it is counterfactual or epistemic. This is an easy decision. Clearly, the reasoning cannot actually be running on ordinary counterfactual lines, because it is not too hard to imagine a scenario where, if there were a king of France, he would indeed be sitting in that chair. Nevertheless, even if I had such counterfactual beliefs, the sentence would still be judged FALSE if the chair is obviously in fact empty. So, the reasoning is running on epistemic lines: the piece of information that there is no one sitting in that chair or that the person sitting in that chair is not king of France is so entrenched that it would not be overturned upon my learning that there is a king of France. The procedure I need is then something like this: to revise D so as to entail p, remove from D anything that is incompatible with p, add p to D, and furthermore discard any information in D that was believed just because p was believed to be false. (24)

Common-sense epistemic revision Remove :p from D. Remove any proposition from D that is incompatible with p. Remove any proposition from D that was in D just because :p was in D. Add p to D. Close under logical consequence.

The resulting picture is that what is involved in pragmatic rejection is roughly the same procedure that would apply if someone who believed that p is false was reliably informed that p is in fact true. That person would then start believing p, stop believing

von Fintel, Would You Believe It? 329 anything logically incompatible with p, and stop believing (presumably becoming agnostic about) anything previously believed just because p was believed to be false.14 Another way to see the picture we get from Lasersohn’s paper is to correlate the judgments about the contrast between the king’s being bald and the king’s sitting in that chair with a contrast between two corresponding even if-conditionals: (25) (26)

Even if there is a king of France (which there isn’t), he is still not bald. Even if there is a king of France (which there isn’t), he is still not sitting in that chair/that chair is still empty.

I would not assent to (25) but I would assent to (26). What is happening in these even if-conditionals seems to be similar to how Lasersohn conceives of the procedure for rejecting sentences with failed presuppositions. The speaker does not really give much credence to the possibility that the antecedent is true, but if it is true and the information state has to be revised accordingly, the consequent will still be true—belief in its truth will be unaVected by the revision. (25) and (26) diVer in exactly the expected way.

Why Not Simpler? Lasersohn’s proposal seems intuitively appealing, since belief revision of the kind sketched in (24) is independently needed. But one might wonder whether something even simpler might not work. The simplest idea, perhaps, would be to keep all information that is logically consistent with f’s presupposition: (27)

Minimal logical revision or Very grudging revision Remove :p from D. Remove any proposition from D that is logically incompatible with p. Add p to D. Close under logical consequence.

There is another way of putting this analysis: all that is needed for a sentence with a failed presupposition to be judged FALSE (rather than evoking the squeamish truthvalue gap judgment) is that it have an entailment that is false but that itself does not entail the falsity of the sentence’s presuppositions. It is permissible that this false entailment itself is only known to be false because its falsity is entailed by the known falsity of the presupposition. But this goes too far, which was already seen by J. D. Fodor (1979), who starts by considering and then rejecting an account very much like the one just oVered. Consider: (28)

#The king of France is bald.

14 At least that is my amateur’s hunch about belief revision. It would be interesting to compare what we are doing here to established results in the literature on belief revision.

330

III. Presupposition and Truth-Value Gaps (29)

F

The king of France is standing next to me.

J. D. Fodor (1979: 201–3) writes: The idea here is that the existence or non-existence of a King of France is irrelevant to the valuation of (29). We need not know whether there is a King of France in order to determine that (29) is false; all we need establish is that there is no King of France standing next to me; that is, either that there is no one standing next to me or that the person or persons standing next to me are not King of France. . . . To put it informally, and in overtly veriWcationist terms, this theory says that (28) has no truth-value because to evaluate it we would need to look at the King of France and we cannot; and that (29) does have a truth-value because to evaluate it we need only look at me and at those people who are near me. But this is unsatisfactory, for it simply shifts the puzzle one step further back. If we can make a list of the people standing next to me and determine that no King of France is on it, why could we not also make a list of the people who are bald and establish that no King of France is on IT?

So, the problem with the simpler mechanism in (27), apart from it being diVerent from realistic belief revision, is that now essentially all relevant sentences are predicted to be judged FALSE, because in principle they could be shown to be misdescribing the world without engaging the presupposition.

The Dilemma As sketchy as Lasersohn’s analysis is, it is demonstrably wrong. Consider: (30)

F

The king of France is on a state visit to Australia this week.

(30) is judged FALSE, which is quite unexpected under Lasersohn’s account. The fact is that most of us judge (30) to be FALSE while our only ground for doing so is our belief that there is no king of France. Imagine, for example, that like me you have no reliable information about the daily political life and current events in Australia. There is no entrenched information, independent of your disbelief in the king of France’s existence, that would ensure that the king of France is not on a state visit to Australia. So Lasersohn would expect the example to be on a par with The king of France is bald, contrary to intuition. There are a lot more examples where (30) came from: (31)

(Coming across an abandoned umbrella on a park bench) F This umbrella was left here by the king of France.15

It is merely our knowledge that there is no king of France which supports our conviction that the umbrella was not left by the king of France. Damagingly to Lasersohn’s 15 Smart alecks might complain that the umbrella could conceivably have been left by the last king of France and just has been lying on that bench for a (very) long time. Such nitpicking can be defused by considering a modernized variant like This cell phone was left here by the king of France.

von Fintel, Would You Believe It? 331 account, while (31) is judged FALSE, one would presumably not assent without hesitation to the corresponding even if-statement: (32)

Even if there is a king of France (which there isn’t), this umbrella was left by someone else.

We are left with quite a dilemma. The version of Lasersohn’s analysis which assumes that the rejection strategy uses common-sense belief revision makes the wrong prediction for (30) and (31). On the other hand, we cannot loosen the analysis too much, say by using the simple idea in (27), since then even The king of France is bald should be judged FALSE.

Footholds for Rejection The question is how to distinguish between these two examples: (33) (34)

#The king of France is wise. The king of France is on a state visit to Australia this week.

F

In each case, we have no reason to reject the sentences other than our knowledge that there is no king of France. We have no independent counter-evidence, evidence that we could hold onto while admitting—albeit not seriously—that there may be a king of France after all. In both cases, there is possible counter-evidence, independent in principle from the existence of the king of France. We could list all wise people and not Wnd the king of France on the list. We could explore what is going on in Australia right now and not Wnd any state visit in progress by a king of France. Based on our knowledge that there is no king of France, we are convinced that these pieces of evidence could be obtained. I take it that it will not help to claim that one could not possibly list all wise people. Here is what might do the trick. In the case of (34) but not in the case of (33), there is a contextually salient entity whose properties (known or not known) are in principle enough to falsify the sentence. In (33), there is no contextually salient entity mentioned (other than the king of France) whose properties could establish that (33) is false. In (34), Australia is made salient and can thus furnish an independent foothold for falsiWcation. The idea then is that the rejection strategy can be based on facts that we know must be there (since we know that the presupposition of the sentence is false) and that we know could be established by examining an entity that everyone involved agrees exists. So, in eVect we are using the following revision procedure: (35)

Conversational revision Remove :p from D. Remove any proposition from D that is incompatible with p.

332

III. Presupposition and Truth-Value Gaps Remove any proposition from D that was in D just because :p was in D, unless it could be shown to be true by examining the intrinsic properties of a contextually salient entity. Add p to D. Close under logical consequence.

We would of course need to explore what counts as a contextually salient entity of the right kind and what exactly it means to say that the intrinsic properties of that entity are enough to falsify the sentence. To a Wrst approximation, contextually salient entities will be those mentioned in the relevant sentence (or perhaps in prior discourse). Perhaps, one could use the discourse referents of File Change Semantics or Discourse Representation Theory. I will not have much to say about this. Crucially, a sentence like The king of France is bald does not make salient or even mention the set of bald people in the world. The predicate bald is not a referring expression. So the account we are now considering correctly predicts that the classic example is not one that can be judged FALSE in the face of presupposition failure. But compare it with the following: (36)

F F

Among the bald people in this world is the king of France. The king of France is one of the bald people in this world.

These sentences do contain expressions referring to the set of bald people in this world. Since we know that the king of France is not among them, we judge these sentences to be FALSE. The contextually salient entity may not have to be mentioned in the same (or any) sentence. David Pesetsky (pers. comm.) reports that The king of France is bald can be judged false if made in the presence of a list that enumerates all the reigning monarchs of the world together with their hairstyle. The contextually salient entity that underwrites our judgments of FALSITY may not have to be an entity of the normal kind. Examples that make claims about particular episodes or situations seem to allow FALSITY judgments: (37)

F F

The king of France is jogging right now. The mathematician who proved Goldbach’s Conjecture died yesterday.

One may think that stage-level predicates, those that make episodic claims such as those in (37), will thus always give rise to FALSITY judgments.16 But there are cases where it seems that the episodes are too diVuse to count as independent footholds for falsiWcation: (38)

#The king of France is having rotten luck these days. #Rotten luck has lately befallen the king of France.

16 The distinction between ‘stage-level’ and ‘individual-level’ predicates is due to Carlson (1977a, b), see also Kratzer (1995) and Diesing (1992).

von Fintel, Would You Believe It? 333 Since individual-level predicates make claims that are in general not localized, one would think that they give rise to more #-judgments than stage-level predicates. Lasersohn in fact considers and rejects the idea that all and only subjects of individual-level predicates give rise to truth-value gaps. His counter-example is: (39)

F

The king of France is on the Rochester faculty.

The argument is that be on the Rochester faculty is an individual-level predicate, but that the sentence (39) is naturally judged false. However, as pointed out to me both by Orin Percus and Irene Heim, be on the Rochester faculty is perhaps not a true individual-level predicate. It can for example be felicitously used as a predicate in there-insertion contexts: (40)

There are many geniuses on the Rochester faculty.

Let me give an example that does make Lasersohn’s point: (41)

F

The king of France owns this pen.

Clearly FALSE. Nevertheless, the deWnite is the subject of an individual-level property. So, not all subjects of individual-level predicates lead to truth-value gaps. Let me expand (41) into a full paradigm: (42) (43)

#The king of France owns a pen. F The king of France owns this pen.

The predicate here is an individual-level predicate in both cases. The diVerence lies in whether there is a speciWc object that the king of France is claimed to own. The intuition is that there is a clear contrast: (41) is much easier to judge FALSE than (42). Consider also the following in contrast to (42): (44)

a.

F

Among the pens in the world, there is one owned by the king of France. b. F The king of France owns at least one among the pens in the world.

These seem again easier to judge FALSE than (42), but surely the semantics is the same. The relevant diVerence seems to be that (44) presents the set of all pens as an entity that we can talk about, while (42) does not.

A Side Remark: Independent VeriWcation Another way to see the intuitive force of the new generalization is this. Imagine you want to convince a doubter of the existence of the king of France. Then, you will want to provide a piece of evidence that can be veriWed and that then (and only then) will support the notion that there is a king of France. Imagine using the following sentences in the scenario given.

334

III. Presupposition and Truth-Value Gaps (45)

Look, of course there is a King of France! a. . . . I had breakfast with him this morning. . . . He’s married to my daughter. . . . He likes my daughter. . . . He owns this pen. b. # . . . He’s bald. # . . . He is married to a Dutchwoman. # . . . He likes his daughter. # . . . He owns a pen.

The sentences that are useless in this kind of argument are exactly those that do not give any foothold for veriWcation independently of the king of France himself.

Salient Footholds and Aboutness The proposal I am developing might seem overly complicated. Why can we not just say that the diVerence between The king of France is in Australia and The king of France is bald is that the former mentions an entity other than the king while the latter does not? That would be much simpler. Well, according to the intuition we are pursuing, one needs more than just some referential expression in the sentence denoting a contextually salient entity: it has to be the case that one could falsify the sentence by looking at that entity. The sentence must say something false about that entity. Here are a couple of nice contrasts that show the force of this requirement: (46)

F

The king of France died in the car accident on the turnpike last night. The king of France was involved in the car accident on the turnpike last night. #?The king of France heard about the car accident on the turnpike last night. F The king of France is in town. #?The king of France is out of town. F

(47)

Just by looking at information about the car accident on the turnpike last night, we would presumably be able to obtain a list of people involved in the accident and also whether they died or not. But we would not have any way of determining who heard about it. Similarly, by looking at the town in question (Paris, I would think) we can Wgure out whether the king is in it. But we cannot Wgure out by looking at the town whether the king is out of it or whether he does not exist. So, the mere presence of a referential item does not necessarily prevent #-judgments. The sentence has to be make an independently falsiWable claim about the entity referred to. By the way, the analyses I mentioned that employ the notion of topic also use the term ‘about’. But it is important to recognize the diVerence between the two ap-

von Fintel, Would You Believe It? 335 proaches. Discourse-based analyses consider whether an exchange is centered around giving information about an issue or about an entity. A sentence is about an entity in this discourse sense iV it answers a question that was asked about that entity. My account asks whether the sentence could be veriWed/falsiWed by looking at a particular entity and its intrinsic properties. If so, it is about that entity. A sentence like The king of France is in Australia, in that light, is both about Australia and about the king of France.17

How Independent does the Counterevidence have to be? Unfortunately, my new account is still not quite correct. Consider in fact my primary example of a sentence that is not rejected as FALSE but gives rise to squeamishness. Add to this just one more example that will illustrate the problem: (48) (49)

#The king of France is bald. #?The man who Sandy went out with last night is bald.

Notice with dismay that both of these examples in fact contain terms referring to salient entities: (48) refers to France, (49) refers to Sandy. Furthermore, there will be in our body of information propositions about those entities which, while compatible with the existence of the king of France, are incompatible with the truth of (48) and (49). We believe that France does not have a bald king. This belief is logically consistent with the existence of the king of France and it is about a salient entity: France. Since we believe that Sandy did not go out with any man last night, we also believe that Sandy did not go out with a bald man last night. The latter belief is logically consistent with Sandy going out with a man last night and it is about a salient entity: Sandy. Why are these sentences not rejected as FALSE? We cannot retreat to the position that the pieces of counter-evidence considered here are ineligible because they were believed to be true just because the presupposition of the oVending sentence was believed to be false. We cannot go that way because of the problem with the state visit to Australia and the umbrella on the park bench in Paris. Consider the contrast again: (50) (51)

#The king of France is bald. Counter-evidence: France does not have a bald king. F The king of France is on a state visit to Australia this week. Counter-evidence: Australia is not being visited by a king of France this week.

17 The sense in which The king of France is on a state visit to Australia is (partly) about Australia involves the concept of aboutness that David Lewis has explored (Lewis 1988), see also Goodman (1961). For some relevant linguistic discussion see Portner and Yabushita (1998).

336

III. Presupposition and Truth-Value Gaps

Both pieces of potential counter-evidence are about salient entities and are believed to be true just because we believe there to be no king of France. What distinguishes them? Here is an idea: the counter-evidence in (51) is in principle epistemically independent of the oVending presupposition. While we believe it to be true just because we believe the presupposition to be false, we could conceivably show it to be true while not showing the presupposition to be false. We travel to Australia and see what is going on. In contrast, the potential counter-evidence in (50) is not epistemically independent of the non-existence of the king of France, even in principle. As soon as we show that France does not have a bald king, we will also have shown that France does not have a king at all. The same diagnosis would seem to apply to Sandy in (49). It is plausible that we could not establish that Sandy did not go out with a bald man without also establishing that she did not go out with a man at all. We are left with a description of what’s going on that looks like this: (52)

Conversational revision (Wnal version) Remove :p from D. Remove any proposition from D that is incompatible with p. Remove any proposition from D that was in D just because :p was in D, unless it could be shown to be true by examining the intrinsic properties of a contextually salient entity without at the same time showing that p is false. Add p to D. Close under logical consequence.

Another Wrinkle: Laws I have concentrated on sentences about the king of France that either do or do not make false claims about some other entity. But the account I formulated, and indeed its ancestor from Lasersohn’s paper, makes a further prediction. If the claim of the sentence about the King of France is incompatible with some general fact (which is so general that it certainly is not believed just because we believe there is no king of France), the sentence should be rejected as FALSE. Relevant examples include: (53)

a. ??The king of France is a woman. b. ??The king of France owns a unicorn. c. ??The king of France owns lakefront property in Bavaria. (Irene Heim, pers. comm.)

While I am not so sure about judgments here, there do seem to be many speakers who judge these to be plainly false. The relevant pieces of counter-evidence here are: kings

von Fintel, Would You Believe It? 337 are male, there are no unicorns, and there is no private ownership of lakefront property in Bavaria.

Strawson’s Curveball Strawson has an example that might threaten our account: (54)

Q: A:

What reigning monarchs are bald? The king of France is bald.

Strawson reports the intuition that The king of France is bald as an answer to the question in (54) is felt to be false. If that is so, it would support his topic-based approach (which it was of course meant to) and remove support from my approach. The idea is that, in the context of the question What reigning monarchs are bald?, our sentence is not about the king of France (in a discourse sense) but rather about baldness. As soon as the oVending deWnite is not what the sentence is about, we should reject the sentence as FALSE. My account has no provision tying the evaluation of the sentence to the question it answers. So it would seem that (54) is predicted to receive a #-judgment. One could quibble with the empirical claim here. Perhaps (54) is still naturally given a #-judgment even in the context of Strawson’s question. But let us assume that the sentence can be rejected as FALSE. The only escape route for my account is to say that there must be a salient entity here about which the sentence makes a false claim. I would suggest that the question in (54) in fact introduces such an entity: the set of reigning monarchs. The answer then makes a false claim about that set: that it contains a bald king of France.

Reformulation?18 Abusch and Rooth (forthcoming) brieXy report on Lasersohn’s analysis and recast his procedure in the following way (actually, this is my recasting of their recasting): (55)

Reject a sentence f (with presupposition p) as FALSE with respect to a body of information D iV there is a D 0 D such that for all D 00 D 0 such that D 00 entails p: for all worlds w compatible with D 00 : sft(w) ¼ 0.

We can reject f as false if our body of information D contains a more limited body of information D 0 which is compatible with the presuppositions of f and that has the property that no matter how it is expanded into a body of information D 00 which entails the presuppositions of f the resulting D 00 will falsify f. 18 This section and the next one brieXy bring up some issues from a new paper by Dorit Abusch and Mats Rooth (forthcoming). I may have more to say about their analyses on some other occasion.

338

III. Presupposition and Truth-Value Gaps

This does seem to recapture Lasersohn’s sketch of an analysis quite faithfully. Unless it is further modiWed in the same ways as I have discussed, it will make the same incorrect predictions. For example, my body of information has a more limited subset which entails that France has no bald king, which can be expanded into a bigger body according to which France does have a king, but no matter how the expansion is done, the resulting set will falsify The king of France is bald. It is possible that my modiWcations will seem more natural when using this reformulation of Lasersohn’s analysis. I leave that to my readers to explore.

Other QuantiWers Abusch and Rooth (forthcoming) are also concerned with explaining patterns of truthvalue judgments, but they mostly deal with the universal quantiWer every and with socalled weak quantiWers like no and two. They argue that in these cases judgments of oddness may trace back to conversational implicatures rather than to presupposition failure. They do not discuss deWnites but for clearly presuppositional quantiWers like both they look favorably on a Lasersohn-style analysis. In fact, they claim that every diVers from both. They consider a scenario where someone reports on a reception and says one of the following sentences. (56) (57)

Every surviving Civil War vet was at the reception. Both surviving Civil War vets were at the reception.

They say that a Lasersohn rejection mechanism applies to (57) and that we can easily imagine this dialogue: (58)

A: Both Civil War vets who live in the city were at the reception. B: False, only Gulf War vets were there.

On the other hand, they report that, according to their intuitions, (56) cannot be rejected as false. If any truth-value judgment is possible it is the smart-alecky (59)

True, but just because there are no surviving Civil War vets.

They argue that every has a non-presuppositional subset meaning, just as many logic textbooks teach us. The oddness of (56) is not due to presupposition failure but because via conversational implicature one infers that the speaker must believe it to be possible that there are surviving Civil War vets, which to us is bizarre, hence the feeling of oddness. I think there is a little bit more to be said here. Let us look at the following scenario. We know that there are no incoming students from Australia this year. But the speaker who asserts the sentences we will contemplate apparently does not share that knowledge and in fact might be mistakenly thinking that there are incoming students from Australia. How would you judge the following sentences? (I have given the usual F/#

von Fintel, Would You Believe It? 339 notation—based on my cursory Weld research—with the addition of some T judgments). (60)

a. b. c.

(61)

a. b. c.

(62)

a. b. c.

T

No incoming student from Australia was at the meeting last night. Every incoming student from Australia was at the meeting last night. #Both incoming students from Australia were at the meeting last night. T This week, I have had breakfast with no incoming student from Australia. T=F This week, I have had breakfast with every incoming student from Australia. #This week, I have had breakfast with both incoming students from Australia. #This year, no incoming student from Australia is bald. #This year, every incoming student from Australia is bald. #This year, both incoming students from Australia are bald. T=F

Some observations about what I found: Every Most of them being sophisticated linguists, my informants clearly saw the possibility of calling (60b) and (61b) true as the non-presuppositional subset meaning would predict. But there was also the sense that this is almost an ambiguity, a hypothesis that Abusch and Rooth (forthcoming) actually consider as plausible. Perhaps, every can be understood in the smart aleck way, but it also has a more prosaic meaning where it is a presuppositional universal quantiWer. Under that reading and assuming the theory of rejection developed here, the sentences in (60b) and (61b) should be judged false. Indeed, one of my informants wrote about (61b): ‘I feel as if it’s deWnitely true or deWnitely false but I’m not sure which.’ The most interesting facet of the judgments with every is that judgments of smart aleck truth or plain falsity are much more readily available with the entity-anchored statements in (60b) and (61b). With predicates like bald as in (62b), informants found every just as weird as other determiners. Both Contrary to Abusch and Rooth’s diagnosis of what is going on in their example (57), informants did not Wnd the sentences with both in (60c) and (61c) false but assigned them to the squeamish type. This fact is of course unexpected from the point of view of my analysis. The feeling one might have is that it is much harder to ignore the failure of the presupposition that there are exactly two incoming students from Australia than the failure of the much less speciWc presupposition that there are such students in the

340

III. Presupposition and Truth-Value Gaps

Wrst place. But making such a distinction in speciWcity between two presupposition triggers is not easily done within the conWnes of my speciWc proposal. No This determiner behaves pretty much as one expects based on the extensive literature on weak quantiWers in recent years (see e.g. Diesing 1992). Together with a stage-level predicate as in (60a) and (61a), no makes a simple set-theoretic claim (that there is an empty intersection between its domain and the matrix predicate) with no presupposition. But together with an individual-level predicate like bald in (62a), no triggers a presupposition that its domain is nonempty. In the absence of a salient foothold for rejection, the sentence arouses feelings of squeamishness. I could and should go on and on, continuing to explore how various presupposition triggers behave when we probe them with the Strawsonian battery of examples. There is bound to be much to be discovered here. But let me be moderate for now and move this chapter towards its end.

6. What are Truth-Value Judgments Really About? My empirical generalization may look truly bizarre. I think that the way to make sense of it is to rethink what intuitions of truth-value judgments are about. As I said early on, truth-value judgments about a sentence are not directly reXective of the semantic value of that sentence. It appears that we prefer to give judgments of falsity even when strictly speaking the sentence has no truth-value. Horror vacui? Perhaps this is a matter of internalized charity toward speakers: making false presuppositions is a graver oVense than asserting something false, so as much as we can we gloss over such a faux pas. But still, if it were just a matter of mapping any untruth to FALSE judgments, I would not expect the pattern I found. If truth-value judgments were primarily about the match between the sentence and the world, then all of the relevant sentences should be judged false because they misrepresent the world in some way or other. Perhaps, the best explanation for the pattern of judgments discussed here can be achieved by assuming that we are dealing with intuitions about possible conversational moves. If a sentence is oVered whose presupposition we know to be unsatisWed, we have to reject the sentence. But there are two rather diVerent ways of doing that. My suggestion is that one of them corresponds to the feeling of squeamishness and the other to a judgment of FALSITY. We get squeamishness if the only way to reject the speaker’s sentence is by challenging its presupposition. If we want to correct the misconception, we will have to deviate from the current trajectory of the conversation and correct the speaker about what the common ground really is like. Suppose that this is a maneuver that we disprefer.

von Fintel, Would You Believe It? 341 Now, if the sentence has entailments that we could in principle falsify independently of discussing its presupposition, we are still in business. It is much easier to see these examples as false, because we do not necessarily have to engage in a debate about the mistaken presupposition. We can move on to a discussion of those entailments and convince the speaker that s/he has no evidence for them. This is what lies behind the intuition that the sentence is false. Of course, it remains a mystery why this procedure puts so much emphasis on salient entities that can serve as footholds for falsiWcation.

7. Conclusion So, whether a sentence with a failed presupposition is judged false or arouses feelings of squeamishness depends on the intricacies of a peculiar fall-back strategy which cares about having salient footholds for rejection. The pattern of truth-value judgments we discovered cannot serve as a direct diagnostic for the presence or absence of presuppositions. The right theory of presupposition will most likely be the one that best explains patterns of presupposition projection, as most researchers have always assumed. But mostly independent of that enterprise, we have seen that truth-value judgments are a rich subject in their own right. Finally, the strategy discussed does not appear to be the only fall-back available. In particular, where presuppositions of uniqueness rather than existence are concerned, people have discussed other ways of ‘repairing’ presupposition failure. Stalnaker (1981, 1984) has argued that conditionals presuppose that there is a unique closest antecedent world. In the face of scenarios where this is unrealistic, he has suggested that a supervenience strategy kicks in. In eVect, the failure of the uniqueness presupposition gets glossed over as long as all of the closest worlds behave the same. In the analysis of donkey sentences, a similar problem arises: see C. Barker (1996) and Geurts (2002) for relevant ideas.

9 Descriptions, Linguistic Topic/Comment, and Negative Existentials: A Case Study in the Application of Linguistic Theory to Problems in the Philosophy of Language Jay David Atlas

1. The Problem Among most philosophers of language and linguists it has remained an unexplained linguistic fact that philosophers’ and linguists’ intuitions diVer markedly on the following two types of statements, the simple declarative statements in (1) and the cleft statements in (2): (1) (2)

a. The king of France is bald. b. The king of France is not bald. a. It is the king of France who is bald. b. It is not the king of France who is bald.

Those with Fregean and Strawsonian intuitions would agree that though (1a, b) would be neither true nor false statements, in stark contrast (2a, b) would be respectively false and true statements. The linguist James McCawley (1981: 241) noted and agreed with these judgments of truth-value but provided no explanation. In this chapter I give an explanation for these data on cleft statements, introduce the notions of topic/comment from linguistics that are required to elaborate the explanation, defend my theoretical approach from an attack by Keith Donnellan, and apply the theory so developed to the vexing and infamous case of negative existence statements. Surprisingly my linguistic analysis of the topics of explicit, informative, negative existence statements, for example, The king of France does not exist, shows that the statement is not about the king of France, and so my analysis contradicts Bertrand Russell’s (1903) analysis in The Principles of Mathematics. It was the ultimately unsatisfactory metaphysical and semantical consequences of his aboutness

Atlas, Topic/Comment, Negative Existentials

343

analysis of negative existence statements in 1903 that motivated Russell (1905) to create his theory of deWnite descriptions. I show that Russell’s (1905) theory of deWnite descriptions is a solution to a linguistic pseudo-problem and, further, that Strawson’s (1964) account of presupposition is not, contrary to the received view, defeated by its alleged (but in fact non-) consequence that The king of France does not exist presupposes that the king of France does exist. Finally I ask what linguistic virtues, if any, Bertrand Russell’s theory of descriptions possesses and draw morals from the history of Russell’s theory for the use of linguistic theory in philosophy of language and philosophical logic.

2. Being Misled by Surface Grammar: Subject Noun-Phrases Versus Focal Noun-Phrases in Cleft Statements Compare a declarative statement (3a) with contrastive stress on ‘Sam’ with the cleft statement (3b): (3)

a. SAM wants Fido. b. It is Sam who wants Fido.

The contrastively stressed statement (3a) has ‘Sam’ as a focal noun-phrase and ‘Someone wants Fido’ as a presupposition, like the cleft of (3b) (see Chomsky 1972). Rhetorically, that is, with respect to focus and presupposition, (3a) parallels (3b). Likewise the normally stressed (4a) and contrastively stressed (4b) parallel (4c). (4)

a. Sam wants Fido. b. Sam wants FIDO. c. It is Fido that Sam wants.

If, like the linguist Gerald Gazdar (1979: 124–5), one were to assume (erroneously) that (5b) parallels (5a), (5)

a. John went. b. It was John who went.

one might also (erroneously) assume that (6b) parallels (6a). (6)

a. Sam wants Fido. b. It is Sam who wants Fido. c. It is Fido that Sam wants.

But as we have just seen in (4), (6b) does not parallel (6a). Statement (6a)/(4a) parallels (6c)/(4c) instead. The order of items in the surface structures in (6) is misleading. One of the oldest lessons, supposedly taught us by Bertrand Russell in ‘On Denoting’ (1905), is that surface grammar can mislead us as to logical grammar, viz. as to the

344

III. Presupposition and Truth-Value Gaps

logically important properties of statements. Certainly that is the lesson that the young Wittgenstein learnt from him in 1912–13 and that Wittgenstein states explicitly in his Tractatus. In this chapter I shall show in what follows that there is an important caveat to be placed on the form this lesson takes. Misled by surface structure one might assume that the logical subject noun-phrase of both statements in (6a, b) is ‘Sam’. On that assumption, in (6b) the focal nounphrase is the same as the logical subject noun-phrase. But in the normally stressed simple statement (6a) the logical subject noun-phrase is ‘Sam’, which is not a focal noun-phrase ‘SAM’. The identiWcation of focus with logical subject would be a linguistic mistake in the normally stressed statement (6a). Likewise the identiWcation of logical subject and focus in (6b) would be a linguistic mistake. Gazdar was misled by surface structure and wrong to pair (5a) with (5b); (5b)/(7b) pairs with the contrastively stressed (7a). (7)

a. JOHN went. b. It was John who went. c. ly(y ¼ John)(GxWx) d. The group of those who went is identical to John.

As Atlas and Levinson (1981: 50–5) have shown, the logical form of the cleft statement (7b) It was John who went is (7c), paraphrased in English by (7d ). (The G operator on a predicate creates a collective term, as linguists call it (Lyons 1977: 315), viz. a groupdesignating term, and the concept of group permits a group of one, viz. John.) Focal singular terms, for example, ‘John’ in (7b), are not subject noun-phrases in logical form, for example, in (7c). The logical subject noun-phrase is the singular term d GxWx e , ‘the one/those who went’. (The special quotation marks around an expression d e ‘ w ’ are Quine’s (1951) quasi-quotation marks.) Note that a collective term is semantically indeterminate between a grammatical singular and plural; for example, one may say ‘The US Administration is . . . ’ or ‘Her Majesty’s Government are . . . ’ Atlas and Levinson (1981), unlike Gazdar (1979), proposed that topic noun-phrases, which are not focal in statements (I distinguish statements from sentences) and not focal nounphrases, be expressed as logical subjects in Russellian logical forms. This proposal allowed for the proper logical treatment of cleft statements. For what is the cleft statement It is Sam who wants Fido about? It is about those (the one) who want (wants) Fido, and says of them (him) that Sam is among them (is him).

3. Strawson’s Suggestions on Topic and Comment Strawson (1971: 89–90) had noted the strong linguistic intuition that (8a) is false rather than neither true nor false, as (8c) would be, the assertion of which falsely presupposes ‘There is a unique king of France’, and McCawley (1981: 241) had noted

Atlas, Topic/Comment, Negative Existentials

345

the strong intuition that (8b) is false rather than neither true nor false. Strawson (1971: 93–4) explained the falsity of (8)

a. The exhibition was visited yesterday by the king of France. b. It was the king of France who visited the exhibition yesterday. c. The king of France visited the exhibition.

(8a) by appeal to the principle that (8a) was not about the king of France. The singular term was not, was not even a constituent of, a topic noun-phrase for statement (8a), and, as Strawson (1971: 93) put it, ‘‘assessments of statements as true or untrue are commonly, though not only, topic-centered in the same way as the statements assessed; and when, as commonly, this is so, we may say that the statement is assessed as putative information about its topic’’. Then Strawson suggested the following principles: (9)

a.

If t is a Topic Noun-Phrase in a statement, and t is a non-referring singular term, then the statement is neither true nor false. (Strawson 1971: 93) b. If t is not a Topic Noun-Phrase in a statement, and t is a non-referring singular term, then the statement is not neither true nor false. (Strawson 1971: 93)

Gundel (1977) and Reinhart (1981) discussed Strawson’s (9a). Atlas (1988, 1989, 1991b) rejected Strawson’s (9a) but accepted Strawson’s (9b) in the form (10): (10)

The Strawson–Grice Condition The existence of a reference of an NP is presupposed in making a statement ONLY IF the NP is a Topic NP (where ‘NP ’ is a metavariable ranging over proper names and simplex deWnite descriptions—not, e.g., d the F of the Ge ). (Atlas 1988: 383–4; 1989: 112; 1991b: 136)

Strawson’s other condition, (9a), the converse of the one that I accept, could be put: (11)

The Unacceptable Strawson Condition If an NP of a statement is a topic NP, then the statement presupposes the existence of the reference of the NP.

For reasons that I discussed in Atlas (1991a), I rejected (11). For example, the assertion of the conjunction ‘There is a queen of England and the queen of England is gracious’, as the earliest linguistic discussions of the Projection Problem for Presupposition (i.e. how the presuppositions of a compound or complex sentence are composed of the presuppositions of its parts; see Karttunen 1973) noted, does not presuppose that there is a queen of England. Then Strawson’s principle (11) would predict that the conjunction is not about the queen of England. But it seems linguistically obvious that the asserted conjunction sentence would, at the least, be about the queen of England. Thus Strawson’s condition (11) makes an incorrect linguistic prediction. So I accepted only (10) of Strawson’s suggested two principles. (This is one of the few conclusions from

346

III. Presupposition and Truth-Value Gaps

the data generated by the treatment of the Projection Problem that had for me more than combinatorial interest, the other being Karttunen’s (1973) refutation of Stalnaker’s (1974) mutual knowledge account of pragmatic presupposition, which led me (Atlas 1975) to take seriously the ‘‘unpresupposed’’ presuppositions, viz. ‘‘accommodated’’ propositions in David Lewis’s (1983b) later sense. See Atlas 1975; forthcoming.) Principle (10) can now explain the heretofore-unexplained linguistic datum of McCawley (1981: 241)—why (12a) is, arguably, neither true nor false, but (12b) is false, despite the occurrence of the vacuous deWnite description. (12)

a. The king of France raises the best racehorses. b. It’s the king of France who raises the best racehorses.

In the cleft statement (12b) ‘‘the king of France’’ is a focal NP instead of a topic NP. So by the Strawson–Grice Condition (10) the existence of the king of France is not presupposed in the statement. Hence, since the nonreferentiality of the singular term can not cause a truth-value gap, McCawley’s (1981: 241) intuition that the statement has a truth-value, FALSE, is explained. Atlas and Levinson’s (1981: 50–5) theory of the logical form of clefts puts the topic NP in the logical subject, so that a cleft of the form (13a) has the logical form (13b): (13)

a. It is John who Fs. b. ly(y ¼ John)(GxFx) c. 9xFx & 8y(Fy ! y ¼ John) d. At least one individual Fs and if anyone Fs, it’s John.

where d GxFxe is a collective or group term d the one=those who Fe , and the predicate says of the group that it is identical to John, in this case a group of one. The distributive reading of the logical form is (13c), paraphrased in (13d). The cleft statement (14a) has most naturally the cooperative, collective understanding (14b), where the predicate ‘lifted the piano’ only applies to the group as a whole and not to the whole by virtue of (14)

a. It’s John, Brian, Rick, Mart and David who lifted the piano. b. lu(u ¼ S[ John,Brian,Rick,Mart,David]) (GxL(x, the piano))

applying to each of its parts and where the group who lift the piano is identiWed with Lesniewski’s mereological sum S of the group’s constituent members (see Ojeda 1993). This approach has the advantage of explaining a heretofore-inexplicable asymmetry between aYrmative and negative statements. Though there is a null set in set theory, there is no null individual in the calculus of individuals, so we have a convenient explanation of the semantical anomaly of the collective interpretation of the aYrmative (15a), where the presupposition that someone lifted the piano and the predication that no one did are contradictory.

Atlas, Topic/Comment, Negative Existentials (15)

347

a. #It’s no one who lifted the piano. b. It’s not no one who lifted the piano. c. :ly(:9x(x ¼ y))(GxL(x,the piano))

The cleft form of the semantically anomalous aYrmative statement would seem to presuppose that someone lifted the piano, and the linguistically acceptable negative statement (15b) It’s not no one who lifted the piano seems to presuppose that someone lifted the piano, yet my semantic account disallows groups of zero elements. There are no null elements as mereological sums; no one does not name an individual or group, and so the aYrmative statement (15a) is anomalous on the collective reading. But why is the negative statement (15b) acceptable? You return to your house to Wnd that the grand piano in the living room has been moved some considerable distance, but the house has been locked and undisturbed. The only living creature in the house was the cat Felicia. So you say to your spouse, It’s not no one who lifted the piano, that is, what is expressed by (15c). What you denied with your not was that the individual or group that lifted the piano is not identical to anyone, by which, of course, you intended to communicate ‘SOMEONE lifted the piano!’

4. Keith Donnellan’s Attack on the Strawson–Grice Condition Contrary to the views of Atlas and Strawson, Donnellan (1981) believes that the later Strawson (1964) of ‘‘Identifying Reference and Truth-Values’’ is mistaken in his use of aboutness, in his appeal to topic NPs; he believes that the early Strawson (1950) of ‘‘On Referring’’ has the more correct intuitions about the truth-value gaps caused by the occurrence of nonreferring singular terms. Though Strawson (1971: 92) brieXy mentions statement-topics and discoursetopics, his theoretically interesting claims are couched in terms of statement-topics. It is hence a slight oddity of Donnellan’s (1981: 135) discussion that he claims that Strawson’s view uses ‘‘as its central idea the conversational circumstances, in particular, what he calls the ‘topic of conversation’ ’’. In fact, when Strawson (1971: 92) Wrst introduces ‘topic’ in his discussion, it is explicitly statement-topic that he characterizes, though he is aware that one can say of a discourse that it has a topic, and his theoretical apparatus explicitly characterizes statement-topics. Donnellan’s (1981: 135) attack is on the adequacy of Strawson’s drawing a distinction between statements with truthvalue gaps and statements with truth-values, by appeal to the failure of reference by only topic singular terms in the former statements. Counter-examples for Donnellan (1981: 135) are comparisons of the existent with the nonexistent. According to Donnellan, where the topic of conversation is the readers of his essay, the statement The readers of this paper will know more about

348

III. Presupposition and Truth-Value Gaps

presuppositions than does the king of France will be neither true nor false, pace Strawson’s analysis. Similarly, Donnellan holds that when he is the topic of conversation and he asserts I am a better volleyball player than the king of France, his assertion is, Donnellan thinks it obvious, neither true nor false. The conclusion that Donnellan (1981: 139) draws is that Strawson’s ‘‘notion of the ‘topic of conversation’ founders on the class of cases I [Donnellan] have called ‘comparisons of the existent with nonexistent’ ’’. The evidence for Donnellan’s claim is explicitly linguistic intuitions; he (1981: 138– 9) remarks that Insofar as I Wnd myself agreeing with Strawson about truth-value gaps, when I consult my intuitions, my intuition is just as strong here that there is no truth-value as it is about any of the more familiar cases . . . It does not matter what one chooses to call the ‘topic of conversation’, our intuitions will tend toward ‘no truth-value’.

(16)

a. John knows more about baseball than does the king of France. b. John’s knowledge about baseball is greater than the king of France’s. c. G(t, t0 ) d. The knowledge of John about baseball is greater than the knowledge of the king of France about baseball. e. John does not know more about baseball than does the king of France. f . The queen of England does not know more about baseball than does the king of France. g. The queen of England knows no more about baseball than does the king of France.

Since there are no arguments, just intuitions about linguistic data, in Donnellan’s objection to Strawson, I will Wrst try to provide a theoretical motivation for Donnellan’s intuitions. Let us begin with a sentence of Donnellan’s sort, viz. (16a) ‘John knows more about baseball than does the king of France’. When the topic NP is ‘John’, Donnellan’s intuition, by contrast with Strawson’s, is that the assertion (16a) is neither true nor false. For the sake of Donnellan’s argument, I want to nominalize (16a) as (16b) ‘John’s knowledge about baseball is greater than the king of France’s’. (Donnellan (1981: 140) actually makes a similar move, so he would not object to mine. As I shall observe later in this chapter, nominalization is also essential in Bertrand Russell’s (1905) original argument for his theory of deWnite descriptions.) If one takes the logical form of (16b) to be (16c) G(t, t0 ), where d te and d t0e are singular terms, d t0e containing as a constituent the vacuous deWnite description ‘the king of France’, for example, as in (16d ) ‘The knowledge of John about baseball is greater than the knowledge of the king of France about baseball’, one might well claim that d t0e , ‘the knowledge of the king of France about baseball’, is a non-referring singular term whose lack of a reference results in a truth-value gap in (16d ), and hence a truth-value gap in (16a). The description ‘the knowledge of the king of France about baseball’ in (16d ) is a complex deWnite description, the case that is explicitly exempted from the application

Atlas, Topic/Comment, Negative Existentials

349

of the Strawson–Grice Condition as stated in (10). This, at least, is an argument for a truth-value gap in (16a). Of course, if one thought presuppositions were involved, one should, as Donnellan would agree, examine the negation of (16a) to see whether presuppositions are preserved, which Donnellan actually does not do in his essay. So, let us consider the negative statement (16e) John does not know more about baseball than does the king of France. Does the negative statement suggest that the king of France has any knowledge of baseball, or even that the king of France exists? I think not, and so neither does the aYrmative (16a) presuppose those propositions. In fact the statement (16f ) The queen of England does not know more about baseball than does the king of France seems Xat-out true, contrary to Donnellan’s intuitions; certainly the more felicitously worded statement (16g) The queen of England knows no more about baseball than does the king of France is true when the queen of England, let us suppose, knows nothing about American baseball. It certainly seems as Xat-out true as the aYrmative statement (17a), with falling intonation on ‘the man in the moon’: (17)

a. Jay knows as much about lichens as the man in the moon. b. I know more about General Relativity than does the man in the moon.

An asserter of (17a) would conversationally implicate (Grice 1989) that Jay knows nothing, or hardly anything, about lichens. And I could assert (17b), with rising intonation on ‘the man in the moon’, and assert another Xat-out truth, contrary to Donnellan’s intuitions, since for a non-physicist I know a bit about General Relativity. The argument that I oVered to motivate Donnellan’s intuitions could be treated as an instance of a principle that Donnellan (1981: 141) explicitly rejects: ‘If there is a principle behind our intuitions it cannot be the simple one that referential terms with no reference result in a truth-value gap when they are used.’ Though it seems unfair of me to pin on Donnellan an argument requiring a principle that he explicitly rejects, I can see no other argument that comes even close to showing what Donnellan wants to claim. This suggests, to me at least, some doubt about Donnellan’s claim. Since Donnellan (1981: 138) agrees with Strawson that when the topic NP of a statement is ‘the readers of Donnellan’s paper’, (18)

Among the readers of Donnellan’s paper will be the king of France.

is false rather than neither true nor false, Donnellan (1981: 139–42) spends the latter part of his essay vainly attempting to Wnd an acceptable principle that will distinguish (18) having a truth-value from (16a) not having a truth-value. In his essay Donnellan considers four possible explanations of his intuitions about the alleged truth-value gap in (16a), only to note, correctly and illuminatingly, that the same explanations would make (18) truth-valueless as well, contrary to his own

350

III. Presupposition and Truth-Value Gaps

intuitions about truth-value. He ends his paper with the sentence ‘I do suggest that intuitions are not simple things to account for’. But the best explanation of Donnellan’s subtle and excellent though vain attempt to account for a diVerence between his linguistic intuitions about the truth-valueless state of (16a) and the truth-valued state of (18) is the following: his intuitions of the alleged truth-value gap in (16a) were mistaken. His comparison statements are no counter-examples to Strawson’s employment of a constraint on the existence of referential presuppositions, viz. that the failure of reference of a simplex deWnite description produces a truth-value gap for the statement containing an extensional occurrence of the singular term only if the simplex deWnite description is a topic NP in the statement. It is fortunate that we can save Strawson’s account from Donnellan’s attack on it, for on it depends the solution to one of the notorious problems in twentieth-century philosophical logic—the problem of negative existence statements.

5. The Problematic Presuppositions of Negative Existence Statements: The Role of Topic/Comment and the Strawson–Grice Condition Hintikka (1986: 267) forcefully put the problem for a Strawsonian theory of presupposition: Strawson and others will have to explain why the existence of a unique present king of France is not presupposed in (1) ‘The present king of France does not exist’. (or, more idiomatically, ‘There’s no such person as the present king of France’) whereas it is presumably presupposed by many other assertions in which the deWnite description occurs, including the famous Russellian example (6) ‘The present king of France is bald’.

Actually Strawson himself never did explain why the existence of a unique present king of France is no presupposed in The present king of France does not exist, but others who have tried include Hintikka (1986) and Atlas (1988). Like Strawson (1964) Hintikka is attracted to the view that statements are construable as answers to questions, and that the questions allow a formulation of what an answer presupposes. In the service of this view, Hintikka (1986: 267–8) continues: The sentence (6) [‘The present king of France is bald’] is most naturally thought of as an answer to the question (7) ‘Is the present king of France bald?’ whereas (1) [‘The present king of France does not exist’] is equally readily construed as an answer to the following one: (8) ‘Is there such a person as the present king of France?’ Now the presupposition of (7) in the precise sense deWned in my theory of questions is equivalent to (9) ‘There exists such a person as the present king of France’ whereas (8) has in my theory a vacuous presupposition.

It is terribly convenient that Hintikka’s question (8) ‘Is there such a person as the present king of France?’ has a vacuous presupposition in Hintikka’s theory of

Atlas, Topic/Comment, Negative Existentials

351

questions, so that its answer (1) ‘The present king of France does not exist’should share the referential presuppositions of its question, viz. none. Of course if one asks why Hintikka believes that question (8) has no presupposition that there is a king of France, his answer is only that it is ‘‘obvious’’ that it does not. After all, why ask a ‘yes–no’ question in which one presupposes a ‘yes’ answer? But then, why assert a denial ‘The king of France does not exist’ in which one presupposes the truth of what is being denied? Is the former question more deeply explanatory of the anomaly of the absence of these referential presuppositions than the latter question? Hintikka (1986) thinks so; I do not. Why does Hintikka accept uncritically his linguistic intuition that ‘The present king of France is bald’ is ‘‘most naturally’’ thought of as an answer to ‘Is the present king of France bald?’ It seems to me that it is just as linguistically natural an answer to ‘Who’s bald?’, ‘Is anyone bald?’, etc. And are Hintikka’s linguistic intuitions really that ‘The present king of France does not exist’ is ‘‘equally’’ natural a reply to ‘Is there such a person as the present king of France?’ Really?! Is it more natural than, say, ‘There’s no such person as the present king of France’? I should not think so (but I wonder how it sounds in Finnish). If you think so, you will Wnd Hintikka’s appeal to intuition compelling. I do not. Not only are these linguistic data uncompelling, the alleged reduction of the presuppositions of statements to the presuppositions of questions entirely begs the explanatory question. Though Strawson himself never explained the puzzle of the presupposition of negative existence statements, we now know what would constitute an explanation. By the Strawson–Grice Condition that the existence of a reference of a simplex NP is presupposed in making a statement only if the NP is a topic NP in the statement, we shall dissolve the puzzle if ‘the king of France’ in The king of France does not exist is not a topic NP. For then there would be no presupposition of the existence of the reference of ‘the king of France’. So we need to consider criteria for topichood of noun-phrases.

6. Some Linguistic Criteria for Topical Noun-Phrases If one wants to identify the topic NP in the statement of the sentence ‘Mikael dined with Jay and Steve’, one simple question to ask is ‘Who or what is the statement about?’ In English the answer is linguistically obvious: Mikael. It is a peculiar feature of English that syntactical subject noun-phrases are typically also topical noun-phrases. So English speakers, uncorrupted by knowledge of the syntax of a formal language like that of Wrst-order quantiWcation theory, correctly take ‘Mikael’to be the topic NP. One might also consider some syntactic variations of the statement, viz. statements of the sentences (a) ‘Mikael, he dined with Jay and Steve’, (b) ‘As for Mikael, he dined with Jay and Steve’, both of which isolate the topic NP. These are paraphrases of the original statement (cf. sentence) in the same context of utterance. These statements

352

III. Presupposition and Truth-Value Gaps

would be felicitous in the context only if it would be appropriate in the context to predicate something of Mikael (see Gundel 1977). Now let us perform a little paraphrase experiment. In a statement that makes a claim of existence, we shall compare for closeness of meaning two pairs, the target statement, T, paired Wrst with one and then another possible paraphrase. Your task, dear reader, is to decide by reXective linguistic intuition, isolated from and independent of all your hopes and dreams for your favorite philosophical or linguistic theory, the better of the two paraphrases for the target existence statement T. Ready? Here’s the Wrst paraphrase pair: (19)

(T) The great-grandson of Jacob Burckhardt exists. (P1 ) As for what exists, the great-grandson of Jacob Burckhardt does.

Now, here is the second paraphrase pair: (20)

(T) The great-grandson of Jacob Burckhardt exists. (P2 ) As for the great-grandson of Jacob Burckhardt, he exists.

Now, which is, intuitively and reXectively, the closer in meaning to ‘The great-grandson of Jacob Burckhardt exists’, (19-P1 ) or (20-P2 )? Really? Is that your Wnal answer? I hope you said (19-P1 ), since (20-P2 ) is the wrong answer. (Over the years I have accumulated large amounts of data from non-linguists and non-philosophers on this paraphrase task. Ordinary speakers of English, in fact 99.9 per cent of this test population, give the correct answer (19-P1 ).) Here is one reason why: a Strawsonian can imagine that a statement of the sentence ‘As for Mikael, he exists’ invokes a presupposition that Mikael exists in the anaphoric use of ‘he’ that is absent from a statement of ‘Mikael exists’, that is, a referential redundancy in the former that is absent in the latter. That makes ‘As for what exists, Mikael does’ a better paraphrase of non-emphatic, informative statements of the sentence ‘Mikael exists’. Parallel arguments support the same conclusion for explicit negative existence statements. A quick argument to this eVect is easily formulated: presuppositions are invariant under main-verb negation. If a statement of ‘Mikael does not exist’ were to presuppose that Mikael exists, so would one of ‘Mikael exists’. But the latter does not; so neither does the former. It is also easy to show directly, for the negative versions of examples like the ones above, that ‘Mikael’ in a non-emphatic, informative statement of ‘Mikael does not exist’ is not a topic NP. By the Strawson–Grice Condition, if it is not a topic NP, the existence of the NP’s designation is not presupposed in a statement of the sentence. (If one restricts the formulation of the Strawson–Grice Condition (10) to simplex deWnite descriptions, I shall treat ‘Mikael’ as the Quinean (1960) deWnite description ‘the Mikaelizer’.) As Strawson (1964) observed, a statement presupposes that a singular term has a reference only if the term is a topic NP. In a non-emphatic, informative statement d t existse or d t does not existe , d te is not a topic NP. Thus the existence statements do not presuppose that d te has a reference. The anomalies pointed out by Hintikka in

Atlas, Topic/Comment, Negative Existentials

353

Strawson’s basic conception of presupposition are removed by a correct linguistic analysis of the topics of explicit and informative existence statements.

7. Some Actual Linguistic Data on Existence Statements During dinner in Cambridge with Stephen Levinson, Mikael Dolfe, a Swedish journalist and Wction writer who had not met Levinson before, asked him whether he had a family, since they were not present at our dinner. Levinson, whose wife Penny Brown and son Nicholas were in Australia, replied (I quote—I wrote it down in my pocket diary at the time, my little eVort at empirical research on discourse), ‘I have a family— they’re not here, but they EXIST.’ Levinson’s emphatic stress on ‘exist’ made sense in the context because he had already established the existence of his family by asserting ‘I have a family’. So his ‘they’, the topic NP in his utterance, did not receive primary stress. In general, topic NPs do not receive primary stress in utterances in which they occur. In his assertion of their existence, viz. of the sentence ‘I have a family’, normal stress patterns in English give ‘a family’ primary stress. In English a speaker does not typically use ‘My family exists’ to establish in a conversation the existence of his family, because normal stress patterns put primary stress at the end of the sentence uttered, while the comment rather than topic role of ‘family’ in ‘My family exists’, which is the same comment role as in ‘I have a family’, would put primary stress at the beginning of the sentence uttered, contrary to the normal stress patterns of English. Thus in a conversation establishing the existence of one’s family, what takes primary stress, and indicates that it is not a topic NP, is the subject NP of ‘My family exists’. The same holds for establishing a claim of non-existence as well. So if it is news to your addressee that there is no winged horse ridden by the Greek hero Bellerophon, one says ‘pegasus does not exist’, not ‘Pegasus does not exist’. I once made this point to the distinguished Princeton University philosopher of mathematics Paul Benacerraf, and in his disbelief at my claim he kept uttering, ‘You think that in denying Pegasus’s existence I say ‘‘pegasus does not exist’’? That’s crazy. I don’t stress ‘‘Pegasus’’ in ‘‘pegasus does not exist’’!’ Why is it that one never has a tape-recorder in one’s pocket when one really needs one, in Princeton, instead of on an island in Papua New Guinea where only 827 people speak a language a grammar for which is currently being written by Stephen Levinson?

8. The Linguistics of Existential Cleft Statements Adequate theoretical attention has not been given to existential cleft statements, for example:

354

III. Presupposition and Truth-Value Gaps (21)

a. #It’s no one who is the king of France. b. It’s not the king of France that exists. c. It’s the king of France that does not exist.

Between the statements (21a) and (21b) in which ‘the king of France’ is respectively a topic and a non-topic noun-phrase, only the latter, the non-‘the-king-of-France’topic-noun-phrase, nonexistence statement, is linguistically acceptable. The alternative negative cleft statement of (21c) ‘It’s the king of France that does not exist’ on some theories of cleft statements entails, and on all theories at least generally conversationally implicates (Grice 1989), that only the king of France does not exist. This is an incorrect prediction, even in the case of a particularized conversational implicature, except in contexts that delimit a list of names or deWnite descriptions, all of which, excepting ‘the king of France’, are singular terms that designate. Hence the closest of the three negative cleft statements (21a, b, c) above to a statement of ‘The king of France does not exist’ is the second cleft statement of (21b) ‘It’s not the king of France that exists’, the topic of which is: what exists. ‘The king of France’ is not a topic noun-phrase in the negative existence statement. As Strawson (1964) observed, a statement presupposes that a singular term has a reference only if the term is a topic NP. In a non-emphatic, informative statement d t existse or d t does not existe , d te is not a topic NP. Thus the existence statements do not presuppose that d te has a reference. The anomalies in Strawson’s basic conception of presupposition are removed by a correct linguistic analysis of the topic noun-phrases of explicit existence statements.

9. Bertrand Russell on Topic and Comment: The Theory of Descriptions as a Solution to a Linguistic Pseudo-Problem In The Principles of Mathematics Russell (1903) thought, prior to ‘On Denoting’ (1905), that statements of ‘Pegasus exists’ were about Pegasus, that is, he thought that in those statements ‘Pegasus’ was a topic noun-phrase. Similarly, in statements of ‘Pegasus does not exist’, he thought that ‘Pegasus’ was a topic noun-phrase. Notoriously he thought that the negative existence proposition was about Pegasus, thus requiring Pegasus to be, even if not to exist. Russell (1903: 44–5) writes: In a large class of propositions . . . it is possible, in one or more ways, to distinguish a subject and an assertion ABOUT the subject . . . The proposition ‘‘humanity belongs to Socrates,’’ which is equivalent to ‘‘Socrates is human’’ is an assertion ABOUT humanity; but it is a distinct proposition. In ‘‘Socrates is human,’’ the notion expressed by human occurs in a diVerent way from that in which it occurs when it is called humanity, the diVerence being that in the latter case, but

Atlas, Topic/Comment, Negative Existentials

355

not in the former, the proposition is ABOUT this notion . . . I shall speak of the terms of a proposition as those terms, however numerous, which occur in a proposition and may be regarded as subjects ABOUTwhich the proposition is.

In his review article on Meinong, Russell (1904: 349) continues: And we distinguish between ‘A is the father of B’ and ‘fatherhood holds between A and B’: the latter, but not the former, is ABOUT fatherhood as well as A and B, and asserts, while the former does not, a relation of fatherhood to A and B. Hence, although, at Wrst sight, the diVerence might seem to be merely a subjective diVerence of emphasis, it results that there is a real LOGICAL diVerence.

This diVerence cannot be incidental for Russell, since he makes essential use of it in a criticism of a view of truth and falsehood later in Russell (1904: 511): If we consider a Relation R between a and b, we should say, when it is false that this relation holds, that there is no such thing as the relation R between a and b. But this argument has been already disposed of, by the contention that the being of this relation is not what the proposition ‘a has the relation R to b’ really aYrms.

The being of the relation is not aYrmed by the proposition because the proposition is not ABOUT the relation. Russell (1903: 49) gives an analysis of the proposition ‘‘A is’’, which, he writes, ‘‘holds of every term (A) without exception. The ‘is’ here is quite diVerent from the ‘is’ in ‘Socrates is human’; it may be regarded as complex, and as really predicating Being of A’’.

In later linguistic terminology, for Russell the topic of the proposition ‘‘A is’’ is A; the comment is that it has Being. But, not for the Wrst time, Russell’s ignoring the use/ mention distinction gets him into trouble. What Russell ignores is that the linguistic observation that a statement contains a topic-designating expression, viz. that the expression is a topic noun-phrase, no more entails that there is a topic or designatum that the statement is about than that a picture is a centaur-depicting picture entails that there is a centaur that the picture depicts. (An exception would be a statement in a ‘‘logically perfect’’ language, in Frege’s sense, and for topic-designating expressions that are ‘‘logically proper names’’ in the sense explored by Russell (1918/1919).) Furthermore, in 1903 Russell simply gets the aboutness of existence statements wrong, and it is aboutness that is explanatorily prior to ‘term’ or reference (Russell 1903: 44–5). As Russell (1919: 169) later remarked, it was ‘‘argued, e.g. by Meinong, that we can speak ABOUT the golden mountain, the round square, and so on; we can make true propositions of which these are the subjects; hence they must have some kind of logical being’’. But it is simply not true that in purportedly informative, true assertions of d t existse or of d t does not existe that d te is a topic noun-phrase. Just because English is peculiar in the wholesale coincidence that the syntactical subject noun-phrase in a sentence and the topic noun-phrase in assertions of the sentence are

356

III. Presupposition and Truth-Value Gaps

often the same, there is no reason to think that in ‘‘the language of thought’’ or in natural languages other than English they should not be distinct. Translations of these existence statements into Japanese, for example, fail to carry the -wa topic marker on ‘Pegasus’. In translations into the Philippine language Tagalog, the topic marker again fails to appear. And in one of the aboriginal languages spoken by some 800 speakers on the northern coast of Papua New Guinea and studied by anthropological linguist William A. Foley (1997), existence statements fail to place an otherwise grammatically required topic marker on the noun-phrase (pers. comm. by W. A. Foley, Max Planck Institute for Psycholinguistics, Nijmegen, 1997). The noun-phrase in informative utterances of ‘Pegasus exists’ or of ‘Pegasus does not exist’ is not a topic noun-phrase in either aYrmative or negative existence statements, and so those statements are not about Pegasus. They are about what exists, or as Quine (1980: 1–19) would say, they are about everything. The metaphysical problem of Pegasus’s being, which it took Russell (1905) the invention of the Theory of DeWnite Descriptions to solve, simply does not arise from the meaning and aboutness properties of negative existence statements, correctly understood (see Atlas 1988, 1989). Russell’s theory was invented to solve a linguistic pseudo-problem.

10. The Philosophical Motivation for Russell’s Theory of DeWnite Descriptions From a historical point of view, it is interesting to note that in the year following the publication of his article on Meinong, Russell in ‘On Denoting’ (1905) espoused precisely the view of falsity that he had rejected in 1904: If ‘aRb’ stands for ‘a has the relation R to b’, then when aRb is true, there is such an entity as the relation R between a and b; when aRb is false, there is no such entity. Thus out of any proposition we can make a denoting phrase, which denotes an entity if the proposition is true, but does not denote an entity if the proposition is false. E.g. it is true (at least we will suppose so) that the earth revolves round the sun, and false that the sun revolves round the earth; hence ‘the revolution of the earth round the sun’ denotes an entity, while ‘the revolution of the sun round the earth’ does not denote an entity. The propositions from which such entities are derived are not identical either with these entities or with the propositions that these entities have being. (Russell 1905: 490–1)

What is historically fascinating is to see that Russell cannot put together the insights of his paper on Meinong (1904) with his theory of deWnite descriptions (1905). In 1904 he abandoned the view that the falsity of aRb meant that there is no such thing as the relation R between a and b, on the (correct) grounds that aRb is not ABOUT R. So R is not a ‘‘term’’ in the proposition.

Atlas, Topic/Comment, Negative Existentials

357

Once he has the 1905 theory of deWnite descriptions, he reverts to an account of relational propositions in which R is a ‘‘term’’, which of course produces famous diYculties for a theory of judgment, diYculties lucidly explored by Russell himself for the next Wfteen years. It is as if he were so delighted that ‘the revolution of the sun round the earth’ does not denote any entity, rather than, as in his 1903 view, denotes a nonexistent entity, that he abandons his earlier insight into the aboutness of relational statements, tacitly expecting that his new theory of denotation will solve whatever problems might arise. The burden is shifted from the semantically more astute observation of 1904 concerning the aboutness of propositions to the 1905 theory of the denotation of singular terms. This is a retreat from the Fregean concern with sentences to the Idealist logicians’ concern with terms. Russell’s new theory of denotation is grafted on to his less astute assumptions of 1900–2. Those assumptions were a blend of Peano’s logic and F. H. Bradley’s Idealist theory of judgment. Only a philosopher steeped in F. H. Bradley’s Principles of Logic (1883), as Russell was, could say, as Russell did in ‘On Denoting’ (1905), that ‘The earth revolves round the sun’ is true if and only if ‘the revolution of the earth round the sun’ denotes an entity, where the relation expressed by the proposition is turned into a deWnite description, and then by Russell’s theory into ‘‘an adjective of the real world’’, to use Bradley’s (1883) expression. Russell’s theory of deWnite descriptions transforms singular judgments into general ones, which is consistent with Bradley’s attack on singular judgments—the same singular judgments that Russell will again defend the existence of in ‘The Philosophy of Logical Atomism’ (1918/1919). Furthermore, the Fregean and Russellian account of existence as a second-order property of properties rather than as a Wrst-order property of individuals is consistent with Bradley’s Kantian doctrine that existence is not a quality. Finally, for a deWnite description to lack a denotation is, on Russell’s theory, for certain properties not to be true of anything. And so, for a relational proposition to be false is for these properties not to be true of anything, that is, not to be ‘‘adjectives of the real world’’. Russell’s 1905 theory of denotation saves F. H. Bradley’s theory of judgment—Bradley’s denial of the existence of singular judgments and Bradley’s account of false judgments. In ‘On Denoting’ (1905) Russell abandons the insight into the structure of propositions gained by his reXections on Meinong’s realism and reverts to F. H. Bradley’s Idealist theory of judgment, now thinly disguised by Russell in Peano’s logical notation as his new theory of denotation of denoting expressions. For all its novel logical machinery, Russell’s theory of deWnite descriptions was a metaphysical and epistemological throwback, an abandonment of the realist insights of Frege and Meinong and of his own related 1904 account of truth and falsity, motivated ultimately by his need to avoid the consequences of his incorrect linguistic and, on his view, logical theory of the aboutness of existence propositions. Once the intuitive, linguistic mistake about topic noun-phrases had been made, his confusion of use and mention and his attraction to G. E. Moore’s realist metaphysics had pushed the pre-1905 Russell into unrestrained

358

III. Presupposition and Truth-Value Gaps

Meinongianism. Meinongianism then forced him back into the embrace of Bradley’s Idealism, suitably Peano-ized by his theory of deﬁnite descriptions.

11. Is Russell’s Theory of DeWnite Descriptions a Good Linguistic Theory? In March 1973, during the Open Discussion of the University of Connecticut conference on ‘‘Language, Intentionality, and Translation Theory’’, Saul Kripke (1974) put the problem as follows: Where someone puts this question in a form like ‘Is the present king of France bald?’, the informant may be puzzled and not say ‘No’. But if instead you put it to him very categorically, say Wrst specifying an armament program to make it relevant and then saying ‘The present king of France will invade us’, the guy is going to say ‘No!’, right? So what type of response you’ll get from the informant will in fact depend on cases. There isn’t the clear verdict in favor of Strawson that would be imagined by concentrating on a narrower class of cases. So it seems that if you want a uniform theory you have to explain one or the other class of cases away. Or so it might seem at Wrst blush anyway. And then the evidence seems hard to balance between which class of cases should be explained away. A Strawsonian might say ‘‘Oh, the ‘No!’ to ‘The king of France will invade us’ is not saying ‘That’s false’ but is a rejection of the statement as having a false presupposition.’’ Because you sometimes say ‘No’ when something is neither true nor false. On the other hand, a Russellian might say ‘‘You’re hesitant to give a verdict when a presupposition isn’t fulWlled, even though you know it’s false.’’ So either man has some class of cases to explain. Isn’t that right? Then the verdict isn’t clear, and this is where the puzzle in the continuing philosophical debate comes from.

In the history of physical sciences, when two quite plausible theories have battled each other to a draw, the stalemate has typically been overcome when it has become clear (a) that the evidence was indecisive because it was the wrong sort of evidence to use in evaluating the competing theories. It was the wrong sort for many reasons, but one reason is typically that the evidence does not reveal (b) that the stalemate arises from a fundamental error shared by what were otherwise competing theories. This, I have argued (Atlas 1989), is exactly the case here. Russell and Strawson were both wrong, for reasons that I have discussed elsewhere and shall not repeat here. (For a related, independent argument, see Katz 1987: 179–80; for discussion see Kent Bach, this volume.) The moral of Kripke’s remarks is that the kind of evidence he considers, assent and dissent to queries in a Quinean (1960) fashion, will not decide between such theories. Quite aside from its mathematical advantages, Russell’s theory of deWnite descriptions has seemed intuitively appealing to many logicians, philosophers, and even some linguists. Even if they accept Strawson’s criticism of Russell, most logicians and philosophers regard the choice between Russell’s classical bivalent theory and a truth-value

Atlas, Topic/Comment, Negative Existentials

359

gap theory as a pragmatic, not a principled choice. But why, despite Strawson’s criticism, does Russell’s theory have some intuitive linguistic appeal? A standard answer is: because Russell’s theory gets the truth conditions right (or almost right, or right enough for deductive purposes) and because, as Russell (1905: 485) claimed, the theory preserves the Indiscernibility of Identicals, the Law of Excluded Middle, and, most importantly, solves the semantic puzzle that a nonentity can be the subject of a proposition, and in particular the semantic puzzle that it must always be contradictory to deny the being of anything. The solution, as Russell (1919: 170) later phrased it, was that ‘The F does not exist’ is true when ‘the F ’ describes nothing real, rather than when it describes something unreal, a case of not describing (not denoting) any entity rather than a case of describing (denoting) a nonentity. Nonentities are not the subjects of propositions, and it is not contradictory to deny the being of anything. Nevertheless, none of this, and especially not the standard remark about the truth conditions, answers my question: why does Russell’s theory have some intuitive, linguistic appeal? Russell (1956: 250) himself, in his ‘‘The Philosophy of Logical Atomism’’, knows the answer, but he does not know that it is the answer. He writes, ‘‘when I say ‘The author of Waverley exists’ I am not saying anything about the author of Waverley’’. My linguistic analysis of Topic/Comment for existence statements shows why there is some intuitive, linguistic plausibility in Russell’s theory. He gets ABOUTNESS right! So the logical structure of an existential statement is not radically diVerent from a grammatical (viz. its topic/comment) structure. Of course Russell’s logical form is diVerent from surface form, but his logical form is consistent with my analysis of the topic/comment structure of informative existential statements. On Russell’s view, ‘Spies exist’ is about a propositional function ‘x is a spy’ and he says of it that it is instantiated (by at least two individuals, if one takes the grammatical plural seriously). But we need not interpret the quantiWcation in this manner. We might view the existential ‘A spy exists’ as about the domain of quantiWcation, what exists, asserting of it that it contains at least one spy. On this view of the matter, logical analysis and linguistic analysis agree. The noun-phrase ‘a spy’ is not a topic noun-phrase in ‘A spy exists’; the statement is not about a spy. But nor is the noun-phrase ‘The tallest spy’ a topic noun-phrase in ‘The tallest spy exists’; the statement is not about the tallest spy—and not because one description is indeWnite and the other deWnite! The same holds for the negative existence statements. On this point Russell’s theory of deWnite descriptions is entirely right. Russell’s problem was not, as has been famously claimed by Quine (1980: 1–19), Russell’s confusing meaning with naming. (Russell never confused linguistic meaning with naming, as anyone who has studied section 51 of Russell (1903) should know.) The problem was Russell’s mistaking what negative existence statements were about, or to put it in his 1903 terminology, mistaking what the subjects of negative existence propositions were. (Indeed, his use of the expression ‘subject’ indicates his confusions, one of use and mention and one of syntactical subject noun-phrase with topical noun-phrase.)

360

III. Presupposition and Truth-Value Gaps

The example of cleft statements with which I began this chapter shows that not every singular term in a statement will designate what the statement is about. So by the Strawson–Grice Condition an extensional statement need not presuppose the existence of the referent of every singular term that it contains. In particular, it does not presuppose the existence of the referent of a deWnite description that is the syntactical subject noun-phrase of an aYrmative or negative existence statement. Informative statements of The king of France does not exist are not about the king of France and so, by the Strawson–Grice Condition on the presuppositions of topic noun-phrases, pace Donnellan (1981) and Hintikka (1986), do not anomalously and inexplicably presuppose the existence of a king of France. Strawson’s (1964) account of presupposition survives its alleged inability to explain the presupposition-free character of explicit, informative, negative existence statements.

PA RT I V

Representation of DeWnites and IndeWnites in Semantic Theory

The chapters in this section all oVer novel semantic accounts of deWnite and/or indeWnite expressions. All in some way challenge the Russellian orthodoxy, although they do not necessarily agree amongst themselves. Russell (1905) analyzes sentences of the form ‘The F is G ’ as existential generalizations that contain no expressions purporting to refer to the F. On analysis such sentences resolve into a complex of quantiWers, predicates and variables, none of which can be construed as singular referring expressions. Frege (1892/1970) on the other hand treated sentences of the form ‘The F is G ’ as subject–predicate sentences, and expressions of the form ‘the F ’ as singular referring terms. The referent of a deWnite description is the unique entity, if any, that it denotes. Frege held also that the referent of a sentence is a function of the referents of its component expressions. Thus, if one of its component expressions lacks a referent, the sentence as a whole lacks a referent too. Since Frege identiWed the referent of a sentence with its truth-value, it follows that a sentence containing a non-referring term (such as ‘the largest prime’) lacks a truthvalue. Since Frege thought that it would be unacceptable to allow bearerless singular terms into a scientiWc language, he proposed to arbitrarily assign referents to nondenoting terms (e.g. in one version of Frege’s view such terms are assigned the empty set as their referent). Russell did not like the artiWciality of Frege’s solution to the problem of nondenoting deWnite descriptions. On the other hand, Russell’s robust sense of reality prevented him from introducing a class of nonexistent entities to be the referents of nondenoting terms, as Meinong (1904/1960) did. Russell’s solution to this problem was to deny that deWnite descriptions (and by extension ordinary proper names) are referring terms. Russell’s views about the nature of singular reference bolstered this conclusion. He claimed that the meaning of a genuine referring term (a logically proper name) is simply its referent, and hence if it lacked a referent it would be meaningless. Since sentences containing nondenoting descriptions such as ‘the present king of France’ are not meaningless, these expressions cannot be genuine referring terms. Besides, Russell thought, one cannot use a referring term unless one is acquainted with its referent. The

362

IV. Deﬁnites and Indeﬁnites in Semantic Theory

only objects of acquaintance that Russell allowed were sense data, universals, and (possibly) the self. (He changed his mind a number of times about whether we can be acquainted with the self.) So, for example, we cannot use the description ‘the current president of the USA’ to directly refer to George Bush, since none of us is directly acquainted in the relevant sense with George Bush. In fact, Russell’s view led to the rather counter-intuitive conclusion that only ‘this’ and ‘that’ (which he took to be names of sense data) and perhaps ‘I’ are genuine referring terms. In other words, Russell’s notion of a referring term was constrained both semantically and epistemologically in such as way as to rule out the possibility that deWnite descriptions are genuinely referring terms. Mark Sainsbury (this volume) challenges Russell’s claim that sentences of the form ‘The F is G ’ are existential generalizations. He argues that we can give an account of deWnite descriptions that delivers the same truth conditions for description sentences as the ones that Russellians would give, yet that treats deWnite descriptions as referring expressions (REs) and the sentences in which they occur as subject–predicate (S–P) sentences. Sainsbury notes that there is a certain asymmetry in the way in which sentences and REs are treated in contemporary semantic theory, which has taken on board many of Russell’s assumptions about reference and REs. The Russellian view is that to understand an RE one must know what object it refers to. But we would not say that to understand a sentence one must know its truth-value, which would be the analog as applied to sentences of the Russellian view about REs. Rather, we would say that to understand a sentence is to know its truth conditions. In Sainsbury’s semantics, REs are associated with reference conditions rather than referents, just as sentences are associated with truth conditions rather than truth-values. A consequence of this shift is that empty names become unproblematic. To understand an RE is simply to know that it purports to refer, not that it actually refers. On this view we can know without further investigation which expressions are the REs. On the contemporary Russellian view, we would, for example, have to know something about astronomy to know whether ‘Vulcan’ is an RE. Sainsbury spells out some of the formal details of his alternative semantic proposal, by stating the relevant clauses of a Davidsonian-style truth-conditional semantic theory. He states axioms for REs and for the S–P sentences toward whose truth conditions the REs contribute. An S–P sentence is true just in case there is something to which the subject-expression refers and of which the predicate is true; falsehood is absence of truth. This is an essentially free logical truth condition, for it treats S–P sentences with non-referring terms as false. Sainsbury’s is a negative free logic, similar to that proposed by Burge (1974).1 1 Karel Lambert coined the term ‘free logic’ in the 1960s. A free logic is one free of existence assumptions with respect to its terms, both general and singular, but whose quantiWers are treated as in standard quantiWer logic. Free logic allows singular terms that do not refer to existing things, such as ‘Pegasus’ or ‘the round square’.

IV. Deﬁnites and Indeﬁnites in Semantic Theory

363

Sainsbury’s account of REs requires an understanding of reference that is at odds with Russell’s views on the matter. As noted, contemporary semantic theory has taken on board many of Russell’s semantic and epistemological assumptions about reference and REs, even if these have been somewhat modiWed. For example, although contemporary Russellians do not agree that the objects of reference are sense data with which we are directly acquainted, they do think that direct reference requires some sort of intimate epistemic contact with the object in question—that we must be en rapport with the object. Sainsbury gives reasons for rejecting the claim that REs require language users to be en rapport with their referents. He also tackles the assumptions that REs must be simple, must have a referent, are not subject to scope ambiguities, are rigid, and are inter-substitutable everywhere with co-referring REs without aVecting the truth-values of the sentences in which they are embedded.2 It is important to note that to treat descriptions as REs in Sainsbury’s free logical sense is not to treat them as referential descriptions in Donnellan’s sense. It is not to oVer an account of the truth conditions for description sentences of the following sort: If there is a unique x such that speaker u (on a given occasion) intends his utterance to concern x and thinks he can achieve appreciation of his intention by using ‘the F ’, then ‘Theref F is G ’ is true as uttered by u (on that occasion) iV x is G. Since on Sainsbury’s account ‘The F is G ’ is given Russellian truth conditions (‘The F is G ’ is true just in case something is uniquely F and it is G ), one may wonder how Sainsbury’s account can deal with the referential/attributive distinction. Sainsbury argues that his view is compatible with either a semantic ambiguity or a Gricean pragmatic account of this distinction. (For further discussion of this distinction, see the chapters in Part II of this volume). Although Sainsbury’s account assigns Russellian truth conditions to description sentences, his referential proposal disagrees with Russell’s about the logical form of description sentences. The logical form of a description sentence of the form ‘The F is G ’ (or ‘The Fs are Gs’) is spelt out in a theorem of a Davidsonian-style semantic theory. In such a theory we strive to give homophonic truth conditions. That is, the logical form of a sentence is the sentence itself, if it can feature in a theorem of an interpretive Because of this, the rule known as Universal Instantiation is qualiWed in free logic. One may instantiate a universal only with terms, t, for which it holds that 9x x ¼ t. A negative free logic is one that regards all subjectpredicate statements involving at least one term that does not refer to an existent entity as false. A positive free logic is one that counts at least some statements containing terms that do not refer to existing things, such as ‘Pegasus is identical to Pegasus’, as true. A neutral free logic is one that counts all statements that contain at least one term that does not refer to an existent entity as truth-valueless. Some (though not all) free logicians agree with Meinong that though terms such as ‘Pegasus’ may not refer to existent entities, they nevertheless do refer. For further explanation of the diVerences between the views of free logicians and those of Meinong, as well as between those of Russell and Frege, see Lambert (1983: ch. 5) and Haack (1996: ch. 7). Lambert (2001) gives a very accessible introduction, which includes details about the internal diVerences between negative, positive, and neutral free logics. 2 Sainsbury (2002: essays IX and XII) oVers further arguments for his account of REs, focusing on proper names rather than deWnite descriptions.

364

IV. Deﬁnites and Indeﬁnites in Semantic Theory

truth theory for the language as a whole. Otherwise it is the closest approximation to that sentence that meets the condition of semantic accessibility. Sainsbury notes that Russell does not distinguish between giving a philosophical analysis and uncovering logical form. It could be that a word that features homophonically in semantic theory deserves substantial philosophical analysis. Sainsbury claims that just this is the case for ‘the’. We may gain philosophical illumination by analyzing descriptions in Russellian fashion, but semantic theory requires the homophonic logical forms assigned to description sentences by his referential approach. While Sainsbury accepts the ‘translational’ approach to natural language semantic theorizing presupposed by Davidsonian-style truth-conditional semantics, Joseph Almog (this volume) mounts a challenge to this approach. The translational approach aims to assign to each sentence of a natural language the proposition it expresses, that is, its truth-conditional content. Almog calls this ‘semantics from below’, since it aims to characterize meaning as much as possible from the perspective of the users of the language and in terms of resources available to these users. Almog argues that for three interesting classes of sentences that he calls the ‘Donnellan classes’, this approach is inadequate. He oVers what he regards as a more adequate conception of semantics, one that he calls ‘semantics from above’. This alternative approach takes the perspective of an ‘omniscient observer of history’ in spelling out the truth conditions for this class of sentences. The three classes of sentences that Almog is interested in are the following: (a) true negative existentials such as ‘Vulcan does not exist’; (b) true empty predicatives, such as the singular ‘Vulcan seemed hot to Leverrier’ and the plural ‘Unicorns seem to be approaching’; (c) true referential uses of deWnite descriptions, such as ‘The man drinking martini is sad’. Almog argues that these sorts of sentences present translational semantics with two problems, which he calls the ‘missing content problem’ and the ‘problem of mind cross-identiWcation’. If translational semantics must associate a content with every natural language sentence, then sentences with nonreferring terms present a problem, since they will induce gaps in content. Also, it will be hard to explain what makes two people co-thinkers when both are thinking about Vulcan, say. There is no externally existing object (Russellian content) to unite these two episodes of thinking and there may be no ‘modes of presentation’ (Fregean contents) shared by these thinkers. Almog rejects Russell’s solutions to these problems. Russell oVers what Almog calls ‘inauthentic’ logical forms for these sentences—ones that diverge from their apparent surface subject–predicate forms. He also rejects the solution oVered by free logicians, who he argues must posit an ontology of nonexistent entities, so that empty names turn out after all to be loaded.3 3 As Sainsbury (this volume) points out, Almog’s characterization of free logic assumes a version diVerent from the sort of negative free logic that Sainsbury himself advocates. See also n. 1 above, where it is

IV. Deﬁnites and Indeﬁnites in Semantic Theory

365

According to Almog, Donnellan begins with the insight that it is not essential to the very idea of a name to have a bearer. What unites all natural language names, whether empty or loaded, is that they have a natural history of use within the language. This use can be investigated and characterized from the point of view of an omniscient observer of history (OOH). Thinking of semantics from the perspective of such an OOH, Donnellan suggests the following truth condition for the true negative existential ‘Vulcan doesn’t exist’: ‘ ‘‘Vulcan does not exist’’ is true iV the history of the use of ‘‘Vulcan’’ ends in a block’. The bulk of Almog’s chapter consists in extending and reWning these insights of Donnellan’s. For instance, he rejects Donnellan’s idea of a block, suggesting instead that all uses of names must have a source, although the nature of these sources may diVer for empty and loaded names. Almog suggests the following as the basic truth condition for simple subject–predicate sentences (suppressing references to worlds of evaluation): A sentence of the form ‘N is F ’ is true as used in context c at time t iV the source of the chain traced by the OOH leading to our use of ‘N ’ in c is F at t. Almog shows how this can be extended to account for referential plurals (‘Unicorns/ bears appear to be approaching’) as well as referential uses of descriptions (‘The man drinking martini is sad’). This latter explanation generalizes in a way that enables Almog to account for referential uses of many other sorts of expressions, including indeWnite expressions. Another virtue of the account is that it can easily solve the problem of mind cross-identiWcation. People may be united as co-thinkers by facts that can be traced by an OOH, even if these facts are unavailable to the thinkers themselves. The various episodes of thinking will be united if they can be traced to a common source. Almog compares this view about the semantics of names to Kaplan’s (1989a, b) account of the semantics of indexicals. Almog thinks there is a great deal of similarity between these accounts. Both reject the idea that giving truth conditions is a matter of associating sentences with a ‘marketplace synonym’. One cannot give the truth conditions for a sentence such as ‘I walk’ by pairing it with some content, because the content of this sentence varies from context to context. Instead, Kaplan oVers truth conditions of the following sort: ‘ ‘‘I walk’’ is true in context c iV the agent of c walks at c’. Kaplan draws a distinction between the character and the content of an indexical. The character associated with an indexical determines its content in a context. For example, the character associated with ‘I’ would be a rule that Wxes its referent as the agent in the context of use. However, Almog thinks that Kaplan does not go far enough in rejecting translational semantics, because Kaplan’s characters are supposed to play a certain cognitive role in the lives of ordinary speakers. For instance, they are supposed to be what competent speakers know when they master the use of such indexical expressions. Almog suggests that Donnellan’s is the more radical view, because it suggests that in noted that not all free logicians treat terms such as ‘Vulcan’ as terms referring to nonexistent entities, a` la Meinong.

366

IV. Deﬁnites and Indeﬁnites in Semantic Theory

doing semantics we need not use only those resources that are available to the marketplace users themselves.4 Peter Ludlow and Gabriel Segal (this volume) are concerned with Russell’s treatment of the contrast between deWnite and indeWnite descriptions. Russell analyzed sentences containing ‘a(n) F ’ as well as those containing ‘the F ’ as existential generalizations. The principal diVerence between these on the Russellian analysis is that the latter entail claims of uniqueness. Consider the following sentences with their Russellian analyses given below: (1) (10 ) (2) (20 )

A job candidate will be here at 2.00 p.m. 9x (Candidate(x) ^ Be-here-at(x, 2.00)) The job candidate will be here at 2.00 p.m. 9x (Candidate(x) ^ 8y (Candidate(y) ! y ¼ x) ^ Be-here-at(x, 2.00))

Clearly the only diVerence between these two analyses is that (20 ) entails that there is only one individual who satisWes the predicate ‘Candidate(x)’. Ludlow and Segal note that in languages other than English, the distinction between deWnite and indeWnite descriptions is not marked, at least on the surface (e.g. in Russian, both ‘the man’ and ‘a man’ would be expressed by the Russian equivalent of ‘man’).5 They suggest there are two possible ways of explaining the discrepancy between English-like and Russian-like languages. It could be that Russian-like languages have deWnite and indeWnite markers, but they are unpronounced elements. These languages would have more semantic structure than meets the eye. Or it could be that English-like languages do not have two semantic markers. Rather they have two surface markers with the same literal meaning. These languages would have less semantic structure than meets the eye. The latter possibility is the one Ludlow and Segal explore. Ludlow and Segal argue that, contra Russell, there is no semantic distinction between (1) and (2). The semantic contribution of both ‘an F ’ and ‘the F ’ is simply ‘9 x F (x)’. Any understanding of deWniteness and/or uniqueness associated with the use of these expressions is something that must be pragmatically derived, using standard Gricean conversational maxims and principles. They point out that the strategy of 4 A related view, inspired by but distinct from Donnellan’s, is developed by Perry (2001: ch. 8). Some skeptical readers have wondered what the truth conditions supplied by an OOH are for, if they do not play any sort of cognitive role in the lives of language users. Such truth conditions would seem to have no role to play in a theory of language understanding or in an account of what we know when we know a language (unless, possibly, we take an externalist approach to such knowledge). Almog would presumably reply that the OOH’s truth conditions play a role in a semantic anthropology. According to Wettstein (1986), the semantic anthropologist ‘is doing the anthropology of our institutions of natural language, and he wants to understand the institutionalized conventions in accordance with which our terms refer. . . . There is no reason to suppose that, in general, if we successfully uncover the institutionalized conventions governing the references of our terms, we will have captured the ways in which speakers think about their referents’ (1986: 124). See also Almog (1984) and Wettstein (1988). 5 i.e. it is not marked on the surface by a single morpheme. In Mandarin Chinese, for example, which also lacks deWnite and indeWnite articles, the distinction between deWniteness and indeWniteness is marked at least partly by word order. It is true however that pragmatic factors also play a large role in conveying deWniteness and indeWniteness in Chinese.

IV. Deﬁnites and Indeﬁnites in Semantic Theory

367

arguing for a minimal semantics and a rich pragmatics is one that has been taken often before. For instance, Kripke (1977) argues that the distinction between referential and attributive uses of deWnite descriptions can be accounted for by assuming a univocal Russellian semantic account for all uses and explaining referential cases as ones where a singular proposition is conversationally implicated. Ludlow and Neale (1991) argue for a similar view regarding the referential, deWnite, speciWc and quantiWcational uses of indeWnites. All indeWnites express existential generalizations, although some uses of an indeWnite may conversationally implicate another (e.g., singular) proposition in addition. In a slightly diVerent vein, Grice (1961: 126–9) argues that the distinction in meaning between ‘and’ and ‘but’ is not one that aVects truth conditions. Any sense of contrast conveyed by the use of ‘but’ is a conventional implicature, as in ‘She is poor but honest’, which conventionally implicates that poverty and honesty do not generally go together. For Grice, there is no diVerence in the contributions to truth-conditional content made by ‘and’ and ‘but’. Their diVerence is a matter of implicature and hence a pragmatic diVerence.6 In the spirit of Grice, Ludlow and Segal propose that the distinction between ‘a(n) F ’ and ‘the F ’ is not one reXected in truth-conditional content, but is rather a distinction in what these expressions conventionally implicate. The use of the latter conventionally implicates that its object is given in the conversational context, whereas the use of the former conventionally implicates that new information is being introduced into the context. Furthermore, there is no conventional implicature of uniqueness associated with singular deWnite descriptions, although in some cases a further (conversational) implicature of uniqueness will be derived. For instance, if a speaker utters the sentence ‘The murderer of Smith is insane’ she will have conventionally implicated that the murderer in question is given in the context, and since she did not use the plural ‘murderers’ she will further have conversationally implied that there is just one murderer in the context, and hence just one murderer in the domain of discourse as Wxed by the context. However, there will be cases in which these implicatures are overridden. For example, where the use of a predicate itself ensures uniqueness, as in ‘the tallest man in the world’, the givenness implicature associated with the deWnite article is blocked. And Ludlow and Segal give several examples of uses of singular deWnite descriptions that are not associated with assumptions of uniqueness, such as ‘John was hit in the eye’, which can be perfectly felicitous without assuming that John is a one-eyed man. The idea that the contrast signaled between ‘a(n)’ and ‘the’ is the given/new contrast was suggested by Heim (1982, 1983a). However, for Heim, givenness is a pragmatic presupposition of the use of the deWnite, rather than a conventional implicature. Heim 6 For a discussion of the diVerence between conversational and conventional implicatures, see Grice (1989: 25–6, 41, 359–65). For a critique of the notion of conventional implicature, see Wilson and Sperber (1993) and Bach (1999b). Note that it is possible to agree that ‘and’ and ‘but’ make the same contributions to truthconditional content and yet to deny that this means that ‘and’ and ‘but’ are semantically identical. This only follows if all aspects of semantics are reXected in truth-conditional content, which is certainly not the case. Indexicals have characters that do not contribute to content. So it does not follow immediately that the diVerence between ‘and’ and ‘but’ is non-semantic (pragmatic) just because it is non-truth-conditional.

368

IV. Deﬁnites and Indeﬁnites in Semantic Theory

recognizes that certain uses of deWnite descriptions are felicitous even though their objects are not given in the context. In these cases she argues that listeners accommodate the presupposition of givenness, by adding information to the set of assumptions shared by conversational participants (to the common ground). This will happen so long as the information needed to accommodate the presupposition is consistent with the common ground and plausible. For example, someone can open a conversation by telling you ‘Sorry I’m late. The dog got out of our yard and I had to get him back in’. Even though you had no idea that the speaker owned a dog, if you trusted her sincerity, you would accommodate the assumption that she has a dog, since it is assumed that she is an authority on whether or not she has a dog. Some assumptions may be harder to accommodate, on grounds of plausibility. If instead of ‘the dog’ the speaker had used the description ‘the giraVe’, it would be harder to accommodate the presupposition of her description, since people do not generally have giraVes for pets. Ludlow and Segal do not discuss the phenomenon of accommodation or say whether and how it would be implemented in their Gricean framework.7 Ludlow and Segal marshal a considerable body of evidence in support of their claim that uniqueness is not part of the meaning of ‘the’. They are able to deal with a range of phenomena and in several cases to give explanations for things that are not so easily explained on the Russellian view. For instance, they can explain why we feel that there is some truth to a speaker’s utterance of ‘The murderer of Smith is insane’ even if it turns out that there is more than one murderer. They can give a simple explanation for the lack of uniqueness in conditional environments, such as ‘If a bishop meets another bishop, the bishop blesses the other bishop’. They can give an elegant account of possessives in English that brings these more in line with parallel constructions in languages such as Italian and French. Finally, Ludlow and Segal note that deWnites and indeWnites seem to behave diVerently in certain environments. Since ‘the F ’ and ‘an F ’ are semantically equivalent on their analysis, they are forced to oVer a pragmatic explanation for these diVerences. For example, they give a pragmatic explanation for the so-called deWniteness eVect noticed by Milsark (1974), and they give a similarly pragmatic explanation for the diVering behavior of ‘the F ’ and ‘an F ’ in predicative environments. The chapters in this section all make novel proposals as to the semantics of deWnite and/or indeWnite descriptions. These chapters add to a growing body of literature challenging the Russellian analysis of descriptions. This topic is likely to spark debate for some time to come. 7 The contrast between Heim and Ludlow and Segal is not necessarily that great, if one agrees with Horn (1996: 310) that conventional implicatures are just (a subset of ) pragmatic presuppositions. Lewis (1983b) introduced the notion of presupposition accommodation. For recent discussions of the notions of pragmatic presupposition, common ground, and presupposition accommodation, see von Fintel (2001b), Simons (forthcoming), and Stalnaker (2002). Besides the notion of givenness, people have appealed to notions such as identiWability and accessibility to explain the contrast between deWniteness and indeWniteness. Abbott (forthcoming) has a thorough discussion of the many suggestions that have been made for capturing this contrast.

10 Referring Descriptions R. M. Sainsbury

1. Introduction One of Russell’s arguments for his theory of descriptions involves contrasting them with names. According to Russell, names, conceived as a logician should conceive them, are expressions which, if co-referring, are everywhere substitutable salva veritate, are not involved in semantically signiWcant distinctions of scope, require acquaintance with their referent in order to be understood, and are such that to suppose that they have no bearer would be to suppose that they have no meaning (i.e. cannot be used in a signiWcant sentence). This conception of names serves as a point of convergence for many distinctive features of Russell’s logic, metaphysics, epistemology, and philosophy of mind. For example, just as acquaintance needs to be a relation immune to the tactics of Descartes’s demon, so a name’s relation to its referent must be somehow guaranteed; just as, in acquaintance, the mind makes ‘direct’ contact with the world, so a name, in referring to an object, somehow makes ‘direct’ contact with it; just as a name is semantically simple, so its referent, Russell was tempted to think, must be ontologically simple (or ‘relatively simple’). The purpose of this chapter is to see what happens to deWnite descriptions if we detach ourselves from Russell’s conception of reference. I argue that, even if we do not wish to depart from Russellian truth conditions for any sentence, we may still treat deWnite descriptions as referring expressions in giving a semantic account of our language. By implication, we thereby come to see that Russell’s theory of descriptions, as normally understood, involves two separable components: there is the claim about the truth conditions of sentences containing deWnite descriptions, and this claim is not challenged in this chapter; and there is a claim about the semantics of deWnite descriptions themselves, to the eVect that they are not referring expressions but quantiWer Many thanks to Jennifer Hornsby, Keith Hossack, Michael Martin, Stephen Neale, David Papineau, Charles Travis, and an anonymous reader for Oxford University Press for comments on an earlier version of this chapter. I would also like to thank the Leverhulme Trust for a Senior Research Fellowship, during the tenure of which this chapter was extensively revised. Finally, special thanks to the editors of the volume, who kindly allowed me to make changes (notably the addition of the Postscript) at a very late stage.

370

IV. Deﬁnites and Indeﬁnites in Semantic Theory

phrases. It is this second claim that this chapter addresses, showing that there is a coherent alternative view consistent with the Wrst claim. The two Russellian claims converge in the framework of distinctively Russellian assumptions, notably those about what is involved in the mind’s contact with the world. Remove these assumptions, and it is hard to resist the thought that deWnite descriptions are complex referring expressions. In the next section, I elaborate this thought. In section 3 I consider objections based upon the idea that the relation I call reference does not deserve to be so called. In section 4 I take into account the distinction between referential and attributive descriptions, and in section 5 I consider whether the referential approach is superior to the Russellian one, especially in the light of the attempt to devise a complete theory of ‘the’, one which includes the cases in which it attaches to a plural or mass predicate.

2. The Proposal Suppose that a good start toward an account of understanding a sentence is to say that it involves knowing the sentence’s truth conditions, and that this knowledge can be embodied in a Davidson-style semantic theory for a language. In this context, what would be the right thing to say about what is involved in understanding a referring expression? To be sure, understanding it requires that one know its contribution to truth conditions; but what is this? The currently conventional answer is that one must know to which object it refers, where this is glossed so that it entails that there is such an object and that one knows, concerning it, that it is referred to by the expression. This gloss is not compulsory, and leads to a disanalogy between the roles of reference and truth in understanding. It would be mad to think that understanding a sentence involves knowing which truth-value it has, for this knowledge is evidently neither necessary nor suYcient for understanding; rather, what must be known are the conditions under which the sentence would be true. The analogous claim for reference would be that understanding a referring expression involves knowing what it would be for the expression to refer, that is, knowing how an object would have to be for the expression to refer to it. Preserving the analogy has the advantage of making empty names unproblematic. Burge (1974) has developed a truth theory which meets the desiderata of the previous paragraph. An approximate example of the envisaged axioms for simple singular terms is: (1)

For all x (‘Hesperus’ refers to x iV x ¼ Hesperus).

As ‘Hesperus’ occurs in quotation marks, the name belongs to the object language; as it is used alongside the identity sign, it belongs to the meta-language. Although I have assumed that the meta-language will contain all the terms the object language does,

Sainsbury, Referring Descriptions 371 this is not essential. The general recipe for writing axioms would say something like: a term is to be associated with a reference condition of the form ‘x ¼ t’, where ‘t’ is the meta-language translation of the term in question. The setting is negative free logic: every atom containing a nonreferring term is false, the quantiWer rules are appropriately adjusted, and the ‘law of identity’ is only ‘8x ¼ x’, rather than the truth of every singular sentence of the form ‘t ¼ t’.1 The reference condition and the logic of identity ensure that singular reference is an ‘at most’ relation: if ‘t’ refers to x and to y then x ¼ y.2 No axiom of this kind could do justice to a semantically complex expression. But the general idea can be implemented, for singular deWnite descriptions, along the following lines: (2)

For all predicates F in the singular, for all x (‘the’ _ F refers to x iV x ¼ the satisWer of F ).

If the meta-language does not itself contain complex referring expressions, a Russellian-sounding axiom could be used: (3)

For all predicates F in the singular, for all x (‘the’ uniquely satisWes F )

_

F refers to x iV x

though this would preclude homophonic theorems. To allow for cases in which F contains an anaphoric pronoun (like the ‘her’ in ‘Sally loves the man who kissed her’) the reference condition must be relativized to an assignment, though this will be ignored here. It is at least on the cards that plural reference (e.g. ‘the men’) can be treated in similar ways (see ‘Plural and Mass DeWnite Descriptions’ in 5.5 below). These truth conditions set context on one side, and so constitute a far from Wnished account.3 The importance of context will not discriminate between Russellian and referential accounts. One can classify context-dependence as linguistic (dependent on previous utterances) and non-linguistic (the remaining cases). Non-linguistic depend1 There are several diVerent logics called ‘free’, and one reason for which, I suspect, semantic theories in free logical frameworks are less common than they should be is that some free logics are quite unsuitable for semantics. The negative free logic presupposed in this chapter is a far cry from what Joseph Almog (this volume) thinks of as free logic: ‘Another response, imagined by various free logicians, is to extend the range of objectual values so as to make seemingly empty names come out as loaded names after all.’ The free logic of the present chapter does not assign values to empty referring expressions. 2 Plural reference is also an ‘at most’ relation. Using capital letters for plural variables, if a plural referring expression (like ‘Ben and Mary’ or ‘the apostles’) refers to X and to Y , then X ¼ Y . This assumes that a plural referring expression does not refer to some but not all of the things it refers to (‘Ben and Mary’ does not refer to Ben). This is semantically required, for a sentence whose predicate is ‘together lifted the boat’ should be true iV true of whatever the subject expression refers to. The apparent oddity is reduced by rephrasing in terms of ‘has X as its referent’. I assume a single use of the expression, singular or plural: the shape or sound ‘Aristotle’ can be used to refer to distinct things. 3 In common with Russellian treatments, they also ignore various arguably distinct uses of descriptions: ‘generic’ uses, exempliWed by ‘The whale is a mammal’ (cf. Strawson 1950), ‘functional’ uses, as in ‘The President is re-elected annually’, and predicative ones, as in ‘Napoleon was the greatest French general’ (cf. GraV 2001).

372

IV. Deﬁnites and Indeﬁnites in Semantic Theory

ence is often signiWcant in cases of ‘under-speciWcation’, when more than one thing satisWes the predicate in the description. One form of linguistic dependence is (or is analogous to) anaphora, for example, ‘A man opened a door for a girl. The girl thanked him.’ The well-known diYculties of doing justice to contextual phenomena appear no greater on the referential account than on the Russellian. The main rival way of dealing with some kinds of contextual variation treats deWnite descriptions as having a semantics which falls short of truth conditions, these being variously supplied by contextually controlled enrichment, and is thus inconsistent with both Russellian accounts and the present form of referential account.4 If the object language contains modal idioms, and the semantics represents these one-dimensionally by quantiWcation over worlds (with domains including trans-world individuals), the axioms become, respectively: (10 ) (20 ) (30 )

For all worlds, w, for all x (‘Hesperus’ refersw to x iV at w (x ¼ Hesperus)). For all worlds, w, for all x (‘the’ _ F refers w to x iV x ¼ the satisWerw of F ). For all worlds, w, for all x (‘the’ _ F refersw to x iV x uniquely satisWesw F ).

The rigidity of names (simple singular referring expressions) is not marked explicitly in the axioms, but will obtain if the meta-linguistic translations are rigid. If a metalinguistic term, t, meets the following condition, so does its object language homonym: for all worlds, w, w0 , and all things, x: if t refersw to x then t refersw0 to x. I assume that a term may referw to something which does not exist at w.5 This is not to assume that there can be names which actually have no bearer, for one could coherently insist, contrary to the claims of the present chapter, that any intelligible name must refer to something with respect to the actual world. As one would expect within a homophonic truth-theoretical approach, (20 ) does not tell us explicitly whether descriptions are or are not rigid designators: object language ones are iV meta-language ones are. Some descriptions are non-rigid if, as seems likely, there is a predicate F , worlds w and w0 , and distinct things x and y such that the following are true: ‘at w (the F is x)’ and ‘at w0 (the F is y)’. Given that ‘the’ can attach to an arbitrary predicate, (30 ) ensures that some descriptions are non-rigid if, as seems likely, some predicates have a single object as their extension at one world, and a distinct single object as their extension at some other world. So, along with other natural assumptions, some descriptions will refer rigidly and some non-rigidly. The rigid reference will be, in Kripke’s terms, de facto: the description ‘happens to use a predicate ‘‘F ’’ that in each possible world is true of one and the same unique object’ (1980: 21 n. 21). It is not possible for a semantically complex 4 This alternative, which has been endorsed by e.g. Stalnaker (1970), Recanati (1993), Bezuidenhout (1997), certainly merits consideration, but it raises too many additional issues to be considered here. 5 Evidence for and against the opinion that this was what Kripke originally had in mind is assembled by Kaplan (1989a: 569–70).

Sainsbury, Referring Descriptions 373 expression to be literally de jure rigid, that is one for which ‘the reference . . . is stipulated to be a single object’ (1980: 21 n. 21), since the semantic properties of a complex must emerge from stipulations concerning their semantically signiWcant simple parts, and cannot be stipulated directly.6 However, one can if one wishes fashion a ‘the’ whose semantic axiom ensures that the referring expression it helps to form is a rigid designator (see ‘theDR ’ below, s. 4); this designator would meet the condition for de jure rigidity, as naturally extended to complex expressions.7 The conclusion agrees with intuition: the most natural way to account for the truth of such modal truths as ‘Someone else might have been the inventor of the zip’ is to suppose that ‘the inventor of the zip’ refers to diVerent things with respect to diVerent worlds, depending upon who invented the zip in the world. An atom is a simple (n-ary) predicate concatenated with n names. On a traditional view, all complexity other than that found in atoms comes from the essentially sentence-forming logical constants. On a more realistic view of natural language, there are other ways in which semantic complexity is achieved. For predicates, these include predicate modiWers (as in ‘bright red’). For subject expressions, these include complex subjects like ‘Ben and Mary’. Given the existence of collective predicates, like ‘together lifted the boat’, it is not plausible that this is an example of essentially sentential complexity, for ‘Ben and Mary together lifted the boat’ is not equivalent to ‘Ben together lifted the boat and Mary together lifted the boat’. Even before considering deWnite descriptions, therefore, the atomic sentence would not represent the basic mode of sentence-formation in natural languages. Rather, we need the notion of a subject–predicate (S–P) sentence. This consists in the concatenation of n subject expressions with an n-ary predicate. There is no ban on the presence of logical constants within either the subject or the predicate. Thus ‘Ben and Mary together lifted the boat’ is subject–predicate despite arguably having a complex predicate (‘together lifted’) and two complex subject expressions (‘Ben and Mary’, ‘the boat’), one of which even contains an expression generally counted as a logical constant (‘and’). A language thought of as built up from atomic sentences diVers sharply from one thought of in terms of S–P sentences. If the language has a Wnite number of simple singular terms and simple predicates, it has a Wnite number of atomic sentences, but it does not follow that it has a Wnite number of S–P sentences. For example, if ‘very’ iterates, there is no limit to this list of predicates: ‘bright red’, ‘very bright red’, ‘very, very bright red’ . . . Each is Wt to form an S–P sentence, by combining with a singular term (and the copula). The unbounded character of S–P sentences is also ensured if ‘the’ can combine with arbitrary predicates to form subject expressions; and if ‘and’ can conjoin names without limit to form complex subject expressions. 6 One could deWne a semantically simple expression as one whose semantic properties can be stipulated. This reduces Kripke’s de facto/de jure distinction, taken precisely as he formulated it, to that between the semantically simple and the semantically complex. However, one can reformulate the distinction to avoid this collapse. 7 It is not semantic complexity as such which induces non-rigidity: ‘Ben and Mary’ rigidly designates Ben and Mary, if each of ‘Ben’ and ‘Mary’ are rigid designators (cf. Hossack 2000).

374

IV. Deﬁnites and Indeﬁnites in Semantic Theory

The notion of subject–predicate sentence I shall work with does not count indeWnite descriptions (‘a(n) F ’) and explicit ‘quantiWer phrases’ as themselves subject-expressions. This is not based on (at least simple-minded) grammatical congruity tests, but rather on inferential properties; I pass over the details. Non-modal truth theory for a language with referring descriptions would need a composition axiom which exploits the notion of S–P. The unary singular case will go along these lines: (4)

An S–P sentence is true iV for some x, S refers to x and x satisWes P.8

Achieving homophonic T-sentences will require a logic which ensures the validity of such inferences as: (5)

From: S–P is true iV for some x, x ¼ t and x is G infer: S–P is true iV t is G.

Extending to the non-unary case is straightforward. Descriptions require an inference in which something like ‘x ¼ the satisWer of ‘‘F ’’ ’ is replaced by ‘x ¼ the F ’, preparatory to using (5). In the non-modal case, ‘for all y, Fy iV Gy’ is suYcient for for all x, x ¼ the F iV x ¼ the G;9 and this in turn is enough (via substitution of equivalents) for the inference just mentioned.

3. Reasons for Thinking This is Not Really Reference On the present view, reference is one essential relation in the composition axiom. It is distinguished from the other relation, satisfaction, by the fact that it introduces an atmost condition: the axioms plus the background logic ensure that for each singular term there is at most one satisWer of its reference condition. In this section, I consider some forms of the objection that reference is more than this.

Referring is Something Only Simple Expressions Can Do This is certainly a Russellian view, and very likely it runs deep in his thought. Justifying it from a neutral starting point (if such a thing exists) would be another matter. It 8 Cf. Larson and Segal (1995: 119). An earlier source (by more than 600 years) of this truth condition for S– P sentences is William of Ockham, Summa Logica (II. 2): ‘it is suYcient and necessary that the subject and predicate supposit for the same thing’ (Ockham 1998: 86). 9 It is a nice feature of Burge’s system that although e.g. ‘the unicorn ¼ the golden mountain’ is false, it is true that, for all x, x ¼ the unicorn iV x ¼ the golden mountain.

Sainsbury, Referring Descriptions 375 confronts the problem of apparent complex referring expressions other than descriptions, like ‘Ben and Mary’, which appear not to be eliminable in every context. And it makes it hard to Wnd a unitary notion of reference common to semantic reference and speaker reference. One motivation for the position is that the reference of an expression has to be something that can be arbitrarily assigned. The motivation is itself threatened by its unintuitive upshots: the denial both of complex referring expressions other than descriptions and of unity between semantic and speaker reference. (A speaker’s choice of words or gestures to refer may be far from arbitrary.)

It is Necessary and A Priori that a Referring Expression Refers This thought was central to Russell’s conception of a name: by deWnition, names name. One can know a truth condition without knowing whether or not the condition is satisWed; and in many cases it will not be satisWed. If the analogy between truth and reference in relation to understanding takes this as its starting point, room is thereby made for the possibility, metaphysical and epistemic, of an expression associated with a reference condition failing to meet that condition, and so being a referring expression which fails to refer. On the view presented here, referring expressions are ones which, in their semantics, purport to refer; whether or not they succeed is another matter, falling in the domain of some science other than semantics. The view makes ‘names name’ analogous to ‘police police’, a speciWcation of what would happen if all went well, rather than of what always in fact happens. I do not intend to try to provide (and do not know how to provide) a knock-down argument in favour of the free logical conception of reference as opposed to the traditional Russellian one.10 It seems to me that once people are satisWed that the former is coherent, its advantages for natural language semantics will be obvious. The point of the present chapter is to argue that the application of the free logical notion to deWnite descriptions is indeed coherent. Russell himself probably wanted it to be a priori, relative to linguistic understanding, whether an expression is a referring expression or not, and so whether or not it refers. This gave him a decisive reason for not counting ordinary names as referring expressions. Modern-day theorists inXuenced by Russell in their treatment of names, like Evans and McDowell, do not suppose that this categorization is a priori (else it would be a priori whether or not Homer exists). A semantic theorist nonetheless needs to make the categorization, and so, if he belongs to the classical mould, must be involved in empirical work, including astronomy (to see if ‘Vulcan’ is a referring expression) and the study of ancient societies (to see if ‘Homer’ is a referring expression). The free logical semanticist has an easier task: one who understands a language is

10 Some relevant considerations are adduced by Sainsbury (2002: essays IX and XII).

376

IV. Deﬁnites and Indeﬁnites in Semantic Theory

thereby equipped to know, without further investigation, which are its referring expressions, though not which of them refer. A connected objection is that reference failure engenders lack of truth conditions. One motivation is that referential success is required for the standard composition axiom to deliver a truth condition, but one needs also to motivate choosing a composition axiom which has this eVect, rather than choosing a free logical one, like (4), which does not.

Understanding a Referring Expression Involves De Re Knowledge Russell thought that understanding a referring expression (i.e. a logically proper name) involves being acquainted with its referent, and more recent theorists in this tradition have held something structurally similar, though without commitment to a Russellian conception of acquaintance: understanding a referring expression involves knowledge (de re) of its referent. On a free logical approach, this cannot be so in general, since not all referring expressions have referents. But can it be so for those expressions which do have referents? On standard views, knowledge of an axiom for a referring expression whose referent is o will be held to be ‘de re’ in one or more of the following respects: (i) O is such that one who knows the axiom thereby knows something about it. (ii) The inference from knowledge of the axiom to the existence of an object, o, about which the knower has knowledge, is formally valid. (iii) The axiom can be known only by one who is en rapport with o. Here ‘en rapport’ marks a position for some unspeciWed but relatively hard to satisfy relation between a knower and an object. Russell’s view entails that all three conditions are met, and the ‘en rapport’ relation is acquaintance. Opinions about (ii) and (iii) can reasonably vary even among those otherwise sympathetic to a Russellian conception of reference,11 but they should all agree on (i). The free logical approach can accept (i) for axioms relating to nonempty referring expressions. Hesperus is such that one who knows that for all x (‘Hesperus’ refers to x iV x is Hesperus) thereby knows something about it; the capital of England is such that one who knows that for all x (‘the capital of England’ refers to x iV x is the satisWer of ‘capital of England’) thereby knows something about the satisWer of ‘capital of England’, that is, about it (the capital of England). ‘Thereby’ does not mean that there is a general inference to this conclusion, regardless of what referring expression appears in the axiom known, but merely that this item of knowledge is or involves knowledge about an object. Russellian and free logical theorists of reference can therefore agree that in some sense (that pointed to by (i)) understanding a semantic axiom for a nonempty referring expression involves de re knowledge of its referent. 11 Thus Kaplan (1989b: 605) explicitly denies (iii): ‘acquisition of a name does not in general put us en rapport . . . with the referent.’

Sainsbury, Referring Descriptions 377

Referring Expressions are Scopeless Since deWnite descriptions are not scopeless, the objector infers that they are not referring expressions. Although it is widely held that referring expressions are scopeless, it is harder to Wnd justiWcations, and such as I can construct depend upon other theses, like ‘It is necessary and a priori that a referring expression refers’ or ‘Referring expressions are rigid designators’ (see below). The proponent of the free logical element of the present proposal will suggest that attributing scope distinctions even to names has some theoretical purpose, whence the existence of scope distinctions for deWnite descriptions is not enough to show them not to be referring expressions.12 Here are two examples of phenomena relating to names which can be addressed in terms of scope. (1)

Some people hear ‘it is necessarily true that Socrates is human’ as false, while hearing ‘Socrates has the property of being necessarily human’ as true. This is prima-facie evidence that names have scope (in the dialects of these people).

An objection is that this is not a scope distinction but rather one between metalanguage modality (masquerading as wide scope) and object-language modality (masquerading as narrow scope). This seems intrinsically dubious, since it is hard to hear the Wrst as something that could be understood quite independently of whether ‘Socrates is human’ is understood. As a scope distinction, it Wts into a theory which seems quite natural. Let propositions be 0-place properties which exist at every world, let sentences be expressions which refer to propositions, and let the application of ‘necessarily F ’ to an object be true iV the object has the property expressed by ‘F ’ at every world at which it exists. ‘Necessarily, Socrates is human’ is then not true, since there is a world at which Socrates does not exist and at which the proposition that he is human exists and is not true; but ‘Socrates is (necessarily human)’ is true, since Socrates possesses the property of being human at every world at which he exists. (2)

One who asserts, as non-Wctional, that Pegasus Xies, has made some kind of mistake. One simple option is to suppose that, in the service of exposing the myth for what it is, we can truly deny what he says. But we do not want to share an opinion with those who, also in the grip of the myth, assert that Pegasus does not Xy (but gallops). So, as Ockham implied (English translation 1998: 123), ‘it is not the case that Pegasus Xies’ is true, and can be used to deny the original assertion, whereas ‘Pegasus doesn’t Xy’ (understood as ‘Pegasus is a non-Xyer’) is false.

In this case it may again be suspected that rather than a scope distinction we have a distinction between meta-language and object language use; and meta-language 12 The formal language of free logic makes room for semantically signiWcant scope diVerences of singular terms and operators. The semantic theorist who uses a free logical framework is not obliged to suppose that such scope distinctions are exploited by any sentences in natural language.

378

IV. Deﬁnites and Indeﬁnites in Semantic Theory

negation is, arguably, quite common (see Horn 1989). But, once again, it is hard to hear ‘It is not the case that Pegasus Xies’ as something which can be understood independently of whether or not ‘Pegasus Xies’ is understood. Free logical (i.e. Ockhamist) truth conditions naturally embed this as a scope distinction: a subject–predicate sentence is true iV the subject refers to something which has the property expressed by the predicate; negation toggles truth and falsity; ‘does not Xy’ is the predicate in ‘Pegasus does not Xy’ but ‘It is not the case that Pegasus Xies’ is the negation of the false ‘Pegasus Xies’. The phenomena are complex, and these examples cannot be considered decisive. They show that, although it may be hard conclusively to establish that simple referring expressions can be involved in semantically signiWcant scope distinctions, it will also be hard to establish that this is not so. It may not be essential to the coherence of the position proposed here that simple referring expressions are scope sensitive. It is not obviously incoherent to hold that simple ones are scopeless but (some or all) complex ones scope sensitive.

Referring Expressions are Rigid Designators Names and descriptions are governed by diVerent substitution conditions in modal contexts (made vivid in slingshot arguments: cf. Neale 1995). Whereas no paradox arises from the supposition that co-referring names are substitutable salva veritate in modal contexts, a paradox does arise from supposing the same about deWnite descriptions. A category of expressions all of which are to be treated by the same semantic relation may diVer in logical/semantic properties. Even if all predicates are to be treated by means of satisfaction, they diVer in their logical properties (e.g. some are ‘rigid’—have the same extension at every world—and some are not). One might see a range of modal expressions as variants on a single semantic theme (diVering, perhaps, in whether they introduce all worlds, or just the logically possible, or just the physically possible, or just the epistemically possible) and so as belonging to a single semantic kind, even if modal equivalence of the diVerent kinds generates diVerent strengths of substitution condition. So we cannot move from the premise (which I do not dispute) that names and deWnite descriptions are governed by diVerent substitution conditions to the conclusion that they are not of a single semantic kind. The contrary view about descriptions can be Wtted into systematic theory. What substitution of t1 by t2 in a context dominated by an operator, v, requires is the truth of v(t1 ¼ t2 ). It so happens that in the case of names, but not descriptions, t1 ¼ t2 entails its necessitation, so names seem eVortlessly to meet the demands for substitution in modal contexts. However, co-referring names are not generally substitutable in hyperintensional contexts, since the relevant v-dominated identity (e.g. ‘John believes that Hesperus is Phosphorus’) may not be true. If substitution inferences were set out in full, with the appropriately dominated identities among the premises, the contrast

Sainsbury, Referring Descriptions 379 between names and descriptions would no longer be at all striking. They would share conditions for successful substitution, but deWnite descriptions would, more often than names, fail to meet them. Russell’s conception of reference and identity dispensed entirely with v-qualiWcations, and the upshot is, as he himself aYrmed, that there are no (or almost no) logically proper names in human languages.13 We get a conception of co-reference that is useful for natural language semantics only by qualifying Russell’s pure vision. Once we start to qualify, however, the idea that genuine co-reference can fail to sustain substitution in hyperintensional contexts but must sustain it in modal contexts becomes hard to motivate. By contrast, the view that the substitution inference can be counted upon only if the strength of the identity matches the strength of the context in which substitution is to occur seems highly natural.14

4. Referential Versus Attributive Everyone agrees that there is some diVerence between ‘referential’ and ‘attributive’ uses of deWnite descriptions, and this raises the question whether the view that deWnite descriptions are referring expressions can allow for this diVerence. There are various conceptions of what the referential/attributive diVerence consists in, and I start by showing how the present proposal can accommodate two versions of the view that there is a semantic ambiguity located in the word ‘the’. Suppose the single word ‘the’ is ambiguous between an attributive version, ‘theatt ’ and a referential version, ‘theref ’; and suppose these have the following impact upon the truth conditions of whole sentences containing descriptions: (Att) (Ref )

‘Theatt F is G ’ is true iV something is uniquely F and it is G. If there is a unique x such that u (on a given occasion) intends his utterance to concern x and thinks he can achieve appreciation of this intention by using ‘the F ’, ‘Theref F is G ’ is true as uttered by u (on that occasion) iV x is G.

The current proposal provides ‘the’ with truth conditions appropriate to ‘theatt ’, but ignores ‘theref ’. (Regarding ‘the’ as a referring expression should not be identiWed with assigning it a contribution to truth conditions which would make it Wt to form ‘referential’ descriptions in anything like Donnellan’s sense.) Referential descriptions would require a further axiom like: 13 Not all direct reference theorists would agree with this: cf. Salmon (1986). 14 Added note: when I Wrst wrote this paper I thought that suitably strengthened identities would validate substitution in all contexts, including hyperintensional ones: in the presence of ‘John knows that Hesperus ¼ Phosphorus’, ‘John knows that Hesperus is visible’ would entail ‘John knows that Phosphorus is visible’. This no longer appears to me to be in general correct. The Postscript suggests a better view.

380

IV. Deﬁnites and Indeﬁnites in Semantic Theory (2ref )

For all x (‘theref ’ _ F refers to x as uttered by u iV x ¼ the thing which u intends his utterance to concern and such that u thinks he can achieve appreciation of this intention by using ‘theref F ’).15

Nothing in the present proposal prevents the addition of such an axiom, which would recognize ‘the’ as ambiguous in the most straightforward way. It would be somewhat misleading to use the terms ‘referential’ and ‘attributive’ to mark the two senses, for in both ‘the’ is treated as a term which takes a predicate to form a referring expression. It is just that the diVerent ‘the’s generate diVerent reference conditions. Applying this to one version of Donnellan’s original distinction, the sense in which an utterance of ‘The man drinking martini is drunk’ is true, even though no one in the context is drinking martini, can be represented as ‘Theref man drinking martini is drunk’; the sense in which an utterance of this sentence is false can be represented as ‘Theatt man drinking martini is drunk’. Both deWnite descriptions count as referential by the standard adopted here. The semantic ambiguity just considered is not the only one that might be represented under the heading ‘referential–attributive’. Another possible ambiguity is between descriptions which, while in some cases designating rigidly, do so merely ‘de facto’ (as a result of the rigidity of their predicates) and descriptions which de jure rigidly designate, as a result of stipulations concerning the word ‘the’. We have already seen how to handle the former. The following modiWcation of (2) will serve for de jure rigidly designating ones: (2DR )

For all x (‘theDR ’ _ F refers to x iV x ¼ the actual satisWer of F ).

Here ‘actual’ is absolute, taking the evaluation just to our actual world. Nothing in the present proposal prevents the addition of such an axiom, and the upshot is a semantic theory which recognizes at least two types of referring descriptions. One of these might be used to represent two readings Stalnaker (1970: 42) discerns in ‘The man in the purple turtleneck shirt might have been someone else’. A true one could be represented as ‘Possibly, theDR man in the purple turtleneck shirt is not theatt man in the purple turtleneck shirt’, and a false one as ‘Possibly, theDR man in the purple turtleneck shirt is not theDR man in the purple turtleneck shirt’. Not all accounts of the referential–attributive distinction see it as a semantic ambiguity in ‘the’. One cannot dismiss such alternative accounts as thereby irrelevant to present concerns, for these accounts may also have implications for the semantics of ‘the’. This is so both for standard Grice-style accounts of the distinction as pragmatic and non-truth-conditional, based on unambiguously Russellian truth conditions, and for accounts of the distinction as pragmatic and truth conditional, based on a subtruth-conditional semantics for the relevant sentences.16 Theorists of the former kind 15 It is no accident, I believe, that attempts to provide this kind of reference condition fail to do justice to the possible semantic complexity of the predicate (e.g. ‘man drinking martini’). 16 References are in n. 4, above.

Sainsbury, Referring Descriptions 381 may see their position as inconsistent with that proposed here, for they oVer the pragmatics in defense of Russell’s theory of descriptions, which indeed excludes descriptions from the category of referring expressions. However, these theorists generally rely not upon this claim, but upon the Russellian truth conditions for sentences containing deWnite descriptions. If a central point of this chapter is granted, namely that one can combine the view that deWnite descriptions are referring expressions with the Russellian view of the truth conditions of sentences containing them, there is room for Grice-style pragmatic accounts of referential–attributive which agree that descriptions are referring expressions. The view that the semantics of deWnite description sentences are ‘under-determined’, that is, are sub-truth-conditional, and require pragmatic enrichment to generate either referential or attributive truth conditions, is inconsistent with the view proposed here. Once context is allowed to enter the picture, there is some slack in what constitutes semantic axioms like (2). Perhaps the axiom is just what is on the page, in which case it is a mere template, requiring interpreters to enrich the predicate F in various contextually prompted ways. Or perhaps there are as many axioms as there are such enrichments, so that each axiom delivers something determinate. These possibilities, however, do not match the view under discussion, for they do not allow for variation in truth conditions along a dimension describable as referential–attributive. On the present view, any truth-conditional variation involves distinct words ‘the’. This view accordingly needs to Wnd fault with under-determination theories, though that will not be attempted here.

5. Logical Form and Truth Conditions The referential proposal agrees with Russell about truth conditions (setting aside possible reWnements adopted in the light of alleged referential–attributive distinctions).17 How can the residual disagreement about logical form matter?

Russell and Davidson on Logical Form Russell thought of the logical form of a sentence, or rather of a judgment, as made plain by its translation into the language of Principia Mathematica. On this view, the proposal to treat descriptions as complex referring expressions is not a logical form proposal, since the language of PM recognizes no such category. In order to have an intelligible debate, I shall adopt a more Davidsonian conception of logical form (cf. Davidson 1980a, b). It was unavailable to Russell but it does justice to a fair amount of 17 Cf. Burge (1974: 314): ‘Whereas we agree with Russell about truth conditions, we disagree with him about logical form.’

382

IV. Deﬁnites and Indeﬁnites in Semantic Theory

his use of the idea of logical form. It is motivated primarily by considerations of semantic theory, taken as a theory which is constrained by the goal of stating something knowledge of which would be enough for understanding the language. On the Davidsonian view, the logical form of a sentence of a natural language is the sentence itself, if it can feature appropriately in a theorem of an interpretive truth theory for the language as a whole; otherwise it is the closest approximation to the sentence which meets this condition of accessibility to semantic theorizing.18 A positive degree of closeness requires identity of truth conditions, and one candidate is closer to the original than another if it involves fewer modiWcations. (This is a vague notion which no one has troubled to make precise.) Within the Davidsonian framework, logical forms (in indexical-free languages) will receive homophonic theorems in the truth theory. This gives another test for divergence of grammatical and logical form, for divergence implies that the oYcial speciWcation of the original sentence’s truth conditions will be non-homophonic. Whereas there is no Wrm distinction in Russell’s approach between philosophical analysis and the Wnding of logical form, Davidson keeps these two activities distinct. It could be that a word for a concept which features homophonically in semantic theory deserves substantial philosophical analysis, in which our standard way of introducing the concept is replaced by something whose very unfamiliarity is illuminating. I suggest that just this is the case for ‘the’. I will argue that we can gain philosophical illumination by analyzing it in the Russellian fashion, but that from the standpoint of semantic theory it should Wgure homophonically, and this requires a referential approach. According to Russell’s theory of deWnite descriptions, their grammatical form is distinct from their logical form. In Davidson’s framework, this means that a sentence containing a description is subject to some non-trivial modiWcation before becoming accessible to semantic theory, and that the proper speciWcation of its truth conditions will be non-homophonic. In particular, it may well be that ‘the phrase [sc. one of the form ‘the F’] per se has no meaning, because in any proposition in which it occurs the proposition, fully expressed, does not contain the phrase, which has been broken up’ (Russell 1905: 488). Russell also speaks of his theory as a reduction (e.g. 1905: 482). One way to express the position is to say that the Russellian logical form of ‘the F is G ’ will not contain ‘the’. There are various ways to implement this, but for deWniteness let us take the following: (6)

The logical form of sentences of the grammatical form ‘the F is G ’ is ‘there is something, x, such that x is F , and anything that is F is identical to x; and x is G ’.

One contrast between Russell’s view and the referential proposal is now stark: the referential proposal sees a sentence of the form ‘the F is G ’ as almost its logical form 18 This skates over some exegetical diYculties, relating especially to Davidson (1980b).

Sainsbury, Referring Descriptions 383 (perhaps bar Wddly modiWcations to make explicit variables and scope). In starkest formulation: (7)

The logical form of sentences of the grammatical form ‘the F is G ’ is ‘the F is G ’.

We now need to ask how we should choose between (6) and (7). The contrasts in Table 1 make vivid the fact that the pre-theoretical view is that descriptions are referring expressions, for intuitions favor the right-hand claims in every case; we need to be dug out of this view by Russellian considerations. If we think that referring expressions must refer, and that all descriptions should be treated by the same kind of semantic axiom, we progress rapidly to Russell’s position. The free logical context removes any logical motivation for the view that singular referring expressions must refer. Reference becomes like Russellian denotation: a referring expression counts as such in virtue of its ‘form’ (and not in virtue of succeeding in referring). Russell’s own conception of reference is not one which can be taken seriously by those who believe that genuine names with the same bearer are not everywhere substitutable salva veritate; and it will appear unattractive to those not moved by Russell’s hope that thinking about the world involves a demon-proof relation. The source of the pre-theoretical view that deWnite descriptions are referring expressions is probably a guess that we apply the same processing strategy to S–P sentences regardless of whether they have names or deWnite descriptions in subject position. The guess could be tested empirically. If it is correct, then from the point of view of an account of understanding, we do well to treat deWnite descriptions as referring expressions (provided that by doing so we can reach the right truth conditions). This is consistent with thinking that Russell’s account can throw some light on the concept associated with ‘the’. The consistency is like that between the claim that the right semantic axiom for ‘bachelor’ says that this word is satisWed by all and only bachelors, while a possibly instructive analysis says that a bachelor is an unmarried man.

‘The’ as a Determiner Some of the ideas in Russell’s theory of descriptions can be implemented with a logical form closer to surface form (the locus classicus is Neale 1990). ‘The’ can be regarded as a determiner which attaches to a predicate to form a quantiWer. Neale says that this is not ‘an alternative to the Theory of Descriptions’ but the theory itself stated in a more ‘congenial’ way (Neale 1990: 45). Neale in eVect takes the truth conditions for whole sentences as the essential characteristic of Russell’s theory of descriptions, independently of the precise logical form which is supposed to generate the sentence-sized truth conditions. A version of Russell’s theory in which ‘the’ is treated as a determiner enables logical form to approach much more closely to grammatical form. On this proposal, the logical form of sentences whose grammatical form is ‘the F is G ’ is something like: the

384

IV. Deﬁnites and Indeﬁnites in Semantic Theory

Table 1. Comparison between Russell’s theory and the referential theory Russell’s Theory of Descriptions

Referential Descriptions

Description sentences are existential generalizations.

Description sentences are S–P.

DeWnite descriptions are quantiWer expressions, like ‘a(n) F ’, ‘no F ’ etc.

DeWnite descriptions are referring expressions, like names (‘London’) and (simple and complex) demonstrative expressions (‘this’, ‘that F ’).

Inference from ‘The F is G ’to ‘something is G ’ is conjunction elimination.

Inference from ‘The F is G ’to ‘something is G ’ is existential quantiWer introduction.a

‘The F is the G ’ is never an identity sentence.b

‘The F is the G ’ may be an identity sentence.

There is no genuine inference from ‘Napoleon is the greatest general’to ‘The greatest general is Napoleon’, for both will be formalized alike.

There may be a genuine, though valid, inference from ‘N ¼ the F ’ to ‘The F ¼ N ’. (The premise needs to be a genuine identity, whereas such descriptions are more often heard as predicative.)

The overall logical form of a pair like ‘The F is G ’ and ‘The F is not G ’matches that of the pair ‘Some cows eat grass’, ‘Some cows do not eat grass’.

The overall logical form of a pair like ‘The F is G ’ and ‘The F is not G ’matches that of the pair ‘John is hungry’, ‘John is not hungry’.c

a Free logic must use a restricted version of existential quantiWer introduction (for example, it cannot accept the validity of the inference from ‘not F a’to ‘something is not F ’). Since negative free logic requires there to be an Sif an S–Psentence is true, it can accept that there is no counter-example to the inference from ‘the F is G ’to ‘something is G ’. b A theorem of PM ( 14.272) shows that such sentences have the inferential powers of identity sentences within extensional contexts. See Russell and Whitehead (1910–13). c Negative free logic leaves room for the possibility that a referring expression (whether name or deWnite description) should take wide or narrow scope relative to negation. It is not obvious that this formal possibility should be exploited in providing semantics for natural languages: perhaps negation always takes wider scope than referring expressions.

x that is F is G. This gives us three views to compare: what I call Russell’s, in which the descriptive phrase is ‘broken up’ at the level of logical form, and two views which retain ‘the’ in the meta-language: the determiner view, and the referential view, both of which oVer logical forms close to, or even identical with, grammatical forms. This chapter largely brackets the determiner view. The view agrees with the referential one on sentence-sized truth conditions, but disagrees with it about the mechanisms that generate these: one view Wngers quantiWcation, the other reference; one view is set within classical logic, the other within free logic. Whether anything substantive hangs on these diVerences is not obvious. I will take it for granted that a proper account of ‘the’ will account for its uses as it attaches to plural as well as singular predicates, and

Sainsbury, Referring Descriptions 385 also as it attaches to mass terms (‘the gold in Zurich’), and that such an account will deal with collective as well as distributive uses of expressions in the remainder of the sentence. A hypothesis, which I here make no attempt to substantiate, is that the determiner view faces diYculties in achieving this level of generality.19 As I now go on to show, the same does not hold either for the Russellian or the referential views.

Plural and Mass DeWnite Descriptions ‘The’ can attach to mass terms (as in ‘the gold in Zurich’) and to plural predicates (as in ‘the people in Auckland’) (cf. Sharvy 1980). Intuitively, these descriptions are perfectly in order, despite the fact that there are many satisWers of the predicates ‘gold in Zurich’ and ‘people in Auckland’. Nothing in ‘the’ has any special connection with any of singularity, plurality, or masshood, and a full account of the word should explain this neutrality: it should be itself neutral and uniform, allowing the variations in question to emerge from the predicate with which ‘the’ couples. One who holds that singular deWnite descriptions are singular referring expressions is likely to hold that plural deWnite descriptions are plural referring expressions. We need to add plural variables, ones which stand in the kind of position Wt to be occupied by plural referring expressions (e.g. ‘Ben and Mary’). The proposed singular semantic axiom was: (2)

For all x (‘the’ _ F refers to x iV x ¼ the satisWer of F ).

Its plural correlate, using ‘X ’ as a plural variable, is: (2P)

For all X (‘the’ _ F refers to X iV X are the satisWers of F ).

An instance of (2P) is: ‘ ‘‘the apostles’’ refers to the apostles iV the apostles are the satisWers of ‘‘apostles’’ ’. Although Mathew and Mark are satisWers of ‘apostles’ they are not the satisWers of ‘apostles’, so their properties alone are not enough to verify a claim about the apostles. (2) and (2P) are essentially the same axiom, modulo number. The composition axiom would be extended to include a plural form. The proposal leads to a straightforward analogue of (7): (8)

The logical form of sentences of the grammatical form ‘the F are G ’ is ‘the F are G ’.

In specifying a minimal departure from Russell’s actual theory of descriptions which would allow it to speak to plural descriptions, the Wrst thing is to add plural variables.20 19 Neale is sympathetic to the idea that a Wnal account of ‘the’ should be general, and his own account extends naturally to plural distributive cases (1990: 46). He explicitly brackets collective readings (1990: 61). 20 I assume that there is genuine and irreducible plural reference and plural quantiWcation (persuaded by Hossack 2000). The historical Russell would no doubt have preferred to reduce this to singular reference and, in the Wrst instance, quantiWcation using classes. Hossack suggested that this would not aVect the contrast between the two conceptions of logical form, which is the theme of s. 5.

386

IV. Deﬁnites and Indeﬁnites in Semantic Theory

The extended Russellian can allow that there are plural referring expressions like ‘Ben and Mary’, while denying that plural descriptions are referring expressions. This is analogous to allowing that singular names are referring expressions while denying that the same is true of singular descriptions. In both cases, the conceded referring expressions suYce to introduce the notion of a corresponding kind of variable. A possible Russellian proposal which mirrors (6) is: (9)

The logical form of sentences of the grammatical form ‘the F are G ’ is ‘there are things, X , such that FX; and X are G ’.

This is incorrect, for it would make Mathew and Mark’s happiness suYcient for the truth of ‘the apostles are happy’. ‘X are G ’ is true of things which are G, so on this proposal ‘the apostles are happy’ is true if Mathew and Mark are happy apostles. Intuitively, we need some notion of maximality: for the apostles to be happy is for all of them to be so. But how should this notion be introduced? Not like this: (10)

The logical form of sentences of the grammatical form ‘the F are G ’ is ‘there are things, X , such that FX; and all X are G ’.

Suppose the department is composed of Ben, Mary, and Joshua. An application of (10) is: The logical form of ‘The members of the department together formed a triangle’ is ‘there are things, X , such that X are members of the department and all X together formed a triangle’. Even if Ben, Mary, and Joshua together formed a triangle, Ben and Mary, while members of the department, did not together form a triangle (how could they?). The Russellian needs further resources. He could allow the positions occupied by plural variables to be occupied also by expressions like ‘all the members of the department taken together’. This is the natural way, but in eVect it concedes a treatment of plural descriptions as referring expressions. Alternatively, he can introduce something other than quantiWer-variable devices. Richard Sharvy (1980) has suggested a single notion which he claims will do the trick. He calls it part of, and we could adapt his suggestion to the approach taken here as follows: (11)

For plural predicates, F , the logical form of sentences of the grammatical form ‘the F are G ’ is ‘there are things, X , such that FX and any Y that are F are part of X ; and X are G ’.

Let us call the italicized portion the ‘maximality condition’. ‘Part of ’ can also be read as ‘are among’. Sharvy’s suggestion avoids both counter-examples: Mathew and Mark’s happiness is not enough for the apostles to be happy, because Mathew and Mark by themselves do not meet the maximality condition and so will not constitute a verifying value for X in ‘there are things, X ’; that the members of the department together

Sainsbury, Referring Descriptions 387 formed a triangle does not require Ben and Mary to, for they do not meet the maximality condition. A nice feature of Sharvy’s account is that we can recover from (11) the condition for the singular description, since, using singular variables, for x to be among y is for x to be y. The condition of unity for singular and plural descriptions is met; the evidently non-homophonic character of the resulting semantic theory cannot (in the present dialectic) be regarded as a defect. Once again, the Russellian approach and the referential approach can both generate the right truth conditions, the former heterophonically, the latter homophonically. The same holds when we consider mass descriptions. Intuitively, ‘The gold in Zurich is worth more than a million dollars’ is true, even though there is gold in Zurich worth less. (11) as such does not address these cases, for they are singular, but Sharvy’s idea is easily adapted: (12)

For mass terms in the singular, F , the logical form of sentences of the grammatical form ‘the F is G ’ is ‘there is stuV, m, such that Fm and any m0 that is F is part of m; and m is G ’.

This tells us, correctly, that the gold relevant to the truth of this sentence is just all the gold in Zurich: gold of which any gold in Zurich is part. (The new style of variable, m, is one Wt to be replaced by a mass term.) On the referential approach, the logical form will be derived from reference axioms. The original (2) could become: (13)

For mass terms in the singular, F , for all m (‘the’ _ F refers to m iV m ¼ the satisWer of F ).

There are satisWers of ‘gold in Zurich’ which are not worth more than a million dollars, for example, the gold in Hans Ernst’s wedding ring. However, the gold in this ring is not identical to the satisWer of ‘gold in Zurich’, for the latter includes other gold. And although the ring perhaps satisWes ‘gold in Zurich’ (at any rate, it satisWes ‘is made of gold and is in Zurich’) it is not a permissible value for m, for such a value needs to be the sort of thing a mass term is true of. So the ring is not wrongly said by (13) to be something to which ‘the gold in Zurich’ refers. This does not show that (13) is adequate, but it removes some grounds for thinking it inadequate. A proper test for a theory of descriptions is that it should deal in a uniWed way not only with singular descriptions but with plural and mass descriptions. We have found no reason to think that the Russellian and referential approaches cannot both do this, in their diVerent ways. So an appropriate question would seem to be: which is correct? Or should we say that semantic facts are indeterminate, that there is no fact of the matter which theory is right? I think we do better to see the two approaches as addressed to diVerent issues. It seems to me likely that the referential approach stays closest to the way in which we actually process language. At least, there is some intuitive support for this, though the

388

IV. Deﬁnites and Indeﬁnites in Semantic Theory

issue deserves empirical investigation. Moreover, the referential approach, unlike the Russellian approach, leads to homophonic logical forms, and this seems to be what is needed by a semantic theory which encodes what we know in understanding language. It would endorse, as impeccable, reports of what is said by uses of sentences containing deWnite descriptions which reuse those sentences, whereas this would not be so if the Russellian approach were applied to understanding. On the other hand, the part of (or among) relation, which appears essential to a Russellian approach and which stands in the way of homophonic logical forms, provides some kind of explanation or analysis of how there can be a uniWed concept the capable of accepting predicates in the singular and in the plural, and mass and nonmass predicates. It is precisely because the part of relation is relatively unfamiliar that it can throw some explanatory light on the uniWed operation of the.21 The main aim of this chapter has been to show that there is room for a genuinely referential theory of descriptions. A subsidiary aim was to show that such a theory is likely to provide logical forms which, from the standpoint of semantic theory, are more accurate than Russellian theories, though the latter have something else to contribute: an explanation of how ‘the’ can exercise its uniWed function in various diVerent constructions.

Postscript I now think that only rigid subject expressions should be counted as referring expressions. My reason is that the speech act of referring involves attempting to single out an entity upon whose states the trans-world truth or falsehood of what is said is to turn. When successful, the referent introduced is the object which matters to truth or falsehood, actual or counterfactual, and so should be constant across worlds, and this will be reXected in the way in which reference is handled in the composition axiom. One impact of the new view is that genuine identities expressed with referring descriptions will be necessary if true, and so the idea of v-qualiWed identity can be scrapped, which is all to the good. Another impact is that, although Russellian and referential descriptions accounts will (in a sense) coincide on the truth conditions of sentences in non-modal languages, they will diverge with respect to modal languages. An interesting contrast will remain, which is structurally similar to that discussed here: it is between rigid referential descriptions, and Russellian descriptions with predicates rendered rigid, for example by the use of a suitable expression for actuality (where this notion is treated as Wxed to our actual world). ‘The actual F is G ’ will have as its Russellian logical form something like ‘Exactly one thing is actually F and it is G ’. By 21 As Sharvy stresses, it is certainly distinct from the concept introduced by the standard use of ‘part of ’ in mereology.

Sainsbury, Referring Descriptions 389 contrast, for referential descriptions, there is no need for a diVerence between surface and logical form. While referential deWnite descriptions and rigidiWed Russellian deWnite descriptions thus diverge in logical form, any sentence which can be seen as containing a referential deWnite description will be assigned the same overall truth conditions as if it were seen as containing, rather, a rigidiWed Russellian deWnite description; so considerations analogous to those raised in section 5 above will be relevant.

11 The Proper Form of Semantics Joseph Almog

The classical form of semantic theory is translational. We assign each target vernacular sentence a(nother) vernacular sentence, a ‘translate’, an entity articulating what-it-says or its content. This is the paradigm method ever since Frege, whether the semantics assigned general propositions or singular ones, whether it rested on a notion of absolute truth or truth at a possible world (index), whether it used Wrst-order or higher order ‘symbolizations’ and/or ‘logical forms’. I believe that three decades ago, Keith Donnellan was the Wrst to suspect that something is fundamentally wrong with the paradigm. He was driven by three classes of ‘outlaws’. The Wrst consists of sentences like ‘Vulcan does not exist’, the second of sentences of the form of ‘Vulcan appeared (to Leverrier) to be hot’ and the third of referential uses of descriptions as in ‘The man drinking the martini is sad’. I will refer to the trio as the Donnellan classes. The result was stated separately by David Kaplan, in his work on indexicals. Kaplan also articulated (i) the inadequacy of translational semantics and (ii) the need for a new form of giving the semantics of the vernacular. This consists of semantic descriptions from what Donnellan called the vantage point of ‘the omniscient observer of history’, and Kaplan characterized as ‘description from above’. The present work is dedicated to explaining what it means to give such a semantics.

This chapter is dedicated to Keith Donnellan. Initially I saw what I call below ‘the incompleteness of translational semantics’ only inside the framework of David Kaplan’s analysis of indexicals. It took me a while to appreciate that Donnellan’s result is in some sense purer. The two Wndings were formulated around the same time (1970–2) in the same institution, by two philosophers, who delved much into each other’s work. Nonetheless, they did not quite realize they hit on what is at bottom the same result. As we shall see, this is no accident. Another manifestation of ‘the incompleteness of translational semantics’ lies in various indirect discourse beliefreport puzzles (e.g. Kripke’s puzzle regarding Pierre). Since the matter is complex, I analyze it in a separate paper, Almog (forthcoming). By ‘semantics’, I target the semantics of ordinary language. Even if the ultimate answer is not diVerent, the quest for the proper form of semantics of formal languages raises its own issues. I owe thanks to Erin Eaker, Mike Thau, David Kaplan, Paolo Leonardi, Tony Martin, Bob Stalnaker, the editors of this volume, and to earlier conversations with Howie Wettstein. In dedicating the piece to Donnellan, I would like to express admiration for his way of doing philosophy, a constant reminder that looking carefully at, rather than past, our ordinary language may hold the key.

Almog, The Proper Form of Semantics

391

1. The Paradigm Questioned: Translational Semantics What is a translational semantics? A simple example, to whose analysis Donnellan made a lasting contribution, regards sentences involving proper names in subject position, ‘Penelope Cruz walks’. All of the following familiar proposals submit a translational semantics: (i) Frege’s theory, at least the one articulated by the lore, providing us with the clause ‘Penelope Cruz walks’ is true iV the leading actress of the movie Woman on Top walks. Such is also (ii) Russell’s classical theory, ‘Penelope Cruz walks’ is true iV there exists one and only one thing such that: it is a leading actress of Woman on Top and it walks. Mill’s ‘directly referential theory’, one revived and embraced by Donnellan and Kripke, is another such proposal: ‘Penelope Cruz walks’ is true iV Penelope Cruz walks. So are also modern variations: modally rigidiWed versions of Frege–Russell, for example, ‘Penelope Cruz walks’ is true iV the actual leading actress of Woman on Top walks and Montague’s higher order ‘Penelope Cruz walks’ is true iV the property of walking has the higher order property of Being-Penelope Cruz-true. These accounts diVer as to what they regard as synonymous with ‘Penelope Cruz’. But they all aim at the same thing: to give the semantics of ‘Penelope Cruz walks’ by providing a synonym, a translate, that speciWes its content. To introduce vocabulary that will help us draw some basic contrasts, we may sum the foregoing enterprise as engaging in a semantics from below. I describe the operation as ‘from below’, because throughout our giving of the semantics, we stay inside the vantage point and resources of the market-place user. We provide the semantics of one arbitrary market-place sentence used by you and me, for example, ‘snow is white’, by co-relating it with another such sentence, one unraveling what you and I said. This much formulates the task of semantics in the informal terms of ‘giving a synonym’ or ‘saying the same thing’. In a more theoretical vein, some have called the entity to be assigned by the translational semantics the proposition expressed by users of the sentence; others have described it as the logical form (henceforth LF) of the sentence (‘the true structure of what the user says (thinks)’). Yet other philosophers, liberated on both the ontological and logical fronts, for example, Russell and colleagues of Donnellan at UCLA working in the tradition of logical semantics (Carnap, Church, Montague, Kaplan, Burge), allowed themselves both vocabularies, which they often used interchangeably. What Donnellan points out is an impossibility result: (D)

The vernacular cannot be given a translational semantics

The Donnellan Classes: What is the Problem? True negative existentials Our Wrst troublesome class regards sentences like ‘Vulcan does not exist’. We have on hand as subject a proper noun followed by the application of negation to the verb. For

392

IV. Deﬁnites and Indeﬁnites in Semantic Theory

example, ‘Cruz does not walk’ is taken as a syntactic twin of our singular negative existential. Such sentences are transformed in classical ‘symbolizations’ or ‘translations’ to LF into the form ‘It is not the case: Cruz walks’. Thus instead of the subject term followed by the predicate ‘does not walk’, we introduce a sentential operator ‘it is not the case that’ that modiWes the non-negated sentence ‘Cruz walks’. I will call this procedure subject-subordination—the vernacular’s surface subject term is mapped into a subordinated clause, modiWed by a sentential operator newly lifted from inside the original complex predicate. The question ahead of us is: how is subject subordination going to enable us to assign truth to ‘Vulcan does not exist’? Suppose we apply subject-subordination to ‘Cruz does not walk’ and reduce it to ‘It is not the case that: Cruz walks’. The truth-value of this last is calculated from how things stand with the objectual referent of the name ‘Cruz’—if that referent, Penelope Cruz, walks, the sentential negation is false; if that woman does not walk, the sentential negation turns out true. No such simple calculation procedure is available for ‘It is not the case that: Vulcan exists’. The absence of a referent of ‘Vulcan’ seems to block our calculations for this sentential operator reduct just as it blocks our calculations for the original subject–predicate ‘Vulcan does not exist’. Simply put, subject-subordination does not seem to oVer a way out. If we allow ourselves the use of the apparatus of singular propositions, we can put the point this way. The individual Vulcan does not actually exist; this precludes just as much the existence of (i) the predicative singular proposition < . . . , not-exist> (the dots indicating the missing Vulcan) as it excludes the existence of (ii) the subject-subordination negative proposition << . . . , exist>, not-true>.1 True empty predicatives Our second problem class involves the likes of ‘Vulcan appeared hot to Leverrier’, and ‘Zeus was worshipped by the Greeks’. We begin, as before, by applying subject subordination. We get ‘It appeared to Leverrier that: Vulcan is hot’. But the sentential operator form oVers no solace. Kripke and Donnellan have given us powerful arguments that the sole semantic value such names have is their referent. If the name refers to no object, it refers to . . . no object and that is that. No embedding inside the scope of operators can resuscitate for the name a semantic content it did not have to begin with. If so, what makes true the whole sentence ‘It appeared to Leverrier that: Vulcan is hot’? Again, we can restate the problem in terms of the apparatus of propositional semantics. If there is no singular predicative proposition of the form < . . . , appeared-hot-toLeverrier>, there is equally no singular subject-subordination proposition of the form << . . . , is-hot>, appears-true-to-Leverrier>. 1 The sophisticated reader may already be thinking here of a third, less standard and trans Wrst-order logic, logical form assignment not involving subject-subordination and closer to the SP visible grammar. We represent ‘Cruz does not walk’ (‘Vulcan does not exist’) by way of ‘Cruz is such that it is not the case that: she walks’, where ‘is such that it is not the case that: she walks’ is the predicate at the LF level, the English analog of the l term (l x Not(Walk(x)) ). This analysis is discussed in detail below. Alas, it will not save ‘Vulcan does not exist’ from Donnellan’s result.

Almog, The Proper Form of Semantics

393

Let us call this Wrst problem the missing content problem. It arises from the fact that the proper noun cannot be translated to—be shown as synonymous with—a descriptive content bearing deWnite description. Our next problem strikes even if we assume such translation-reductions to descriptions. Let us pretend in what follows that in our speech ‘Vulcan’ does indeed mean ‘The Mercury Perturber’. But now let us roll time back to the mid-nineteenth century before any such name was introduced by the French astronomer Leverrier. Believing himself to be seeing in the telescope the heavenly body he seeks, Leverrier says ‘The yonder planet appears near the sun’. The rumors of the sighting spread to the British Royal Society. Without any clear sense of their exact origin or of Leverrier’s theory about the Mercury perturbations, a British astronomer, Sir Charles, asserts: ‘This heaven-knowswhere-it-is-planet (‘‘Charlie’’, to-give-it-a-name) seems small’. We, later users, now face the task of reporting this French–British sequence of events. We use the subjectpredicate (henceforth, SP) sentence (G1) below. I designate it using the letter ‘G’ in acknowledgement of Peter Geach, who was the Wrst to smell the problem it raises for sentential operator-reductions: (G1-SP)

Vulcan appeared to Leverrier near the sun and seemed small to Sir Charles

The report is natural and, cunning logicians aside, would be judged true. How is the translational semanticist to explain this? He would apply subject subordination and give us instead of the original SP sentence, a sentential operator reduct, designated here (G1-OP) to signify it is the operator-transform of (G1-SP): (G1-OP)

It appeared to Leverrier that: Vulcan is near the sun and it seemed to Sir Charles that: it is small

We may separate logical and conceptual worries regarding (G1-OP). Logically, the pronoun ‘it’, in the second clause, has the head noun, ‘Vulcan’, in the Wrst clause, as its grammatical antecedent. But it is very diYcult to provide a systematic-logical treatment of such pronouns. Familiar theories—treating the pronouns as bound variables or abbreviated (contextual) descriptions—soon run into problems. Notice that the most natural treatment, using an existential quantiWer on the outside to quantify in and bind the cross-reference variables, is excluded by the unavailability of the objectual value—Vulcan. Like remarks apply to any attempt to provide semantics in terms of singular propositions corresponding to the embedded clauses (‘It is near the sun’, ‘it is small’); there are no such propositions.2 The character of the bond between antecedent and pronoun emphasizes the logical angle. But behind the formal binding issue lurks a more ‘conceptual’ problem. We 2 Almog (1998a) discusses in detail why the two types of explanations do not explain well—while abiding by natural desiderata—Geach’s anaphoric structures. For more on a positive solution, see s. 3 below.

394

IV. Deﬁnites and Indeﬁnites in Semantic Theory

start with an SP structure that has many complex predicates (conjoined) but one uniWed subject—Vulcan: one subject then, with many complex predicates tucked on to it. The operator-transform inverts things: we get the predicates simpliWed but we engender a plurality of unruly subjects, the unbound dangling pronouns. On the analogy of the famed cross-world identiWcation problem in the metaphysics of modality, I would like to call ours the problem of trans-mind cross-identiWcation. Here is Leverrier’s mind and here is Sir Charles’s mind—on what basis do we cross-identify the distinct minds, make them co-thinkers, minds focused on the same thing? It is natural to look inside the heads of those about to be cross-identiWed: perhaps they share a common content, a common way of thinking, an identical mode of presentation of the purported body, in virtue of which we might say: ‘There, it is because of this common internal element of thought that we made them co-thinkers’. But what if there is no such common internal factor shared by Leverrier and Sir Charles? After all, the deWnite descriptions the two used to display their cognitive states were rather diVerent: ‘the yonder planet’ and ‘the heaven-knows-where-it-is-planet’. The circulated rumors surely did not provide any unique mode of presentation of the posited item. On the other hand, there is of course no external item, an existing object, out there to unify our co-thinkers (and idem, for an attempt to unify by means of singular propositions). If we have no entity—object or content, the familiar semantic values of translational semantics—to unify by, how do we cross-identify the minds of Leverrier and Sir Charles? In confronting this second, mind cross-identiWcation, problem, we may appreciate that it is the mirror image of the missing-content problem raised earlier. In facing the missing content problem, the translational semanticist told us that if we allowed a higher semantic value, some sort of ‘content’—an identifying property or intension or sinn, etc.—to be associated with our key term, for example, ‘Vulcan’, we would be out of the woods. But here, in our second problem, when we need to co-relate the two thinkers, the higher semantic value is exactly what does not guarantee the cross-identiWcation—Leverrier and Sir Charles are co-thinkers in spite of their associating diVerent contents (identifying properties, modes of presentations, etc.) with the purported heavenly body. What makes the two co-thinkers is not a common internal content but an external fact, probably unknown to both, but available to what Donnellan called ‘the omniscient observer of the history’—an ‘omniscient’ investigator, having all the pertinent facts accessible by stipulation, who is out to survey the history of the language and of the mental life of our pair of cognizers. Such an omniscient observer can link the cognitive states of the pair by tracing back, as a matter of historical fact, their mental episodes as having their origin in the same source-phenomenon. But now we confront our puzzle. How are we to inject this unifying information available to the omniscient observer into the mundane heads of the common users, Leverrier and Sir Charles? How are we to make a common natural ‘semantic content’ out of this complex historical information amassed by the omniscient observer?

Almog, The Proper Form of Semantics

395

True referential uses We consider in this class subject–predicate sentences with a deWnite description as subject, for example, ‘The man drinking the martini is sad’. Donnellan (1966) argued that such simple sentences (prior to any embeddings in the scope of sentential operators) display two uses of the deWnite noun-phrase. First, there is the attributive use, which may be glossed by the paraphrases (i) ‘the man with the martini, whoever he may be, is sad’ and (ii) ‘whoever is the unique martini drinker, sad he is’. In addition, Donnellan spots the referential use. To elucidate it, Donnellan asks us to imagine a party in which I am faced with a sad man, Mr Smith, holding a martini glass. Unbeknownst to us it is Wlled with water. Hidden from our eyes, in the kitchen, there is, as luck would have it, a happy man, Mr Jones, drinking a martini. He is the only one to so drink in the party. According to Donnellan, we judge my use of ‘The man with the martini is sad’ as true and made true by the individual I had in mind, viz. Mr Smith, in spite of his not drinking a martini. Furthermore, my use rests in no way on Mr Jones, who is merely what Russell would call the denotation of the attributive use of the description. Now, as is well known, Donnellan’s distinction has engendered much discussion. Some, for example, Kripke, have claimed the phenomenon Donnellan pointed out— the referential use—has nothing to do with the semantics of the language but merely with ‘pragmatics’. Furthermore, Kripke adds that for all that Donnellan has shown, the English sentence ‘The man drinking a martini is sad’ has univocal truth conditions, indeed those given by Russell’s quantiWcational reading: ‘the man drinking a martini is sad’ is true iV there exists one and only one thing such that: it is human and it drinks martini at the party and it is sad. Thus, according to Kripke, the truth of the sentence I used depends on how things stand with Jones. Any further ‘reference’ to Smith is mere speaker’s reference, a matter not unique to deWnite descriptions but pervading the use of language: proper names, demonstratives, adjectives, verbs, etc. can all be subverted by a speaker who has in mind something and uses a linguistic device non-conventionally to get to that intended target. My task here is not to dissect what was intended by Donnellan’s distinction. Nor am I bent on rebutting his critics or criticizing his own way of drawing the distinction. For the record, I will just mention that I believe (and here I go beyond Donnellan) that there is a semantic ambiguity in English for deWnite noun-phrases (and other referential denoting phrases, e.g. indeWnites); it has nothing to with what I, the speaker, intend, but is rather written into the conventions of the language; it is a distinction that is not recoverable by scope distinctions (which are separable and needed anyway); what is more, the distinction has nothing to do with epistemic glosses Donnellan falls back on (in the attributive use, the speaker has no knowledge of the satisWer; in the referential use, he knows who it is or is acquainted with it). I have argued this matter—really a plurality of issues—elsewhere and here none of this will matter.3 3 I have dealt with the Donnellan–Kripke dispute in detail in Almog (1998b). I there claim that there is a semantic ambiguity in English (indeed for all denoting phrases, e.g. ‘Every philosopher in the room’); that

396

IV. Deﬁnites and Indeﬁnites in Semantic Theory

In the present chapter, I am focused on one issue only. I would like to assume, with Donnellan, that there is a referential use in which (i) Smith is picked out by a rulegoverned procedure and (ii) the English sentence, as I use it, is true. (i) and (ii) lead Donnellan to a fundamental problem. The problem is raised by Donnellan (1966). Donnellan is faced with Linsky’s earlier claim that the referential use is neither true nor false. Not so, claims Donnellan. My use of the sentence in the party is very true indeed. The problem is not with our truth-value assignment; we have a clear truth-value all right. The problem is rather in assigning my use the statement it expresses. This then is Donnellan’s key question: what is the statement expressed by my use of the sentence? The terminology of ‘statement’ is Strawson’s. I will assume that we could well ask: what is the proposition I express? Or, we may wonder, what is the content I expressed? In a nutshell: what is it that-was-said by JA, on that occasion? I assume that one semantic content associable with the sentence is the attributive content, the one Kripke sees as calling for Russell’s quantiWcational truth conditions. As mentioned in the foregoing footnote, I disagree. But, for our present purposes, very well, let there be such a Russellian attributive–quantiWcational content. What matters is that it is not the one I express in the party. But then, what determines what I do express? Many modern readers have suggested that, rather than expressing the Russellian general proposition (there exists one and only one thing such that . . . ), the proposition I express on this occasion is singular, viz. that Smith is sad. But even if this is correct on this occasion (or in proposition-free terms that Donnellan prefers, even if I had said-of Smith that he is sad), this does not solve the main problem. The main problem is this. There seems to be no systematic rule-governed way of associating with an arbitrary referential use of ‘The man drinking the martini is G’ the statement (proposition, content, etc.) it expresses. On diVerent uses, diVerent objects would be referred to (at my next party, I may well pick out Jones, who will be in front of me, with a martini glass, etc.). Thus the situation here seems even more worrisome than with indexicals. Consider the sentence ‘I am sad’. We may wonder what is the proposition it expresses. The answer is: there is no such unique proposition. DiVerent uses will determine diVerent propositions. Nonetheless, there is a rule-governed determination of the pertinent proposition. We may write with David Kaplan: For any context c, ‘I am sad’ is true at c iV the agent of c is sad at c. Donnellan observes that no such rule, conventionally associated with the lexical meaning of the quoted singular term, seems forthcoming for referential uses of descriptions. We can say to ourselves: we sure know one thing—whoever is picked by the description had better be sad. But what is picked by the description? No translational condition, derived from the English meaning of the lexica in the description, seems Kripke’s quantiWcational-Russellian conditions do not get right the attributive use (of the deWnite description) and that Donnellan’s ‘have in mind’ conditions do not get right the referential use.

Almog, The Proper Form of Semantics

397

forthcoming on the right hand side of the biconditional. At the same time, we all know ‘what’s going on’. We can gloss it informally: whoever was the individual that the agent had in mind, and thus the individual to whom he refers by using ‘the man drinking the martini’, had better be sad. And yet, this longish description ‘whoever was on the mind of the agent while he used ‘‘the man with the martini’’ ’ is surely not what the agent himself had in mind; or said; or expressed as his content. Back in the party, I surely was not saying: the individual who is at the source of what I now have in mind and am about to refer to with ‘The man drinking the martini’ is sad. That is not what I said. And so our question is: can we or can we not write down systematic truth conditions—whether translational or not, but at least as good as those assigned systematically to indexicals—for the referential use of descriptions? Can we write a general condition of the form: ‘The man drinking the martini is G’ is true on the referential use in a context c iV. . . 4

2. Singular Negative Existentials Donnellan’s Dilemma for Logical Form Semantics: Classical Logic I begin our discussion of Donnellan’s troublemakers with the simplest of all, singular negative existentials. One way to develop a ‘feel’ for just how translation-semanticsresistant this form is is to consider the logical tradition’s most sophisticated attempts to defuse the problem. I have in mind a methodology due to Russell, with later variations due to modern ‘free logicians’. Donnellan’s analysis is that the standard logical techniques do not solve the problem. I agree. Russell’s key notion is that of a logically proper name. I will here stress the italicized adverb. I read the whole phrase to mean: a proper name adequate for the purposes of logic. To qualify for the title, Russell posits both epistemic and logical conditions. The epistemic condition got most of the attention of philosophers—the referent of the name must be known (to the thinker) with immunity to Cartesian doubt. The logical condition is that the sole linguistic contribution of the name is its actual bearer. Thus, if the name has no actual bearer, Russell concludes, quite apart from any epistemological issue, that the name is not a logically proper name. Classical logical languages (e.g. the standard predicate calculus) stipulate in Russell’s vein the logical condition—the individual constants include only those that actually 4 A suspicion related to the present one, that referential descriptions threaten standard ‘translational’ semantics, is discussed in this volume in Sainsbury’s chapter. He too contemplates a separation of (i) the truth conditions of a sentence from (ii) the logical form (proposition) expressed by the sentence. An interesting development of this separation is provided by Eaker (2002).

398

IV. Deﬁnites and Indeﬁnites in Semantic Theory

designate. In consequence, no true sentence of the form ‘Not (Exists(a))’ may be expressed. Thus, even if, per impossible, Russell were to allow vernacular proper names the assignment of what Donnellan approves as an authentic logical form—the form ‘Fa’—one making them individual-constant like, there would be no way, at least by the canons of standard predicate logic, to cater for the true ‘Vulcan does not exist’. We could provide for the (logically) true ‘Neptune exists’ and (logically) false ‘Neptune does not exist’ but not for ‘Vulcan’ sentences. As Russell was to stress (and predicate logic was to standardize), the classical rule of existential generalization (EG) is valid over any such use of a logically genuine name ‘a’. But if we use EG and if we make ‘Vulcan does not exist’ true, we verify ‘there exists at least one thing such that: it is Vulcan and it does not exist’. This would be unacceptable. So, as long as we keep EG, we had better stipulatively bar any true premise of the form ‘Vulcan does not exist’. Donnellan does not agree with this stipulative logical methodology. For Donnellan, ‘Vulcan does not exist’ is intuitively true. How could logic stipulate away what the history of the world and of our language make so plainly true? Let us take stock and reXect on the previous paragraph. In the austere set-up of the syntax and semantics of classical predicate logic, and with no mention of propositions, Donnellan already provides a baggage-free formulation of his fundamental dilemma, Donnellan’s dilemma. To articulate the ‘invisible’ logical form of vernacular sentences, we rely on the visible forms of predicate logical formulae. One could propose that a syntactically authentic LF assignment to (symbolization of ) subject–predicate vernacular sentences with proper names, for example, ‘Vulcan (not) . . . ’, would call upon the atomic form ‘Fa’ and its sentential negation ‘Not (Fa)’. By the step I called above subject-subordination, our original ‘Vulcan does not exist’ reduces to ‘it is not the case that: Vulcan exists’. But now, says Donnellan, even with this reduction, we are trapped by a dilemma: if our syntactic form-assignment to the term ‘Vulcan’ is authentic, our semantics is Xawed—we cannot calculate the right truth-values for negative existentials. The other horn of the dilemma—Wne truth-values, Xawed syntax—is even less appealing and was already rejected by Donnellan in his attack on Russell’s ‘truncated description’ theory of names. We get the right truth-values but at the price of fabricating syntactically complex, and thus inauthentic, logical forms (symbolizations) for vernacular names.5 5 I believe that Donnellan and Kripke have given deWnitive reasons why vernacular names are not truncated descriptions. But I should like it noted that, even if we ignore their arguments and follow the Russell horn of the dilemma, we do not actually get a uniformly correct account of the truth-values. To see this, we need to consider the following two types of ‘negative existentials’. (i) For present-time and actuality-bound negative existentials, e.g. ‘Vulcan does not exist’, Russell calls upon the small scope reading of the surrogate description, e.g. ‘It is not the case that: there exists at least one thing such that: it is uniquely perturbing-Mercury’. However, this will not work when it comes to (ii) negative existentials looking away from actuality and the present time. Consider ‘Bill Clinton will not exist’ and ‘Bill Clinton might not have existed’. Russell has to call wide scope readings to the rescue: e.g. ‘There is at least one thing such that: it is uniquely an ex-American president married to an American senator from New York and it will be the case: he does not exist’. Saved then? Not Russell. For him, it is from the frying pan into the Wre: the last conjunct in the wide scope reading is ‘meaningless’, for it contains ‘will . . . not

Almog, The Proper Form of Semantics

399

Donnellan’s Dilemma for Logical Form Semantics: Free Logics Is Donnellan’s dilemma inescapable? It may seem that we have reached an over-pessimistic conclusion just because classical predicate logic is too restrictive. What if, following a variety of ‘free logics’, we were to insist on making the LF-assignment to ‘Vulcan does not exist’ (i) syntactically authentic, but, at the same time, we would (ii) look for a ‘liberated’—freer—semantics, for example, one in which the classical rule of EG is invalidated? When this course is followed, all that may be concluded from ‘Vulcan . . . ’ is ‘there be at least one thing such: it is Vulcan . . . ’. This last involves no claims of actual existence; indeed free logicians often add to the matrix ‘There is something such that: it is Vulcan . . . ’, the conjunct ‘ . . . and it does not actually exist’. When we follow the free logic course on EG, we are in for a criticism from Donnellan—we have taken an ‘easy’ way out, we have not really solved the problem of true singular negative existentials. This calls for a detailed explanation. In the single rule of ‘existential generalization’, classical predicate logic semantics joins a variety of independent steps regarding (i) predication, (ii) naming and (iii) quantiWcation. Let us separate then the three theses embraced by predicate logic: (i) Predication is actualistic—the truth of a predication, of an ‘Fx’ frame (under an assignment), entails the actual-existence of a satisWer of ‘F ’. This idea involves no allusion to (a) the semantics of proper names (indeed it would apply even if the language had no singular terms) nor do we mention (b) the semantics of quantiWcational operators (e.g. the idea applies just in the atomic fragment to the satisfaction-conditions of predicates). (ii) Naming is actualistic—classical logic submits that primitive subject terms have only actualia as designata. Again, we should note that this is not a thesis about (a) predication or (b) quantiWcation. It is a thesis about the range of designata (referents) of primitive subject terms (e.g. proper names). (iii) QuantiWcation is actualistic—the English quantiWer locution ‘there is at least one thing such that . . . ’ and the Wrst-order operator ‘(Ex) . . . ’ are to be read actualistically, as ranging over a domain of actual existents only. Three diVerent ideas then: we may deny one while sustaining the two others. I focus here on one such combination, noted by Donnellan. He submits that, sometimes, the truth of a predication does not force the existence of any actual object bearing the property (satisfying the predicate). Some predications, for example, ‘x runs’, may so entail. But not all of them. For example, ‘x does not exist’ does not (as we shall see, ‘x appears to be approaching’does not, and so on). exist’ as a predicate of an individual. Russell thinks any such use of ‘exist’ is a category mistake. Yet harder problems arise for Russell from the likes of ‘Spinoza existed once but he no longer exists’, if the variables of the wide scope quantiWer range only over present existents (as they should). I have analyzed this last problem in the papers mentioned in the above footnotes.

400

IV. Deﬁnites and Indeﬁnites in Semantic Theory

In denying (i), actualist predication, we are not thereby forced to deny in turn (ii) and (iii): we are not thereby suggesting that names may designate merely possible individuals or/and that the quantiWer-locution is to be read non-actualistically. We may, Donnellan did, simply deny that predication is actualistic, while holding on to both (ii) and (iii), namely that naming and quantiWcation are actualistic all the way down. This sends us back to the invalidity of the classical rule of EG. It is indeed not valid for subject–predicate vernacular sentences. But this invalidity does not submit that possibilia are designated or that our quantiWers are possibilistic. The rule is not valid because there are subject–predicate premises with empty subject terms ‘N’ that are true and yet (i) the subject–predicate ‘N exists’ is false as is (ii) the quantiWer generalization ‘there is at least one thing such that: it is N and . . . ’. We can now see why Donnellan thought free logic merely distracts us from his dilemma. Free logics mistreat the empty name ‘Vulcan’. For free logics either (i) make the semantics of the name be that of a deWnite description, for example, ‘(the x) (Mercury-Perturber (x))’, a description itself understood without existential presuppositions and satisWed by a merely possible individual, or alternatively (ii) the name ‘Vulcan’ is taken to directly refer to that posited merely possible individual. From Donnellan’s standpoint, this is no solution to the problem but rather what Russell described as ‘a paltry evasion’. Driven like Russell by a robust sense of reality, Donnellan wants us to recognize at the outset that ‘Vulcan’ is an empty name, in the strict sense that it is both (i) a name (not a description) and (ii) an empty name, not one that does refer after all to a twilight zone entity, a mere possibilium. I am not arguing here that this is the correct stance—that (i), metaphysically, there are no such merely possibles and that (ii), semantically, names cannot name ‘them’. Right or wrong, this view is presupposed by Donnellan (as it is by Russell): our designations (and in turn, quantiWcations) may include only actualia. So, even if free logics may be commended for sticking to syntactically authentic LF assignments, they are to be criticized for serving themselves to semantically inauthentic values. We are still trapped by Donnellan’s dilemma—no way to provide ‘Vulcan . . . (not) . . . ’ sentences an account that is both syntactically and semantically authentic.

Summary: Donnellan’s Ur-Observation about Naming The basic point that diVerentiates Donnellan’s approach to empty names from the spectrum of technical Wxes just reviewed is conceptual. It concerns the very idea of a name, what is and what is not constitutive to being a name. Herein lies the most fundamental observation of Donnellan, one that provides the foundation to his own account for negative existentials. I will call it Donnellan’s ur-observation: (DUO)

It is not essential to (the very idea of ) a name to have a bearer.

Almog, The Proper Form of Semantics

401

The theorists whose cunning Wxes were just reviewed, be they from the descriptivist tradition (e.g. Russell) or the Millian tradition, be they classical or free logicians, have insisted that the very essential function of a proper name is exhausted by its standing for an object. Driven by this idea about the essence of names, some philosophers impose on the vernacular a constraint dear to the classical logical tradition: if an individual constant does not have an objectual value, it is not part of the language. This is stipulated into the vocabulary used by such theories, in what they call a genuine (or ‘logically proper’) name. Another response, imagined by various free logicians, is to extend the range of objectual values so as to make seemingly empty names come out as loaded names after all. It is in this way that we get as contents propositions involving non-existing constituents and mere possibilia in the role of referents. The two reactions diVer in the ‘remedy’ they propose. But they agree on the fundamental diagnosis: a genuinely empty name is a mutant instance of the species of names. Something has to be done to bring it into line—expel it from the language or load it with a value. Donnellan views the logical methodology—on either remedy—as overly prescriptive. His primal motif is ‘Don’t Wght ordinary language’. If we are to describe the language rather than Wght it, ‘Vulcan’ must be recognized as genuine a name and as proper, logically speaking, as ‘Neptune’ is. What is more, the insistence on the objectual value of the expression—the object it ‘carries’—obscures for Donnellan the critical fact that the semantics of proper names, loaded or empty, may be articulated prior to the discovery of whether a given name is loaded or empty. This uniformity holds for a variety of basic issues: the falsehood for both loaded and empty names, on the modal front, of disguised description theories; the presence of very little identiWcatory information ‘in the head’ of ordinary linguistically competent users; the importance of social practices insuring that the names ‘catch on’ after an initial individual introduction act; Wnally, the critical role of historical transmission in connecting that initial use and contemporary uses. The foregoing traits are fundamental to our institution of names and are all prior to and independent of checking out whether the name indeed hooks up with a real object. I class Donnellan’s principle (DUO) as fundamental because it pleads for a shift in what should be viewed as the essence of naming: the introduction of subjects-ofdiscourse rather than the carrying of real objects under one’s shoulder. The insight will guide various generalizations and reWnements suggested below.

3. Donnellan’s Omniscient-Observer-of-History Semantics Donnellan’s positive proposal, henceforth Donnellan’s rule, regarding negative existentials goes like this:

402

IV. Deﬁnites and Indeﬁnites in Semantic Theory (DR)

A sentence of the form ‘Vulcan does not exist’ is true iV the history of use of ‘Vulcan’ ends in a block.6

The proposal calls for some annotations. The Wrst four are Donnellan’s own. The remaining are driven by the present generalization.

Non-Translational Truth Conditions The right-hand side of the (DR) biconditional is not a translation of the object language—English—negative existential. Donnellan emphasizes that his rule does not provide a content-entity: what-the-sentence-says or the proposition it expresses. It rather states for each sentence a certain condition. The condition articulates when and only when the sentence is true.

The Omniscient Observer The condition of truth is not given by the market-place user of the sentence. It is articulated by what Donnellan calls ‘the omniscient observer of history’, he who, by looking from outside at the history of uses, describes in his own sophisticated language of communication chains, intentions, etc. the workings of the market-place language. In a vocabulary familiar from the semantics of formal languages, the meta-language (ML) deployed by the omniscient observer and the object language (OL) used by the market-place speaker are each both richer and poorer than the other in certain respects. On the one hand, the meta-language deploys rich concepts of chains of communication, etc. and non-local forms of description surveying the whole of the language’s history. On the other hand, the object language allows uses of empty names and true applications of negations of the existence predicate, devices the omniscient observer does not serve himself to.

Inter-Linguistic Conventional Translation Though the ML does not translate the OL negative existential sentence, it accounts, again ‘from above’, for inter-linguistic translation among object languages. Donnellan observes that it is an empirical fact that ‘Pe`re Noe¨l est ge´ne´reux’ is the conventional translation of ‘Santa Claus is generous’; furthermore, ‘Robin des Bois est ge´ne´reux’ is not such a translation. This would not be reXected by any content-entity assignment— 6 Donnellan (1974) uses intuitive examples to gloss when we have on our hands a ‘block’. It is essentially a negative idea—the historical explanation of the present use of a name, e.g. ‘Santa Claus’, ends up with no clear unique original candidate that is an existing object. As we shall see below, when we dissect the notion of the source of a communication chain, the notion of ‘block’ requires some reWnements. In particular, we will need to attend to cases, emphasized to me by Mike Thau, in which there is a real object, say, a peasant standing on a mountain in a thunder storm, causing, in some sense, the speaker’s introduction of a new name for a thunder god, ‘Thor’. Nonetheless, ‘Thor’, as we use it, is not a referring name borne by the real peasant; it is an empty name of a god.

Almog, The Proper Form of Semantics

403

an object, a conventional null object (the empty set), a singular proposition. All such assignments homogenize all empty names, making them all have the same ‘value’. On the other hand, historical tracing does explain why we inter-translate ‘Santa Claus’ and ‘Pe`re Noe¨l’; furthermore, we are told (i) why we cannot use as a translate of ‘Santa Claus’ the French name ‘Robin des Bois’ but (ii) why ‘Robin des Bois’ is the translate of ‘Robin Hood’. The basic observation of Donnellan is this: inter-linguistic translation is explained not by a ‘from below’ matching of contents; it is explained by an omniscient observer’s matching of origins of word-uses.7

Uniformity of Explanation Donnellan emphasizes the importance of giving a uniform explanation for the workings of referring and empty names. This emphasis is an oVshoot of his ur-observation (DUO). A single historical mechanism governs the transmission of names, empty or loaded. The rule of tracking the source is the same for both; and the resolution of questions of co-reference—with or without reference—is the same. It is not that in the case of ‘Cicero is bald’ and ‘Cice´ron est chauve’, we give one, content-bound explanation, but in the case of ‘Pe`re Noe¨l’ and ‘Santa Claus’ we fall back on another, ‘from above’ and content-free explanation in terms of matching-origins. The semantics does not consist in (i) a primal stage, the assignment of an object (content) to a name, that (ii) in a subsequent stage induces sameness of content (same object assigned). Things run the other way round: by matching of word origins, we reach Wrst the resolution of sameness of truth conditions and inter-linguistic translatability. Whether any such name has, in addition, an existent entity as a bearer is the posterior question.

Extending Donnellan’s Observations Names of past existents Donnellan speaks of treating uniformly two categories of names: empty names, for example, ‘Vulcan’, and referring names, for example, ‘Neptune’. I would like to amend it to a three-way classiWcation; the added category consists of names that refer to individuals that no longer exist. The category is interesting because it combines features of both empty names—the purported entity does not exist, and loaded names— the name does refer. In this way, the third category helps to pry apart metaphysical and semantical questions. Metaphysical questions prime investigations of what there is— what objects and propositions exist at a given time (in a given possible world). Semantical observations concern the mechanism by which names (more generally, single words) are propagated in history, from dubbing to contemporary use. 7 A similar explanation will apply to common nouns, e.g. the conventional translation of ‘unicorns’ as ‘licornes’ and not as ‘griVons’.

404

IV. Deﬁnites and Indeﬁnites in Semantic Theory

Subject versus predicate trouble8 Donnellan’s rule of truth (DR) is speciWc to this one predicate ‘ . . . does (not) exist’. Donnellan does not give us a general rule for calculating from (i) the history of a name N and (ii) the nature of the predicate ‘is (not) F ’, and (iii) the facts of the world, the truth-value of the sentence ‘N is not F.’ I see Donnellan as fusing two issues. (a) When is a name N empty? (b) When does the predicate ‘does not exist’ apply truly? The notion of ‘ending in a block’ combines the two by assigning ‘true’ to the whole sentence ‘N does not exist’, regardless of the separate questions (a) and (b). The rule tells us nothing about how to evaluate sentences like ‘Neptune is (not) inhabited’ or ‘Vulcan is (not) in my room’. A generalized Donnellan rule should treat the name N prior to the insertion of this or that speciWc predicate. This leads to two separate desiderata regarding such subject– predicate sentences. The Wrst concerns the subject, the second the predicate. The Wrst is that the question whether a name N is empty must be settled before the slotting of any predicate. The second desideratum regards the ‘ . . . is (not) F ’ part. Some predicates are existence-receptive, for example, ‘ . . . explodes (is exploding)’. Tensed, they require existence at the pertinent time of application, for example, past existence of the subject term referent as in ‘N exploded (was exploding)’. In a second class, we Wnd the existencerejective predicates—true application entails the non-existence of the referent of the subject term. Natural examples are provided by ‘ . . . does not exist’, ‘ . . . is unreal’, ‘ . . . is illusory’, ‘ . . . is a pseudo (false, hallucinatory, etc.) object’. Finally, we have the existence-neutral predicates. For example, ‘ . . . was (is) believed by Leverrier to be hot’ and ‘ . . . was sought by Leverrier’. Such predicates could go either way: with ‘Neptune’ they apply truly to an existent, with ‘Vulcan’, they apply truly without an existent. The foregoing three-way classiWcation of predicates is independent of the theory of the subject position proper name N. It would apply even if all names were disguised descriptions. The source of a communication chain The notion of ‘block’ is essentially negative—something at the beginning of the chain is not ‘standard’. But I believe this judgment goes against Donnellan’s own ur-insight (DUO), the very idea of a name in the vernacular should not have an existing referent (bearer) built in; an empty name is as genuine a name. This observation about the object language—English—should be reXected in the meta-language used by the omniscient observer to describe our name-using practices. It is so reXected if we do not stipulate into the meta-language deWnite description ‘the source of the chain leading to our use of N ’ the idea that the source must be a real existing object. 8 The present discussion of issues extends the earlier separation, in the context of the classical vs. free logic debate, of naming actualism versus predicate actualism versus quantiWcation actualism.

Almog, The Proper Form of Semantics

405

Here then is an alternative conception that respects DUO. It also has the added virtue that it attends to the fact that a myriad of causes might generate a communication chain. I do not presuppose any more that the Wrst link in the chain (if you will, ‘the Wrst cause’) is a real existing object. What I do presuppose is that, if a chain exists, a source exists—no chain without a source. This much seems an elementary truth of metaphysics: chains are real relations between worldly events. Notice that, on the present conception, given an expression that looks like a name, say ‘Zinguela Serenissima’, it is not a priori that a source of a chain exists. The reason is that a chain may not exist here in the Wrst place. ‘Zinguela Serenissima’ has no source behind it because it was never introduced in history as a name and, consequently, never caught on as a name. Whether there is a chain, and thus a source, is a historical question. Let us assume now our candidate name N is historically real and a chain exists. Thus, a source-of-the-chain exists. The remaining question is what kind of phenomenon was the source of the use of N. In the case of ‘Neptune’, it was a real object—the planet. In the case of ‘Vulcan’, it was a real phenomenon—the perturbations in the orbit of Mercury. In yet another case, the origin of the common noun ‘unicorn’, it is most likely that it was a misperception of rhinoceroses in India that got things going. In the Loch Ness situation, the source of the name ‘Nessie’ may have been a misinterpretation of the ripples on the loch’s watery surface. And so it goes—the real phenomenon that is the source may or may not be a real object. So, in the general case, we associate, with our vernacular name N, the omniscient observer meta-language description ‘the source of our contemporary use of N ’.9 This much governs any use of N, whatever predicate we tack on. Next, we append to N a particular predicate, say the existence-entailing ‘ . . . sings’. This would require that the source of the chain would be (i) a real existing object, (ii) that sings. This much is imposed not by the name, whose source might not be a real object, but by the conditions of true application of this speciWc predicate, ‘ . . . sings’. Not so for existencerejective predicates. For example, for ‘N is unreal (is a Wgment of our imagination, is illusory, etc.)’, we get: the source of our use of N is (respectively) not a real object, is a Wgment of our imagination, is an illusion. This applies also to the case ‘N does not exist’. Here the truth condition is: the source of our use of N is not a really-existing object. That makes ‘Vulcan does not exist’ true. Finally, we come to existence-neutral predicates, for example, ‘N was sought by Leverrier (believed to be hot, etc.)’. We now require that the source of our use N was sought by Leverrier (believed by Hob to be hot, etc.). This much does not demand that the source be a real existing object. Whatever that source was—a real existing object, perturbations in the orbit of a real object, misperception of ripples on the water—the pertinent source was sought, was believed to be hot, etc. 9 Some indexing of this description taking care of ‘our use’ has to be added. See below the discussion of generation of use situation.

406

IV. Deﬁnites and Indeﬁnites in Semantic Theory

In a nutshell, our general meta-language condition for a name asserts that (i) the source is always real but (ii) may not be a real existing object. This conception satisWes well what I take to be Donnellan’s fundamental observation—empty names abound in the vernacular and are genuine names indeed. The requirement of a real object as a source comes not from the name but from the application conditions of speciWc predicates. More on the source of a chain and reference-Wxing10 My old friend Mike stands on a little hill in the midst of the outback, waving a hammer. A storm is gathering, lightning strikes, and a deafening thunder rolls. Impressed by Mike’s dramatic posture and scared out of their wits, the locals decide to introduce the name ‘Thor’ for the thunder god. I later report ‘Thor does not exist’. My report is true. Now, our generalized Donnellan condition submits that this would be the case iV the source of my use of ‘Thor’ is not a real object. But is not the source of the use a real individual, my old friend Mike? To deal with such apparent troublemakers, we need to keep some distinctions. First, I am here focused on those cases in which (for a given name N), ‘N does not exist’ is true. Our task is to explain the ground of this truth. For example, the ancients thought that (what we view as) the planets were gods. Let us say they called, at least in the evening, the second planet from the sun ‘Booz’ and worshipped it as a god. Of course, we might think to ourselves: ‘Really, Booz is just Venus. Venus exists, so Booz exists’. Suppose that is how we think of this case (I don’t, see below). But then, ‘Booz does not exist’ is false. As such, it is not a sentence I am here analyzing. So, one of our pretheoretic tasks may be to agree on which intuitively given cases are to be judged as reported truly by ‘N does not exist’. To reiterate, I am focused solely on such reports, true negative existentials. I come now to my second observation. Within the class of true negative existentials, I would separate the following cases: 1. Shifted reference. Let us assume, ‘St Claus’ originally successfully named a real man, Nicolas. Later, the use of the name shifted, making our present use of ‘St Claus does not and never existed’ true. So, here the source of our contemporary use, in the sense that matters, is not the real man Nicolas but the shifted use. No doubt, the shifted use was caused by some real enough natural events related to Christmas celebrations among real human agents. But the source of our current use—as I intend ‘source’—is not any one of these natural events or causes. The source is our intention, at one point, to pick out this unreal Wgure. 2. Mistakes within a theory. A second category concerns the likes of ‘Vulcan does not exist’—no shifted reference, just a mistaken positing of an object to explain real 10 In this sub-section I am indebted to Mike Thau (who insisted on the ‘Thor’ case). Similar worries have been expressed to me by Erin Eaker, David Kaplan, and many years ago, by Howie Wettstein and Rogers Albritton.

Almog, The Proper Form of Semantics

407

astronomical events. Yet again, we should not confuse a variety of real enough causes with what I target by the source of our (and of Leverrier’s) use. Doubtless, Leverrier was not a daydreamer or mystic. He did attend to real causes, among them the real behavior of a real planet, Mercury. All of those real causes brought it about that Leverrier intends to use ‘Vulcan’ for a new item he thought was out there. This ‘thinking about’ is the source of his (and our) use of ‘Vulcan’. 3. Ungrounded frameworks. A third case arises for those who reason as follows: Venus does exist but the religious ancients were all wrong—their god Booz does not exist, there is no such heavenly god. So, ‘Booz does not exist’ is true when I say it now. And if so, it was always true, even of the ancients, though nobody then realized it. We could see that ‘from the omniscient observer’s point of view’, when the ancients were then using the sentence ‘Booz exists’, the observer would smile to himself and say ‘forget it’. Of course, we may view this third case as a species of the second—a mistaken theory involving the positing of gods (rather than planets). I prefer to view it as substantially diVerent—what leads us astray is radically false beliefs about the sort of entities involved (gods, ghosts, etc.) rather than mistakes within a given scientiWc framework. What makes ‘Booz does not exist’ true? The source of my (our) use is the ancient religious use. And when they introduced it, they had in mind a god. Reviewing the history, we go back to bestowal and replay their introduction and it is obvious— from above—that we should say: they failed to refer; we should not say: they successfully referred to Venus but were a bit wrong about its nature. It is in this way that I want to think of the above-mentioned story of ‘Thor’. My old friend Mike’s standing on a hill caused the believers to introduce ‘Thor’. But Mike is not the source of their and our use. The source is their thoughts—let us pick out the god of thunder . . . 11 Double indexing In speaking of the application conditions of a predicate, in for example, ‘Spinoza does not exist’, we must separate two types of information. The Wrst regards the time of use of the name, the second the time of evaluation of the application (truth) of the predicate. Failure to separate the two types of information may lead us to describe two very diVerent situations as ones at which Spinoza does not exist. There is Wrst the situation that held in the year ad 1001, with the language spoken by the users of that period. 11 I should like it noted that this reference-Wxing thought is so to speak ‘referential’, not ‘attributive’. The believers name someone they are already and independently cognitively focused on. To direct the new-coming audience, they pick the item they have in mind as ‘the god of thunder’, whether it is really the god of thunder or not (we should recall that ‘Hesperus’ was so Wxed with Venus, via ‘the evening star’, even though Venus is not a star). That is, the name-introducer is already ‘acquainted’ (in the normal sense of this word) with his intended target. His use of the ‘reference-Wxing’description is ‘referential’, it merely serves as an ‘acquaintance proxy’, as a guide to those who are not already so focused. In my view, all name introductions are driven by such referential uses of descriptions (viz. no name introduction is fully logically attributive).

408

IV. Deﬁnites and Indeﬁnites in Semantic Theory

At that time, our name ‘Spinoza’ did not refer (yet) to any individual whatsoever. In contrast, we have a second situation, for example, in the year ad 2001, where in our language(s), the name does refer albeit to an individual that no longer exists. Not only can we say in our time, of our time (as evaluation-time), that Spinoza does not exist in it; we can say truly now that he didn’t exist then, e.g. in ad 1001. Our condition of truth must thus separate, what I call the time of the generation of use from the time of predicate application, what has been called by David Kaplan and Hans Kamp the time of evaluation. Used in 2001, ‘Spinoza does not exist’ is true at any point (before and after) outside his lifetime. But set in a context of use before the name was part of the language, the sentence is semantically non-evaluable. Double indexing again: modal matters Kripke observes that ‘Vulcan does not exist’ may be true of a counterfactual situation w, even though the name ‘Vulcan’, as the inhabitants of w use it, does trace an object that exists in w. And vice versa: the name ‘Neptune’ may not trace, as used in w, a locally existing object but our ‘Neptune exists’ is true of w, for the yonder planet does exist there. More generally, our account of names must respect their modal rigidity, their picking out throughout counterfactual set-ups the very item they actually pick out. In contrast, the meta-linguistic description used by the omniscient observer ‘the source of our use of N ’ is obviously not rigid.12 It has been suggested that we could set things straight by inserting into the description a ‘rigidiWer’. For example, ‘the actual source of our use of ‘‘Neptune’’ ’ is said to counterfactually track the planet as reliably as ‘Neptune’ does. The suggestion misunderstands the role of the meta-language description.13 The rigidiWcation is driven by the search for an object language synonym, by a commitment to a translational semantics clause that speciWes what is said: ‘ ‘‘Neptune is blue’’ is true iV the actual source of our use of ‘‘Neptune’’ is blue’. Taken as a contentspeciWer, the biconditional obscures the question—whose description are we dealing with here? For the rigidiWed source-description is not a market-place synonym of the name. The description is the omniscient observer’s rule for tracking the origin of N. This much is prior to any modal evaluations—the description is simply not what the name means; it provides an explanation of the history of the name’s use. The meta-language description applies to the generation-of-use situation. This explains not only the correct mode of modally evaluating ‘Neptune does not exist’ in a 12 I assume, with the logical–semantic tradition, that this very name ‘Neptune’ may be considered in a counterfactual situation in which it has quite a diVerent historical origin. The tradition assumes that neither (i) the name’s having a referent nor (ii) the actual historical origin of a name’s use, are essential to the name’s identity. As a metaphysician, I would deny this, asserting the essentiality of origins of any historically real item, names included. But I ignore here this metaphysical issue. 13 There are other problems with the synonymy or true modal equivalence claims. In truth, the description does not track Neptune as well as the name. The diVerence comes in counterfactual worlds w, where Neptune does not exist. But this does not matter for present purposes, where my reservations about the rigid description apply at an earlier stage.

Almog, The Proper Form of Semantics

409

counterfactual world w, but also why Kripke has it right regarding the falsehood of our use of ‘Vulcan (unicorns) might have existed’.14 The latter may seem true, if we apply the meta-language description to the counterfactual world w—an entity is traced over ‘there’ and made into the referent. It thus seems that Vulcan (unicorns) does (do) exist in w and so our ‘Vulcan (unicorns) might have existed’ is actually true. Nonsense, exclaims Kripke. Apply ‘The source of our (proper, common) name ‘‘Vulcan’’ (‘‘unicorn’’)’ to our generation of use situation. It traces no referent (planet, species). This Wxes things once and for all—we have no object to take with us to counterfactual evaluations, in particular no object to take to w. Thus, ‘Vulcan (unicorns) might have existed’ is false. But, as Kripke correctly insists, it is equally misleading to put things in terms of ‘Necessarily: Vulcan (unicorns) does (do) not exist’. Says Kripke (1980: 24): ‘Perhaps according to me the truth should not be put in terms of saying that it is necessary that there should be no unicorns, but just that we can’t say under what circumstances there would have been unicorns.’ Consider for contrast ‘Necessarily: the round square does not exist’. This is something we can say: in each world of evaluation, we have some condition to evaluate. The evaluation simply leads to no entity that is the round square. Not so with ‘Vulcan’ and ‘unicorns’. It is not that there is this content (proposition)—that Vulcan does not exist or that unicorns do not exist—and it keeps being true in all worlds. It is rather that: given our generation of use situation, no world w can make false our historically Wxed sentence ‘Vulcan (unicorns) does (do) not exist’. Kaplan’s related result I speak throughout the chapter of Donnellan’s result but, around the same time (early 1970s), David Kaplan hit on a structurally similar result—the impossibility of a translational semantics for the vernacular.15 Though the two were close interlocutors, somehow the similar discovery was not recognized. There are many reasons for this. One reason is that Kaplan’s class of examples—indexical sentences like ‘I walk’—was quite diVerent from Donnellan’s diet of subject-phrases (proper names and descriptions). Secondly, when it came to Donnellan’s paradigm case, the singular negative existential, Kaplan experimented with a translational semantics involving ‘gappy’ propositions, a fact that may have prevented him from seeing the import of Donnellan’s case. But third and most important, the semantics Kaplan did give to indexicals, by means of his ‘characters’, did not suggest to Kaplan at the time the radical breakdown spotted by Donnellan, whereby the semantics of the market-place language may only be given from the ‘outside’, by the omniscient observer of history. Kaplan’s clause for ‘I walk’ reads: ‘For any context c, ‘‘I walk’’ is true at c iV the agent of c walks at c’. Kaplan emphasized:

14 See Kripke (1980: 156–7). I have tried to defend Kripke’s intuitions in the above cited papers. 15 I refer to Kaplan (1989a). Kaplan (1989b) ampliWes on some of these themes.

410

IV. Deﬁnites and Indeﬁnites in Semantic Theory

(i) The right-hand side of the biconditional did not provide a synonym of the market-place indexical sentence. (ii) The right-hand side did not articulate what a given user of that sentence says, for example, when used by Kaplan, ‘I walk’ says that David Kaplan walks. (iii) The description ‘the agent of the context’ occurs in a meta-language that is itself, in principle, indexical free. It is, Kaplan here calls upon Quine’s notion, a ‘language of the laboratory’ in which the scientiWc description of our marketplace language is stated. (iv) When it comes to modal evaluations, for example, of ‘I might have been walking’ (asserted by Kaplan), the description ‘the agent of the context’ is not applied to the possible world of evaluation. We rather apply it before any such evaluation, in the generation of use situation, to Wx the referent of ‘I’, for example, Kaplan. Then we take that referent, the man Kaplan, not the reference-Wxing description, and check, in a given world, whether he satisWes the pertinent predicate. So far, so good. Features (i)–(iv) replicate the proWle we have seen Donnellan ascribe to his omniscient observer’s description. But Kaplan’s analysis had another dimension— one pertaining to cognition—that distinguished it from Donnellan’s. In a section called ‘Epistemological Remarks’, Kaplan (1989a: 529–40) described his characters as cognitively accessible to the market-place user, as providing after all, a sort of classical ‘cognitive meaning’. And so interpreted, as descendants of Frege’s sinne, the characters do not lead to Donnellan’s radical realization about how—and in whose language—we give the semantics of market-place English. Indeed Kaplan went to describe the characters as: (v) providing what the ordinary competent speaker knows when he masters the indexical expression; (vi) generating a priori pieces of knowledge as in ‘I am an agent’, ‘This is an object demonstrated’, ‘He is male’, etc.; (vii) explaining from inside the user’s head Frege-like identity puzzles, for example, the cognitive value of informative identities like ‘I¼He’ and ‘This¼That’; (viii) providing what the speaker has in the head when he says truly ‘this pink rat does not exist’ (pointing out a mistake to a hallucinating friend). In spite of there being no proposition expressed, the speaker apprehends and communicates a signiWcant character that provides the cognitive value for both him and his interlocutor. The subsequent use of characters by Kaplan, Perry, and many others to replicate Fregean explanations of cognitive puzzles has done much to propagate this impression of a ‘cognitively accessible meaning’. Technically, characters do not oVer a translation—they fail on the modal front to simulate the rigidity of the object language indexicals. Nonetheless, they seem to oVer a from-below description of what is in the

Almog, The Proper Form of Semantics

411

head of the market-place user. Indeed, this encouraged Perry, Stalnaker, and others to introduce on top of what ‘I walk’ says—the mundane proposition—a second, hyperproposition, roughly ‘the agent of the context walks’. The hyper-proposition is modally distinct from the mundane proposition the market-place ‘I walk’ expresses. But it may seem epistemically and semantically equivalent for every speaker with the content of the object language sentence ‘I walk’. The divide between the object language ‘I walk’ and its meta-language character has been minimized.16 Kaplan’s characters present a dual proWle: semantically, recall (i)–(iv) above, they are on a par with Donnellan’s omniscient observer description. On the other hand, as (v)– (viii) show, they have been assigned cognitive roles Donnellan never assigned his metalanguage description. I think that Donnellan had it right on just how radical the breakdown of translational semantics is—the semantical rules describe our language ‘from above’ and are not cognitively accessible meanings. Driven by his observations, we need to rethink the Fregean roles assigned to characters: are they indeed cognitively available to the ordinary user? In whose language are they articulated? And if not accessible to the common user, will they still explain Frege’s informative identities? On reXection, we see that characters are not pieces of knowledge of ordinary users but descriptions from above of our conventional practices, descriptions provided by the omniscient observer of history about the rules governing our indexical expressions. As such, they do not generate a priori truths; they could turn out false of ‘I’, ‘This’, etc. if suitable evidence about indexical-and-demonstrative using practices came in; in turn, quite a diVerent, not from-inside-the head but from-the-omniscient-observer’soutside-perspective, explanation of Frege’ s informative identities is called for (for proper names as for indexicals). This at any rate is my reading of Kaplan, with the help of twenty-Wve years’ hindsight. So read, Donnellan and Kaplan made a simultaneous discovery about the very form of the semantics of the vernacular.17

16 But see next footnote for a later reconsideration by Stalnaker of these cognitive uses of characters. 17 My focus in this work is Donnellan’s ideas. I only mention in passing, without genuinely arguing, that this is how we should rethink Kaplan’s characters. Detailed arguments, focused on Kaplan’s work, are provided in Almog (forthcoming), with special emphasis on a ‘from above’ rather than from ‘inside the head’ explanation of what makes an ‘a ¼ b’ identity informative. A similar reconsideration of characters as cognitive meanings and of the role of the ‘hyper-proposition’ is undertaken by Stalnaker (2001). The resulting ‘meta-perspective’ strikes me as operating at the level of Donnellan’s ‘from above’ external observer. Having mentioned Kaplan’s dual approach to characters, let me add the following ‘updating’ annotation. In more recent work, Kaplan (forthcoming) oVers a more ‘from above’ approach to characters. He also gives further examples, beyond indexicals, of the non-translational nature of semantic rules. His new examples involve the phenomenon of expressives, locutions that, by convention, express an attitude on behalf of the speaker. Among his examples Wgure beguiling cases such as ‘ouch!’ ‘oops!’, ‘Kaplan is a honkey’ (as contrasted with the non-expressive ‘Kaplan is Caucasian’), ‘That bastard Kaplan was promoted’, and other colorful such. This later discussion of semantics emphasizes much more, in the Donnellan vein, the non-translational character of the meta-language, the descriptions being given by a scientist operating from above and, most critically, the non-availability of the analyses ‘in the heads’ of common users.

412

IV. Deﬁnites and Indeﬁnites in Semantic Theory

The basic truth condition With the foregoing annotations under our belt, I am at long last ready to state my generalization of Donnellan’s truth conditions. It reads: (GD)

A sentence of the form ‘N is F ’ used at context c is true at t iV the source of the chain traced by the omniscient observer of history (OOH) leading to our use of N at c is F at t.18

Let us try the condition on the three types of names and the three types of predicates segregated. Consider Wrst a name of a present existent, ‘Penelope Cruz’. Applying our condition, we get ‘P. Cruz walks’ is true iV the source of our use of ‘Cruz’, viz. that woman, walks; ‘P. Cruz does not exist’ is false because the source of our present use, that Xesh and blood woman, does really exist; Wnally ‘P. Cruz is believed to be in town’ may well be true if the source, that real woman, is believed to be in town. I consider next a name of a past existent. ‘Spinoza walks’ is not true now for the source of our present use of the name, that man, exists no longer and thus cannot be presently walking. But ‘Spinoza walked’, as we use it, may well be true, if evaluated at an appropriate moment in the seventeenth century: the source of our use did walk then. Next, ‘Spinoza does not exist’ is true now—the source of our current use of the name, that man, no longer exists. Finally, ‘Spinoza is believed to be in town’ may well be true if the source of the name’s use is believed to be in town, a belief that may well arise with those unappraised of the vicissitudes of that source. Finally, consider ‘Aphrodite’, the name of the nonexistent goddess. ‘Aphrodite walks’ is false; walking demands a presently existing individual and the history of our use of ‘Aphrodite’ does not track as a source any such individual. ‘Aphrodite does not exist’ is presently true. I do not have enough information about the real source of our use; I take it to be ancient Greek tales of an irresistibly alluring divine being. Be that as it may, the source was not then (at the time of introduction)—a real existing person. Finally, ‘Aphrodite is believed to be in town’ may well be true: for example, the source of our use of the name may well be taken, by her uninformed followers, to be in town. This may apply to readers of a text in which she is mentioned by name (or auditors of an oral tale in which she is spoken of ) who take the account to be a factual report of an alluring present-day personage. But it may also apply to true believers, rare as they might be, in the existence of Greek gods (e.g. present-day worshippers of Zeus, etc.).19 SP structures versus indirect discourse It is a tribute to the generality of the OOH account that we have an uneventful passage from single-case predications, for example, ‘Vulcan seemed to Leverrier near the sun’, 18 We may further relativize the evaluation to a possible world. 19 And so it goes with our third-category predicates, e.g. for ‘Aphrodite appeared alluring to ancient Greeks’ to be true we need to check, roughly, whether the source of our use ‘Aphrodite’ appears to be alluring, not whether it is alluring. We must remember that throughout our third class of predicates, we are to check out what

Almog, The Proper Form of Semantics

413

to multiple predications, for example, as in ‘Vulcan seemed to Leverrier near the sun and appeared small to Sir Charles’. We recall that on the method of sentential operator reduction this step trapped us with Geach’s problem of inter-sentential ‘free anaphors’ and the attendant puzzle of cross-identifying the minds of co-thinkers: for example, the problem generated by ‘It seemed to Leverrier that: Vulcan is near the sun and it appeared to Sir Charles that: it is small’. It is a virtue of the truth condition Donnellan supplies that it extends smoothly from single predicates to multiple predications. The OOH demands that the source of our use of ‘Vulcan’ seem to Leverrier near the sun and appear small to Sir Charles and . . . We encounter here on a more general scale a fundamental opposition between the method of sentential operator reduction and the Donnellan-induced SP truth condition, what each Wnds as the ‘paradigm’ case and what each views as a diYcult limit case. The sentential operator methodology regards SP structures as ‘hard cases’ in need of reduction to sentential operator forms. It reduces (i) the single case SP ‘Cruz does not walk’ to a sentential operator form ‘It is not the case that: Cruz walks’. In turn, (ii) the SP multiple predication ‘Cruz walks but was yesterday (might have been) sitting’ reduces to a sequence of operator sentences: ‘It is the case that: Cruz walks and it was the case (might have been the case) that: she sits’. The trans-mind cross-identiWcation case is not far behind. We reduce ‘Penelope Cruz is feared by Hob to be in town and believed by Nob to be admirable’ to ‘Hob fears that: Cruz is in town and Nob believes that: she is admirable’. We may recall that this two-step methodology engendered problems of anaphoric connection. There is (i) the logical problem—the pronoun ‘she’ in the later clauses is unbound—and (ii) the attendant conceptual problem—how do we cross-identify the referent of ‘she’ (in the scope of the various later modalities and attitudes) with the referent of the head noun, viz. how do we keep track of Miss Cruz throughout all these embeddings? In answer to the conceptual problem, I see the method of ‘reduce-it-all-to-sentential-operator-form’ as a form of translational semantics: we seek an object language expressible content (‘propositional constituent’), shared by the head noun and the late pronouns, to ground the fact that a single subject—Penelope Cruz—is referred to throughout the anaphoric chain. We may say: the unity of the subject (predicated) is guaranteed by the unity of an identifying content, a content that cross-identiWes the subject, Penelope Cruz, in the minds of various thinkers (Hob–Nob, etc.), mentioned in the anaphoric chain. The OOH methodology reverses here the order. It regards SP structures as the problem-free and simple cases to start with. To go back to Geach’s old story: (G1-SP)

Cruella (the witch) is believed by Hob to be in the neighborhood and believed by Nob to be dangerous.

seems to be the case, what appears, what is believed, etc. of the source of the name, not what is actually historically true of that source.

414

IV. Deﬁnites and Indeﬁnites in Semantic Theory

We understand this case as a simple case of predication of the source of ‘Cruella’ by a string of predicates. Next, we regard as derivative the indirect discourse (sentential operator) form: (G1-OP)

Hob believes that: Cruella is in the neighborhood and Nob believes that: she is dangerous.

With hindsight, I read our inability over the years to give a unifying content (propositional) semantics to (G1-OP) as not surprising: in the most general-complex cases, indirect discourse cannot have a translational semantics. We unify Hob and Nob not by a common internal content but by the external information available to the OOH. This last makes the truth condition of the indirect discourse structure non-translational, whether the head noun is referring (‘Cruz’) or is empty (‘Vulcan’). Let me gloss this in some detail. In reports of the (G1-SP), the name ‘Cruella’ (‘Vulcan’, ‘Cruz’, etc.) is used by the reporter. Neither Hob nor Nob may have any inkling of it. Indeed, Hob and Nob may have no inkling of each other’s use. For example, Hob may have used ‘A red toothed witch is in the neighborhood’ (in short, he used A) and Nob may have expressed his belief with ‘the ominous looking woman is dangerous’ (in short, he used B). It is now the reporter who coordinates the foregoing separate activities into a single report. He selects his favorite subject term, in our case ‘Cruella’; he Wxes it as the coordination point—the coordinated subject—for the report that is to come; next, he goes on to predicate it with ‘believed by Hob to be in the neighborhood’ and ‘believed by Nob to be dangerous’. For this to be true, our three protagonists need to be historically cowired: the origin of the reporter’s ‘Cruella’ is the very object that is the source of Hob’s state and, independently, of Nob’s state. It is important to take notice of the following fact of reporting: the reporter does not record in his report how either Hob or Nob expressed their states; he himself may not have a record (knowledge) of their mode of referring to Cruella. What matters is that, as a matter of historical fact, it is this common source that induced their states: Hob’s use of A, Nob’s use of B, and the reporter’s coordinative use of ‘Cruella’. Referential uses of descriptions I come Wnally to the third Donnellan class—referential uses of descriptions. My question was: can we formulate a rule-like condition, in the manner of Kaplan’s characters, for referential uses of descriptions? I propose the following: (Ref )

For any context c, ‘The F is G ’ is true, on the referential use, with respect to c, at the moment of evaluation t iV the source leading to the agent of c’s use of ‘The F ’ is G at t.

As before, the right-hand side truth condition is not translational: it does not provide the content asserted by the market-place user. What is more, very much like Donnel-

Almog, The Proper Form of Semantics

415

lan’s rule of historical chains for proper names, the condition is articulated by the omniscient observer. This means that we Wnd in it (i) allusions to facts about historical relations not accessible to either the local market-place speaker or hearer and (ii) the rule draws on complex notions (sources of chains, transmissions, intentions) that go beyond the repertoire of the mundane user. What is going on in a referential use is this. The omniscient observer looks at, for example, the party situation, ‘from above’. He notices that the martini-drinker, Jones, away in the kitchen, impinges in no way on my cognitive state. He observes that Smith, water-drinker though he is, is the pertinent source (perceptually, in this case) that leads to my production of the phrase ‘the man drinking a martini’. The observer can say: before JA ever produced this phrase, he was already thinking about—he was cognitively bound to—Smith. JA could have produced a variety of expressions to ride back to Smith on that already-existing wire. JA could have used: ‘He’, ‘This’, ‘That man’, ‘You’, ‘Smith’, ‘Jones’ (a misnomer of Smith), ‘A man in front of us I would not like to point to’, etc. In all of these cases, we would have a referential use—the use of the linguistic device would be parasitic on a prior cognitive bond, a thinking about, tying me and Smith. On this omniscient observer reading of the rule, referential uses do not guarantee special epistemic privileges to the speaker. There need not be recognition on the part of the speaker of who it is he has in mind (thinks about). Recall the speaker, here JA, has in mind Smith because the history of the world put Smith in my mind, by placing him right there in front of me. Having Smith in mind, thinking about him, provides me with no special knowledge about Smith or special tracking (and identiWcatory) capacities often invoked (including some of the time by Donnellan) in analyses of referential uses. In the same vein, a referential use should suggest no special authority on the part of the speaker. Referential uses are often read to suggest that it is the speaker’s ‘decision’ or ‘intention’ or ‘stipulation’ that determines what the descriptive phrase refers to (a latent ‘humpty-dumptyism’). The picture many have in mind is of my head initiating the act of reference. It thus seems that it is absolutely up to me who the referent is. On my understanding, this is a misreading of both (i) Donnellan (what he means by ‘the individual intended’ or ‘individual had in mind’) and (ii) a misreading of what is really going on in these referential uses. The account of what I ‘have in mind’ does not start with what my head ‘stipulates’. My head shows up in the ‘middle’ of the story. The history of a given referential use begins with a source situation external to me, out in the world—be it an object (e.g. Neptune) or some misperceived phenomenon (e.g. Vulcan). The source generates a chain making me have something in mind. Thus, what I have in mind (the item I intend) is what the history of the world made my mind have (intend). This is reXected in our rule for the truth of ‘The F is G ’. It reads: ‘the source of the agent’s use of ‘‘The F ’’ is G ’; it does not read ‘the item that the agent stipulates by his use of ‘‘The F ’’ is G ’.

416

IV. Deﬁnites and Indeﬁnites in Semantic Theory

The referential use promises neither epistemic nor stipulative privileges. On the other hand, referential uses induce modal privileges, viz. rigidity. As before, the omniscient observer description applies to the generation of use situation, not to the time-and-world of evaluation. We trace the source of my use of ‘the man drinking a martini’. We locate that object: Smith. If we now predicate him with ‘will be (might have been) happy’, we are checking whether Smith, regardless of any drinking behavior at some time t or world w, is happy at t (or: at w). In turn, there is no checking in t or w, of whether Smith is the source therein of my use of any phrase whatsoever (I may well not exist at t or w). We simply check of Smith—the source of my use of ‘the man with the martini’ in the generation of use situation—whether he is happy at t or w. Finally, the referential use Wts nicely my earlier cross-mind Hob–Nob observations: (i) about the key notion of ‘sameness of cognitive focus’ (co-thinking) and (ii) the mechanisms of object-language attitude reports. First, regarding sameness of cognitive focus. Let us go back to our party. Looking at Smith in the corner, I say to my friend Francesca: ‘The man drinking the martini bianco is sad’. A few steps away, you, Reginald, are looking at Smith in the corner, and you say to your friend Isabella: ‘The man drinking the martini rosso is sad’. Finally, in the kitchen, our common friend Jerry, looking at Jones, who is, as a matter of fact, drinking a martini bianco, says to Emanuela: ‘The man drinking the martini bianco is sad’. Observing things from above, the omniscient analyst will tell us that you (Reginald) and I are co-focused, we share a cognitive state, in spite of our heads not agreeing on descriptive content; what is more, both of our descriptions do not denote Smith; and yet, it is he, Smith, however linguistically speciWed by the two of us, who is the common source of our shared cognitive state. In contrast, even though Jerry uses the very descriptive phrase I use, he and I are not co-focused; we are not thinking about the same item. I should like it noted that this omniscient observer classiWcation would run the same way if the phenomenon observed was, as in the Hob–Nob cases, hallucinatory, for example, pink rats due to our over-drinking at the party. The Joseph–Reginald pink rats are shared, both of us are in a common cognitive state of agitation over those rats (even if we pick them up by two referential descriptions diVering in lexical meaning). On the other hand, Jerry’s kitchen-drinking-induced pink rats would put him in a diVerent cognitive state. The history-induced classiWcation of cognitive states impacts the mechanism of object-language attitude reports. This has been noticed by Donnellan (1966: 302– 4). He notes there that if I used referentially ‘the man with the martini is sad’, the reporter R may truly report me with a host of co-referring expressions of his choice, for example, ‘Joseph said that the president of the college (Smith, that man, etc.) is sad’. I propose to extend Donnellan’s original aperc¸u to our cases of a co-focused plurality of agents (Hob–Nob cases). Reporting the party situation, our reporter R may say truly ‘Joseph said that Smithy is sad and Reginald did so too (said that too)’. This is a true report even though neither you nor I used the expression ‘Smithy’. We may be totally

Almog, The Proper Form of Semantics

417

unaware of this nickname of his. The report is true because R’s use of ‘Smithy’ traces back to the very source of my use of ‘the man drinking the martini bianco’ and the source of your use of ‘the man drinking the martini rosso’. The three uses of the three diVerent expressions are generated by the same source. The three of us are thus cowired. Like remarks would apply if the source phenomenon was misperceived (Vulcan, pink rats, witchy Cruella, etc.). I could say referentially ‘The witch with the huge axe is dangerous’. You, prompted by the same scary footprints, could say ‘The ghost with these enormous boots is dangerous’. Reporting from the scene the public scare caused by the footprints, the TV live reporter would frame things truly by saying ‘Joseph said that Cruella is dangerous and so did Reginald’. And yet, neither I nor you Reginald are familiar with the name ‘Cruella’. Let me summarize my Wndings about the referential use. On the present analysis, Donnellan hit on a fundamental phenomenon that he illustrated using deWnite descriptions. The phenomenon itself does not necessarily involve (deWnite) descriptions. Ironically, in this Kripke was right, though, of course, Kripke makes the point to dismiss the importance of Donnellan’s discovery, whereas I make it to emphasize its fundamentality. What Donnellan discovered concerns the very idea of thinking-about an object. And Donnellan’s point is this: a thinker may be Wxed on an object of thought (more generally: has in mind something even if a misperceived source) prior to and independently of any linguistic reference. The cognitive bond between the historical source and the mind of the thinker is already Wxed by the time the thinker selects a linguistic device to ride back, along the established cognitive wire, to the source. The use of the linguistic device is to communicate to others the cognitive bond he is already part of (what he already has in mind). In this respect, whatever expression the thinker selects for communicative purposes (proper name, demonstrative, deWnite description, indeWnite), the expression is governed in the thinker’s use by the prior cognitive bond. It refers to whatever was the source of the agent’s use of the expression. Last but not least, the governing rule is not what the thinker himself says; it is a description due to the omniscient-observer semanticist of how the expression is used.20 20 As just mentioned, Kripke is right in saying that the phenomenon of referential use is not speciWc to deWnite descriptions (to reiterate, he so means to banish the Donnellan referential use from semantics; my motivation is the opposite). Indeed an extension of the present kind of omniscient observer rule applies both to (i) referential uses of names and (ii) indeWnite descriptions. Suppose we give the correct rule for my party’s referential use of ‘Jones is sad’, where I use ‘Jones’ while having in mind the right-before-my-eyes, Mr Smith. The rule would read: check sadness of whoever was the source of the agent’s use of ‘Jones’; in our case, this would round up Mr Smith, not the man in the kitchen, Mr Jones. Of course, this complex rule is not stating the content, the what-I-said; all I said was of the very man, Smithy, that he is sad. The foregoing account extends in a natural way to referential indeWnites, e.g. as in the anaphoric chain ‘A man came into my oYce. He was agitated. He tried to sell me a book.’ Donnellan (and Strawson before him) pointed out that such indeWnites introduce speciWc subjects of predication rather than issue a quantiWcational claim (about the non-nullity of the number of satisWers of some feature). For Donnellan, my referential use of the indeWnite is that of a crypto-name, as if I said ‘A man, namely Smith, came into my oYce . . . ’. In Almog (1998b), I argue that indeWnites are semantically ambiguous; furthermore, that one of their linguistic functions is referential—to introduce particular subjects into discourse (though the manner in which they do so is by a conventional rule, not Donnellan’s manner which

418

IV. Deﬁnites and Indeﬁnites in Semantic Theory

4. Summary I have argued that the three Donnellan’s subject–predicate constructions escape the methods of translational semantics. Reduction to sentential operator form, as in ‘It is not the case that: Vulcan exists’ and ‘It appears that: Aphrodite is approaching’ oVers no salvation. Furthermore, in the referential use of ‘the man drinking a martini’, there is nothing like an operator we could somehow extract from inside the simple sentence. Philosophers and linguists have tried to Wnd some such hidden operator (‘I assert that . . . ’). To me, it is clear that the sentence ‘The man drinking a martini is sad’ is an operator-free subject–predicate sentence, if any vernacular sentence is. The problem for translational semantics we encounter here is not due to any operator, visible or imaginary. The problem is due to the key nominals that occur in the simple sentences: ‘Vulcan’, ‘Aphrodite’, ‘the man with the martini’. It is impossible to give the semantics of these nominals by translating them, namely by specifying for each a Wxed content its user expresses. In turn, we cannot compose a natural contentproposition expressed by subject–predicate sentences involving these nominals, even when it is clear to us such sentences (‘Vulcan does not exist’, ‘Aphrodite appeared alluring to ancient Greeks’, and ‘The man with the martini is sad’) are . . . true. This last fact is Donnellan’s absolutely primal datum: in spite of our not having a clear way of assigning a content, the subject–predicate sentences of all three classes are obviously true. And there is no mystery about their truth. Speaking ‘from the outside’, we know full well, by citing the history of the use of the nominal, how it is that the sentences are true; also, what, in the world, made them true. This suggests the problem lies in our method of description of the truth–value calculation. With translational semantics, we tend to presuppose that the truth-value calculation must come from below, by looking for the objectual semantic value, be it an object or a sense, assigned by the semantics to the pertinent component expression. We then construct a conventional semantic content for the sentence as a whole—the proposition it expresses—and get a truth-value by matching this content with worldly facts. Donnellan urges that this content-bound method is not a possible truth-value calculation procedure when empty names are present. On reXection and by uniformity considerations, the content method may not be the truth-value calculation procedure we follow even when the names are loaded or when a nonempty referential deWnite description is used. We may hope that a semantics of a language would produce for us objects-ofthought or contents or logical forms or translations, something to wrap our mind around. So driven, we are inclined to regard the empty ‘Vulcan’, ‘unicorns’ and the Smith-bound shifty referential use of ‘the man drinking the martini’ as aberrations. is more speaker-relative). Indeed, substitution of a given indeWnite on both sides of the biconditional of the rule (Ref ) gives us such a conventional rule, the omniscient observer rule for referential indeWnites (in subject position).

Almog, The Proper Form of Semantics

419

But if we view the task of semantics as accounting, given the history of the language and the world, for the intuitive truth-values of our sentences, we might see in these ‘aberrant’ examples rather standard cases—the history of use of the key nominals, together with the history of the world, provide nicely for the truth-value of the sentences, viz. the truth of ‘Vulcan does not exist’, ‘Aphrodite appeared alluring to ancient Greeks’, and ‘The man drinking a martini is sad’.

12 On a Unitary Semantical Analysis for DeWnite and IndeWnite Descriptions Peter Ludlow and Gabriel Segal

1. Introduction Philosophers of language (and semanticists) do not agree on much, but few have felt reason to doubt that there are at least two kinds of descriptions in natural language: deWnite descriptions (e.g. of the form ‘the F ’), used in sentences which say that there is a unique satisWer of F, and indeWnite descriptions (e.g. of the form ‘an F ’), used in sentences which claim only that something or other satisWes F. However, if analytic philosophy had spawned in languages other than English and German, this assumption might not appear so innocent. In many languages—even many Indo-European languages—we do not Wnd these two kinds of descriptions, at least not on the surface. In Russian, for example, neither deWnite nor indeWnite articles appear on the surface: ‘the man’ and ‘a man’ would both be expressed by the Russian equivalent of ‘man’. If we suppose that natural languages do not diVer from each other in their logical form,1 then there are a limited number of possible paths we can take to explain the discrepancy between English and these other languages. The Wrst path is simply to say that, for example, Russian has both deWnite and indeWnite descriptions, but that they do not appear on the surface; they are simply unpronounced lexical items. In other words, in Russian and similar languages, there is more semantic structure than meets the eye. The second path is to say that languages like English that appear to have deWnite and indeWnite determiners do not actually have two determiners in the sense of two logical We are indebted to Delia GraV and Stephen Neale for discussion, to Kent Bach, Emma Borg, Ray Eluguardo, Kasia Jaszczolt, Ruth Kempson, Mike Nelson, and Jason Stanley for comments on an earlier version of this chapter, to Anne Bezuidenhout and an anonymous reviewer for comments on the penultimate draft, and to Zolta´n Szabo´ for comments on multiple drafts. 1 This assumption could be challenged, of course, but work in generative linguistics over the last three decades has convinced us that diVerences between natural languages are often superWcial, and that at the level of description we are interested in those diVerences are vanishingly small. The thesis is empirical, however, and may yet turn out to be false, but it has been fruitful to date and we accordingly adopt it here.

Ludlow & Segal, On a Unitary Analysis

421

operators with distinct semantical contributions. Rather, we can say that English has two surface grammatical elements (‘a’ and ‘the’) with the same literal meaning. As we noted above, few have taken this option, but it has been on the table for some time. Earlier attempts to defend the thesis include Kempson (1975), Breheny (1999), Szabo´ (2000), and Zvolensky (1997), and the idea is at least entertained in Heim (1982) and Kamp and Reyle (1993). In this chapter, we revisit the position and enlist a number of new arguments in defense of it. Before we get into the arguments for the proposal, it is Wrst necessary to see that there is a great deal of intuitive appeal to the idea (as strange as it may seem at Wrst blush). Accordingly, we will begin with a discussion motivating the general position, and a possible technical execution of the idea. We will then turn to some of the arguments for the position (in section 3) and will conclude (section 4) with some general observations about why the idea initially seemed so strange.

2. The Intuitive Appeal of the Basic Thesis The basic idea under consideration is that for both ‘An F is G ’ and ‘The F is G ’ the literal semantics is exhausted by the interpretation that [9x: Fx][Gx) receives in a standard truth-conditional semantics. In other words, in English there is less semantic structure than meets the eye.2 It has been widely noted that there are many cases in which we use deWnite determiners without expressing uniqueness. Consider the following examples:3 (a) (b) (c) (d ) (e) (f ) (g) (h)

John went to the dentist. John was hit in the eye. The boy scribbled on the living-room wall. (Du Bois 1987) Towards evening came to the bank of a river. (Christopherson 1939) Let’s go to the pub. (British English) John drove into the ditch and had to go to the hospital. Take the elevator to the sixth Xoor. (Birner and Ward 1994) No problem, I’ll get the maid to do it. (Epstein 2000)

2 A third possibility, consistent with what we say here, is that there is no genuine Russellian determiner at all (either deWnite or indeWnite) but rather that the surface determiners are, strictly speaking, free variables which might be unselectively bound by discourse elements, and which, when unbound, would be interpreted like an indeWnite description. Heim (1982) and Kamp (1981) have proposed this analysis for the determiner ‘a’, but so far as we know, a similar proposal has not been made for the determiner ‘the’ (see Kamp and Reyle 1993: ch. 3) for discussion of puzzles surrounding the treatment of ‘the’ in Discourse Representation Theory, DRT). The point of this chapter for DRT would be that the analysis can be extended likewise to the determiner ‘the’. We pass over that possibility here because we do not want to be deXected by discussions of the relative merits of DRT. 3 We are indebted to Anne Bezuidenhout for bringing some of these examples to our attention and to Abbott (1999) for tracking down their sources in the literature.

422

IV. Deﬁnites and Indeﬁnites in Semantic Theory

Our view is that these cases are not aberrations, but in fact more closely reXect the semantics of so-called deWnite determiners. What we intend to show is that in those cases where there is apparent deWnite force or uniqueness, pragmatic considerations are in play. The basic argument is straightforward. Just as Kripke (1977) argued that we need not posit an ambiguity in ‘the F ’ to account for the referential and attributive uses of descriptions, and just as Ludlow and Neale (1991) argued that we need not posit a semantical ambiguity to account for the referential, speciWc, deWnite, and quantiWcational uses of ‘an F ’, one can likewise argue that one need not draw on a semantic distinction between ‘the F ’ and ‘an F ’ to account for the diVerent uses to which these descriptions are applied. SpeciWcally, ‘the F ’ and ‘an F ’ literally make the same contribution to truth conditions: they both have the semantics that we would ordinarily represent as ‘(9x)Fx’ in Wrst order logic. The deWnite use of both ‘the F ’ and ‘an F ’ is derived by standard Gricean pragmatic resources.

3. Speaker’s Reference/Semantic Reference and Beyond Donnellan (1966) pointed out that deWnite descriptions can be used in (at least) two diVerent ways. On a so-called attributive use, a sentence of the form ‘The F is G ’ is used to express a proposition equivalent to ‘Whatever is uniquely F is G ’. For example, on seeing Smith’s badly mutilated corpse, Robinson might say ‘The murderer of Smith is insane’ thereby expressing the idea that some unique individual murdered Smith and whoever that individual is is insane. On a referential use, a sentence of the form ‘The F is G ’ is used to pick out a speciWc individual, x, and say of x that it is G. For example, suppose Jones is on trial for Smith’s murder and is ranting and raving in the dock. Brown says ‘The murderer of Smith is insane’, thereby expressing the idea that Jones is insane. The proposition thus conveyed is true whether or not Jones murdered Smith. Donnellan suggested that Russell’s quantiWcational account of deWnite descriptions might capture attributive uses, but that it does not seem right for referential uses. Kripke (1977) responded by arguing that a univocal, quantiWcational account of deWnite descriptions could account for both referential and attributive uses. The distinction between the two uses would thus be a purely pragmatic matter. In essence, Kripke’s argument goes like this. Assume that ‘the’ has a univocal, quantiWcational semantics. Then speech acts and acts of communication would still proceed in the way Donnellan says they do. Suppose, by hypothesis, that at the trial when Brown says ‘The murderer of Smith insane’ the literal meaning of the words is that exactly one person, x, is such that x murdered Smith and x is insane. Brown would still, in that context, succeed in communicating the proposition that Jones is insane.

Ludlow & Segal, On a Unitary Analysis

423

How does this work? The basic idea is simple. Brown speaks while Jones rants and raves in the dock, or shortly after. Following Grice’s (1975) maxim of quality we may suppose that Brown is attempting to say something true. He must, therefore, believe that exactly one person, x, murdered Smith and x is insane. He must have grounds for this belief. Now either these grounds are speciWc, based on beliefs about some particular individual believed to be the murderer, or they are not, in which case they would be based on general features of the case (such as the appalling condition of Smith’s body). But there is no evidence that Brown has any general grounds for the belief and good evidence that he has speciWc grounds—for someone accused of Smith’s murder is highly salient and acting strangely. Moreover, in the particular conversational context, an expression of a belief based on purely general grounds would not be relevant: general features of the case are not under discussion at the time. The audience may thus infer that Brown would not have said what he did unless he believed that Jones was insane. Moreover Brown knows all of the above and knows that his audience knows it, and so on. Therefore, Brown has communicated the proposition that Jones is insane. Kripke’s argument is that the univocal semantics plus independently motivated pragmatics suYces to account for the data. There is therefore no need to posit an ambiguity in ‘the’. Doing so would be redundant. Ludlow and Neale (1991) apply the same general considerations to the analysis of indeWnite descriptions. According to Ludlow and Neale, there are a number of possible uses to which we can put indeWnite descriptions, including referential uses, speciWc uses, deWnite uses, and purely existential uses. A referential use (as with deWnite descriptions) is a use in which an indeWnite is used to refer to some particular individual (e.g., if I say ‘A red-haired student in the front row is cheating on the exam’ when there is one and only one red-haired student visible in the front row). A speciWc use is one in which I intend to communicate that I have singular grounds for my utterance but I don’t intend my audience to identify the individual (as when, while proctoring an exam, I say ‘A student is attempting to copy his neighbors’ exams. Everyone please keep your work covered.’). A deWnite use is when I intend to communicate that exactly one individual satisWes the description, but I do not intend to communicate who the student is, or even that I know who it is (‘A student robbed the answer sheet from my oYce leaving his Wngerprints, but we haven’t identiWed the prints’). A purely existential4 use is simply one that corresponds directly to the literal meaning of the existential quantiWer (‘The exam scores are too high. A student must have circulated the stolen answer sheet’). As argued by Ludlow and Neale, all of these uses can be derived from the simple quantiWcational interpretation of the indeWnite. That is, semantically there is a unitary analysis of indeWnite descriptions, but following the Gricean considerations high4 Ludlow and Neale use the phrase ‘purely quantiWcational’, but that phrase is more general than they intend.

424

IV. Deﬁnites and Indeﬁnites in Semantic Theory

lighted by Kripke (1977) and Neale (1990), this single interpretation can be used in all of the ways just outlined. Of particular interest to us now is the idea that pragmatics can (and routinely does) deliver both referential and deWnite uses of indeWnite descriptions. What this means is that an indeWnite description can be put to all the uses that a Russellian deWnite description can. But if that is so, then why do we need Russellian deWnite descriptions at all? Answer: we don’t.

4. Extending the Gricean Strategy As noted above, our central idea is that ‘a’ and ‘the’ are two expressions with diVerent spellings but the same meanings—synonyms, rather like ‘gray’ and ‘grizzled’ or ‘grisly’ and ‘gruesome’. Literally ‘The F is G ’ and ‘An F is G ’ both express the same thing as ‘[9x: Fx](Gx)’. When we use ‘The F ’ referentially, it is via the same mechanisms that we use ‘[9x: Fx]’ referentially. More signiWcantly, we hold that even the uniqueness implication of ‘The F ’ is a matter of use, not what is literally said. The mechanisms that allow us to use ‘The F ’ deWnitely—i.e. to communicate that there is a unique satisWer of F—are similar to those that we employ in using ‘An F ’deWnitely. In the next section, we will provide some detail about the nature of these mechanisms. First, however, we have to get clear on the true nature of the diVerences between ‘a’ and ‘the’. Obviously there are important diVerences between our application of the terms ‘the’ and ‘a’. A student who did not know these diVerences would certainly have problems on his English language exams. But it does not follow that these diVerences in application are part of their semantical content. Indeed, many synonyms customarily are put to diVerent uses. Take, for example, Grice (1961, 1975) on the distinction between ‘but’ and ‘and’. On Grice’s story, ‘but’ and ‘and’ literally mean the same thing, but diVerent ‘conventional implicatures’ are associated with them; ‘but’ implicates a sense of contrast between the conjuncts. Similarly it is reasonable to hold that ‘unless’ has the same semantics as ‘or’ in its exclusive use, but implicates that the proposition expressed by the initial conjunct is the more likely or default hypothesis. We want to tell a similar story about ‘a’ and ‘the’. However, our story is not that the uniqueness implication of ‘the’ is part of its conventional implicature. We believe that the conventional implicature of ‘the’ is actually much weaker: it is simply that the noun-phrase with the determiner ‘the’ carries the conventional implication that the object under discussion is given in the conversational context. Noun-phrases fronted by the determiner ‘a’ are conventionally implicated to involve new information. The basic observation about new and given information is not novel, of course; many traditional studies of grammar make precisely this claim (see Christopherson 1939; Givon 1984), and the observation has not escaped the attention of lexicographers. Early on in the lengthy entry for

Ludlow & Segal, On a Unitary Analysis

425

‘The’ in the Shorter Oxford English Dictionary, we Wnd this entry: ‘the . . . Marking an object as before mentioned or already known or contextually particularized’. Now clearly, just as when we use ‘but’ in a conjunction we are not literally expressing a claim that the Wrst conjunct ordinarily contrasts with the second, we believe it is clear when we implicate that something is given information we are not explicitly saying: ‘this is given information’. It is simply something that competent users of English can infer from the conventional implicature inherent in ‘the’.5 Givenness is not the complete story, however. When the predicate by itself ensures uniqueness, the implicature gets overridden. If it is obvious that there can only be one F, then no issue can arise about singling out a speciWc F as already mentioned or contextually particularized. Consider, for example, superlatives (‘the tallest man in the room’) and titles (‘The President of the United States’). There is no diVerence in truth conditions between ‘[ix: x is the tallest man in the room](Gx)’ and ‘[9x: x tallest man in the room](Gx)’. The same point holds of titles. Given that there is, by deWnition, only one President of the United States, the choice of the logical determiner is utterly irrelevant to the truth conditions of the resulting sentence: ‘[ix: x is the POTUS](Gx)’ has the same truth conditions as ‘[9x: x is the POTUS](Gx)’. In these cases, uniqueness is neither implicated, nor implied by the type of determiner. Rather it simply results from the descriptive material itself.6

5. The Formal Derivation So far we have been describing in an informal way the Gricean mechanisms that give rise to uniqueness implications. We now want go into more detail about the nature of these mechanisms, and show how a uniqueness implication might be explicitly derived. We will begin with an illustration, drawn from Neale (1990), of how such Gricean derivations work, and then we will proceed to give derivations for both the deWnite and

5 We are not wedded to the idea that this additional information be accounted for in terms of conventional implicature. We are aware that the notion of conventional implicature has fallen into disfavor in some quarters (see, e.g. Sperber and Wilson 1986/1995: 182; and Bach 1999b), and we note that all of what we say here can be recast in terms of explicature or inference or by one’s favorite account of how the new/given distinction is to be understood. Our key point is that the new/given distinction (however understood) plus pragmatic resources of some form will get us to deWniteness. Anne Bezuidenhout has observed that it might be possible to build the new/given information into the semantics of the determiner. In principle we do not have a problem with that move either—it would mean that ‘the’ and ‘a’ do diVer in meaning, but not in the way we ordinarily suppose. 6 Other cases where the givenness implications fail are the cases cited in examples (a)–(h) above. It is worth noting that many of these examples involve possessive notions of some form (the eye ¼ his eye, the living-room wall ¼ the living room’s wall, the bank ¼ the river’s bank). One attractive proposal would be that sometimes the determiner signals inalienable possession of some form rather than givenness. As noted above, givenness is certainly not the whole story here.

426

IV. Deﬁnites and Indeﬁnites in Semantic Theory

referential uses of indeWnite descriptions. All these derivations will Wt the following template. S has expressed the proposition that p. There is no reason to suppose that S is not observing the Gricean cooperative principle (CP) and maxims of communication. (c) S could not be doing this unless he thought that q. (d ) S knows (and knows that I know that he knows) that I can see that [he thinks]7 the supposition that he thinks that q is required. (e) S has done nothing to stop me thinking that q. ( f ) S intends me to think, or is at least willing to allow me to think, that q. (g) And so, S has implicated that q. (a) (b)

Neale’s example of how this works involves a case where the speaker and hearer both know that Harry Smith is the chairman of the Flat Earth Society. The idea is that the speaker (S) can use the deWnite description ‘The Chairman of the Flat Earth Society’ to refer to Harry Smith. The derivation goes as follows: (a) (b) (c)

S has expressed the proposition that [ix: Fx](Gx). There is no reason to suppose that S is not observing the CP and maxims. S could not be doing this unless he thought that Gb (where ‘b’ is a name).

Gloss: On the assumption that S is observing the maxim of relation, he must be attempting to convey something beyond the general proposition that whoever is uniquely F is G. On the assumption that S is adhering to the maxim of quality, he must have adequate evidence for thinking that the F is G. I know S knows that b is the F, therefore S thinks that Gb. S knows (and knows that I know that he knows) that I know that b is the F, that I know that S knows that b is the F, and that I can see that he thinks the supposition that he thinks that Gb is required. (e) S has done nothing to stop me thinking that Gb. ( f ) S intends me to think, or is at least willing to allow me to think, that Gb. (g) And so, S has implicated that Gb. (d )

In the case where we do not refer to an individual known by name, but rather to someone in the perceptual environment, we replace step c of the derivation with c 0 . (c 0 )

S could not be doing this unless he thought that Gb (where ‘b’ is a demonstrative that refers to an individual salient in the environment).

Gloss: On the assumption that S is observing the maxim of relation, he must be attempting to convey something beyond the general proposition that whoever is uniquely F is G. On the assumption that S is adhering to the maxim of quality, he must 7 The additional ‘he thinks’ in square brackets is not in Grice’s original formulation, but is added by Neale.

Ludlow & Segal, On a Unitary Analysis

427

have adequate evidence for thinking that the F is G. It is not plausible to suppose that he has just general grounds for this belief, therefore he must have object-dependent grounds. I can see that there is someone in the perceptual environment who could be taken to satisfy the description ‘the F ’, and I can see that S can see this. Therefore the grounds for his assertion that the F is G are plausibly furnished by the belief that Gb. Crucially, we can also see that a similar derivation is possible for a deWnite use of an existentially quantiWed expression—that is, the use of an expression of the form [9x: Fx](Gx) to communicate that there is a unique satisWer of the description. Let us return to the case in which S comes upon Smith’s badly mutilated corpse and says ‘The murderer of Smith is insane’ and consider how the addressee might derive the implicature of uniqueness. (a ) S has expressed the proposition that [9x: Fx](Gx). (b) There is no reason to suppose that S is not observing the CP and maxims. (c ) S could not be doing this unless he thought that [ix: Fx](Gx). Gloss: By invoking the determiner ‘the’, S intends to communicate that whatever F or Fs he is talking about is/are given in the conversational context. By refraining from using a plural noun, S intends to communicate that just one F is given in the conversational context. If there were more than one F given in the context, S would have used the plural deWnite description (otherwise S would Xout the maxim of quantity). But the only way a unique F could be given in the context, is if there is just one F in the domain of discourse as Wxed by the current context. S knows this, and so must be assuming that there is exactly one F in the domain of discourse. Therefore the grounds for S’s literal assertion that [9x: Fx](Gx) are plausibly furnished by the belief that [ix: Fx](Gx). (d )

(e ) ( f ) (g )

S knows (and knows that I know that he knows) that I know that he thinks there is only one F and that I can see that S thinks the supposition that he thinks that [ix: Fx](Gx) is required. S has done nothing to stop me thinking that [ix: Fx](Gx). S intends me to think, or is at least willing to allow me to think, that [ix: Fx](Gx). And so, S has implicated that [ix: Fx](Gx).

We conclude that there are no formal obstacles to the analysis of descriptions like ‘the F ’ as mere existentially quantiWed noun-phrases. As we will see in the next section, there are good philosophical reasons for supposing that this is in fact the correct analysis.

6. Positive Evidence If we pursue the path outlined above, then a number of familiar puzzles about deWnite descriptions begin to dissolve. These puzzles range from the problem of misdescription

428

IV. Deﬁnites and Indeﬁnites in Semantic Theory

to the behavior of ‘the’ in conditionals to the general detachability of the uniqueness implications of ‘the’. In this section we will canvass some of these familiar problems with the standard Russellian view and show how an existential analysis of ‘the’ solves those problems.

The Argument from Misdescription As Neale (1990: 91–3) observes, one of the great advantages of the Gricean distinction between the proposition said and the proposition meant, is that it oVers an account for our being in two minds about certain famous ‘misdescription’ cases. For example, in Donnellan’s case of Smith’s murderer discussed above, I might say ‘Smith’s murderer is insane’, and still say something true even if the crazy man at the defense table is entirely innocent of the charges and the actual murderer, who is miles from the scene, is quite sane. Of course, there is also some pull to say that in such a case one is saying something false as well. As Neale observes, this is simply a case where what we literally said was false, but what we intended to communicate—the proposition meant—was true. The two-level theory thus accounts for our conXicting intuitions.8 Neale observes that things work in the opposite direction as well. Hornsby (1977) gives the case of my observing the man ranting at the defense table and saying, ‘The murderer of Smith is insane’, not realizing that the man at the table is (i) innocent and (ii) quite sane, while the actual murderer is at large and certiWably insane. Again we are pulled in two directions about the truth of what I say, and again Neale observes that the distinction between the proposition meant and the proposition expressed allows us to explain why. In this case, the proposition expressed is literally true, but what I intended to communicate is mistaken. The problem is that there are other cases of misdescription which remain problematic for the classical Russellian analysis of descriptions—in particular for attributive uses. Suppose, for example, that at the scene of the crime Detective Watson says ‘The murderer of Smith is insane, whoever he is’. But now suppose that there is not one murderer, but two and that one or both of them is insane. Has Watson said something true or false? Again, we seem to be in two minds about this. There is a clear sense in which Watson is communicating something true, but there is also a sense in which his remarks are mistaken. If we suppose that the English determiner ‘the’ actually has the semantic content of the existential quantiWer, then our conXicting intuitions have a ready explanation. What Watson literally said was true; there is in fact a murderer of Smith who is insane. But what Watson was communicating to his audience—that there is only one such murderer—is mistaken. 8 Subsequent work by Neale (1999) and a number of others has gone on to propose that two distinct propositions are being expressed. Everything we say here can be recast under that assumption.

Ludlow & Segal, On a Unitary Analysis

429

The distinction between the proposition meant and the proposition expressed is a crucial part of the solution, but it cannot be the whole solution. Solving the problem of misdescription also requires that we give up the idea that uniqueness is part of the meaning of ‘the’. When Gricean mechanisms are combined with an existential analysis of the deWnite determiner, the residue of the argument from misdescription Wnally dissolves away.

Conditionals Another class of constructions that has been a nagging problem for the classical Russellian analysis of ‘the’ has been its behavior in many conditional environments. The problem, very simply, is that the determiner fails to have the expected uniqueness implications in these environments. Consider the following examples from Ludlow (1994) and Moltmann (1996).9 (1) (2) (3) (4)

If a doctor and a lawyer enter the room, the doctor will sit down Wrst. If anyone here has a dog, he has to register the dog. If anyone has a dime, he should put the dime in the meter. If a bishop meets another bishop, the bishop blesses the other bishop.

Obviously these sentences do not imply that there is a unique doctor, dog, dime, etc. What can account for this lack of uniqueness—particularly if uniqueness is supposed to be part of the meaning of the determiner? As a Wrst stab, one might suppose that on a Lewis-type analysis of conditionals, the description will be relativized to a world, so that we will not be committed to the absurd view that these sentences entail that there is only one doctor, dog, dime, etc. This move helps a little, but now the problem of incomplete descriptions returns with a vengeance.10 Since each nearby world is going to have several doctors, dogs, dimes, etc., there is no unique doctor, dog, etc. to be picked out by the description. But the situation is even worse than standard solutions to the problem of incomplete descriptions—those which hold either (i) that the unique object can be picked out by Xeshing out the description or (ii) that the unique object is picked out by restricting context.11 Regarding strategy (ii), what sense does it make to talk of restricting the 9 The Wrst, from Ludlow (1994) is attributed to Mark Johnson. The others, from Moltmann (1996) are attributed to Robert Fiengo. As Moltmann notes, similar phenomena also arise in so-called donkey-anaphora sentences. Consider: ‘Everyone who has a dog has to register the dog’. 10 The problem, as initially brought to attention by Strawson (1950) is that if I say ‘the table is covered with books’, I do not mean to be suggesting that there is only one table in the world, but that seems to be what the Russellian theory of descriptions is literally committed to. As Neale (1990) explains, two strategies have emerged for trying to answer this objection. The Wrst strategy (adopted, e.g. by Sellars 1954) is that the unique object can be picked out by using context to Xesh out the description. The second strategy is to say that the unique object is picked out by allowing the context to restrict the domain of quantiWcation. See Lepore (this volume), Stanley and Szabo´ (2000a), Neale (2000) for a sample of the discussion. 11 Both strategies face severe diYculties. The Wrst one needs an account of what determines the features of context that complete description and how the hearer is supposed to discover them. For example, suppose that

430

IV. Deﬁnites and Indeﬁnites in Semantic Theory

context in other worlds? Even if we could make sense of this idea, example (4) soon gives rise to new problems. What restriction on context is simultaneously going to include the event of a bishop blessing another bishop and allow that there is only one bishop in the speciWed context? This is like having a context in which we eat cucumber sandwiches, but in which there are no sandwiches. Regarding strategy (i), there is simply no way to suYciently Xesh out the description, even employing the device of packing demonstratives into the description. Why? Because we crucially understand that any further descriptive facts concerning the two bishops, for example, are irrelevant to the truth of what we say. The bishops can be tall, short, Italian, or African, and they can certainly be well out of range of our demonstrative acts; it simply does not matter. Indeed it cannot matter, since in the case of conditionals we want our claim to cover all bishops—indeed, all possible bishops. These considerations do not turn on the speciWc theory of conditionals that we choose to employ. For example, if we adopt an event-based analysis of conditionals in the spirit of Lycan (1984) and Kratzer (1989), we fare no better. (A version of this idea with application to descriptive pronouns is explored by Heim 1990a; Ludlow 1994; Moltmann 1996.) Applied to this case, the basic idea would be to try and give an event/ situation-based analysis of conditionals so that the deWnite descriptions with their uniqueness implications are always relativized to an event or situation. So e.g. in the case of (1) the gloss would be that ‘for every event in which a doctor and a lawyer enter the room there is a related event in which the unique doctor in that event will sit down Wrst’. As noted by Heim and Ludlow, this solution gives rise to a number of puzzles that need to be confronted in the case of descriptive pronouns, and Moltmann argues that it fails altogether for both deWnite determiners and descriptive pronouns.12 Smith enters a room, leaving the door open, and Brown comments: ‘You left the door open’. There are indeWnitely many possible completers for the description e.g. ‘the door/that I am now staring at/through which you just entered/to this room/over there/’. It is diYcult to see what would make one of these, rather than any other, the real completer. Moreover, if Smith is to understand what Brown is saying, then he must know what the completer is. It is hard to see how he could know this (for discussion see Wettstein 1981; Larson and Segal 1995.) The problem with the second strategy is that it requires an account of some rather special mechanisms of domain restriction. Consider the following example. We are watching a basketball game and a particular player, P, miraculously shakes oV his defender, shoots, and scores. I say ‘The man is a genius’. On the face of it, a number of men are in the domain of discourse. On the hypothesis that ‘the’ is a Russellian quantiWer, the domain has shrunk to include just P and exclude all the other men. Compare the case in which we are watching a basketball game and several players cooperate in a masterful play. I cannot simply say ‘Every man is a genius’ and thereby felicitously communicate that every man who was involved in the masterful play is a genius. Nor can I say ‘Most men are geniuses’ to communicate e.g. the proposition that most men on the team that just scored are geniuses. These cases indicate that the proposed domain-shrinking mechanisms that account for incomplete descriptions are not the same as those that operate with other quantiWers, like ‘most’ and ‘all’. Until some account of these special mechanisms is oVered, the proposal remains merely promisory. 12 For example, one problem revolves around example (4). Very simply, no matter how Wnely we carve up events, we still need to be left with two bishops in the event in question. But then there simply cannot be a unique bishop in the event. (Heim attributes the observation of this problem to Hans Kamp and Jan van Eijk.) Ludlow proposes a possible solution that would allow the introduction of thematic roles into the actual descriptive content of the noun phrases, so that one would get ‘the bishop theta-1’ and ‘the bishop theta-2’. The only problem is that the usual theta-roles (agent, patient, theme) do not seem to be suYcient for distinguishing bishops that stand in the meeting relation (or any other symmetrical relation). One can introduce theta-role

Ludlow & Segal, On a Unitary Analysis

431

Perhaps it is better to just cut the Gordian knot and reject outright the idea that the uniqueness clause is part of the meaning of ‘the’.

Possessives Possessives, like ‘my F ’ and ‘your F ’ provide further evidence for the existential analysis of ‘the’. Since Russell (1905) these have typically been treated as equivalent to ‘The F of mine’, ‘the F of yours’, etc., and this treatment is vindicated by the explicit use of the deWnite article in possessive constructions in many languages: for example, Italian ‘Il mio libro’ (‘the my book’). It seems to us, however, that these uses are more correctly analysed as being equivalent to [9x: x is an F of . . . ] (e.g. [9x: x is a book of mine]). Obviously, if I say ‘This is my book’, or ‘She is my student’, or ‘You stole my idea’, there is not even the illusion of uniqueness. So why should we saddle our analysis of possessives with a clearly incorrect uniqueness clause? It is of course possible to contend that this merely shows possessives to be indeWnite descriptions and that it has nothing to do with the proper analysis of the English determiner ‘the’, but this contention strikes us as disingenuous. Languages with surface elements which are candidates for being genuine deWnite determiners are far and few between. In addition to English, the Romance languages oVer some candidates like the Italian ‘il’, ‘la’, ‘lo’ and the French ‘le’ and ‘la’. The problem is that in Italian these elements appear explicitly in possessive constructions (as in ‘il mio libro’—‘the my book’) yet carry no uniqueness implications. This fact undermines the plausibility of Italian determiners being genuine Russellian deWnite determiners. One can of course dismiss the Italian determiners and argue that the English ‘the’ is diVerent, but then the defense of the Russellian nature of English ‘the’ looks like an ever lonelier task.

The ‘DeWniteness EVect’ At Wrst blush the ‘deWniteness eVect’ appears to be a problem for the unitary analysis of descriptions. We concur with Szabo´ (2000) that it is not problematic, and we would go

labels arbitrarily, but without some semantical content to those labels the proposal teeters on the brink of vacuousness. Another problem with the relativization-to-event strategy involves an extension of the ‘sage plant’ sentences discussed in Heim (1982) (Heim credits Barbara Partee with the original sage plant examples; Moltmann credits the observation about sage plant sentences with explicit descriptions to Bob Fiengo): ‘If a man bought a sage plant, he bought two others along with the sage plant.’ Suppose we take the description to be relativized to events, so we can gloss this as ‘for every event in which a man bought a sage plant, there is a related event in which he bought two others along with the sage plant in that related event’. The problem is that the consequent event has to contain at least three sage plants. Notice also that it will not help to say that ‘the sage plant’ is elliptical for ‘the sage plant he bought’ since by hypothesis there are three sage plants that he bought. Other maneuvers are possible (e.g. ‘the sage plant in question’ or the ‘sage plant in the antecedent event’) but how are the unique denotations of these descriptions to be identiWed? The problem is parallel to distinguishing the two bishops in (4).

432

IV. Deﬁnites and Indeﬁnites in Semantic Theory

further and argue that, once properly understood, it provides more evidence for the unitary analysis. The apparent problem is that deWnite determiners are prohibited from being inserted in certain grammatical environments, where indeWnite determiners are permitted. The paradigm, Wrst observed in Milsark (1974) involves so-called there-insertion contexts: The basic observation from Milsark is that there-insertion is only possible with a particular class of determiners. So, for example, one gets the following distribution of facts. (5) (6) (7) (8) (9) (10)

There’s a fox in the henhouse. There’s the fox in the henhouse. There are three foxes in the henhouse. There’s every fox in the henhouse. There are no foxes in the henhouse. There is every fox in the henhouse.

The crucial facts for our purposes concern (5) and (6). It appears that indeWnite descriptions can be inserted in these environments and that deWnite descriptions cannot, and it is supposed that this is a reXex of something semantical. A number of attempts have been made to give a semantical account of this phenomenon (see e.g. Barwise and Cooper 1981, Keenan and Stavi 1986, Higginbotham 1987) but none of these attempts have been entirely satisfactory.13 This failure should be no surprise if the considerations adduced above are correct. The eVect cannot be semantical in character because there is no semantical diVerence between deWnite and indeWnite descriptions. There is, in fact, compelling evidence to support the contention that this is not a semantical eVect. For example, a description with a uniqueness clause can be inserted in these environments. (11)

There’s exactly one fox in the hen house.

This shows that it cannot be the uniqueness clause in ‘the’ that is blocking thereinsertion. One natural explanation for the distribution of there-insertion facts is that they are a reXection of pragmatics rather than semantics; perhaps (as suggested in McNally 1992, 1998; Zucci 1995; Ward and Birner 1995; and Lumsden 1988) it is the implication of givenness that is blocking there-insertion here (e.g. since ‘there is/are’ is a device typically used to implicate that new information is being introduced there is a kind of pragmatic clash when ‘there is’ signals new information and ‘the’ signals old information). This explanation is further supported by facts like the following: (12)

There are none of the foxes in the hen house. 13 See Ludlow (1991) for discussion.

Ludlow & Segal, On a Unitary Analysis (13)

433

There are none of the foxes in the universe of discourse in the henhouse.

In the Wrst example ‘none of the foxes’ suggests that some foxes are already under discussion, which clashes with the implication of ‘there are’ that new information is being introduced. The second example shows a case where this givenness implication of the noun-phrase is defeated by the introduction of talk of the full universe of discourse, and hence, as predicted, there-insertion is possible despite our using the determiner ‘the’. The upshot is that by unifying our semantical analysis of ‘the’ and ‘a’ we are able to see why attempts to give a semantical account of the diVerences in behavior between ‘the’ and ‘a’ consistently founder. It also steers us in the apparently more fruitful direction of investigating these facts as reXexes of pragmatics alone.14

Predicational Uses of DeWnite Descriptions One of the central observations about indeWnite descriptions is that they have predicational in additional to quantiWcational uses. For example, it has been observed that ‘a lawyer’ in the following example appears to be a simple predicate rather than a quantiWed noun-phrase. (14)

John is a lawyer.

It has also been argued that ‘the lawyer’does not have this sort of use. (15)

#John is the lawyer.

So there is supposed to be an asymmetry between these two cases. If there is such an asymmetry, and if it is a semantic asymmetry, then it appears we have a reason for distinguishing ‘the’ and ‘a’ as being semantically distinct. But is there a semantic asymmetry? We think not. In the Wrst place, one can introduce a semantic equivalent of a Russellian description in a predicational environment and have a perfectly grammatical sentence. (16)

John is the unique satisWer of the predicate ‘lawyer’.

As Smiley (1981) has observed, there are many cases in which descriptions of the form ‘the F ’, when in object position, routinely behave like predicates.15

14 But see Abbott (1999) for criticism of the givenness account of there-insertion contexts. Abbott’s positive proposal is that these sentences are blocked because the combination of an existence claim with the existential presupposition of the deWnite determiner creates a ‘conversational tension’—to wit: ‘it is not felicitous to assert something while presupposing it’. This story strikes us as untenable since one can perfectly well say ‘The mayor is a mayor’, and while it may be uninteresting it is not ungrammatical in the way that (6) is. Abbott’s account would also seem to falsely predict that (11) is ungrammatical, since ‘exactly one’ carries the same existential presupposition (or same existential entailment) that a deWnite description does. 15 For an interesting discussion and defense of the thesis that deWnite descriptions are always predicational, see GraV (2001).

434

IV. Deﬁnites and Indeﬁnites in Semantic Theory (17) (18) (19)

John is becoming the biggest jerk in town. John was the mayor of Boston. John is the lawyer that I was talking about.

If one opts for a semantical asymmetry between deWnite and indeWnite descriptions then these facts are diYcult to explain.16 But if deWnite and indeWnite descriptions diVer pragmatically then we would expect there to be a pragmatic account of their diVerent behaviour in predicate phrases. And such an account is not hard to Wnd. Earlier we noted that it is a conventional implicature of the use of ‘the’ that it apply to given information. Anomalous cases, such as ‘John is the lawyer’ occur when the implicature of givenness is present but unfulWlled, that is, when no F (no lawyer) is given. Felicitous cases are those where either the implicature is cancelled (16, 17, and 18), or it is not cancelled but is fulWlled (19, 20, and 21): (20) (21)

a. I heard that Smith oVended a lawyer in court yesterday. b. Yes, unfortunately John is the lawyer. a. I heard that Barry has joined your legal team as an expert in constitutional law. b. No, Barry is the DNA expert.

Of course it might be argued that these cases with ‘the’ are not really cases of predication, but can be assimilated to identity statements: (22)

[the x: lawyer x] (John ¼ x)

But notice that the same can be said for cases involving ‘a lawyer’. (23)

[an x: lawyer x] (John ¼ x)

Our interest here is not in defending the predicational theory of determiner phrases, but simply to point out that there is no interesting semantical asymmetry between the determiners as regards such environments. Or rather, that a number of philosophical puzzles can be avoided if we reject such asymmetries.

A Note on Focused Descriptions Abbott (1999) has argued that certain focus facts involving deWnite descriptions tell against their analysis as being merely quantiWcational. Consider the following sentence. (24)

That wasn’t a reason I left Pittsburgh, it was the reason.

16 See Enc¸ (1986) for an attempt to explain these facts by positing implicit temporal indices within the descriptions (e.g. ‘the Mayor of Boston’ might be understood as ‘the Mayor of Boston at time t’). This is certainly a possible gambit, but without serious changes it weds us to a semantics of tense and metaphysics of time that at least one author of this chapter would reject for independent reasons. See Ludlow (1999).

Ludlow & Segal, On a Unitary Analysis

435

According to Abbot, an utterance of (24), focused as indicated, implies that the speaker has only one reason for leaving Pittsburgh, not many. Now the idea is that this supports the uniqueness analysis of deWnite descriptions over the givenness analysis of deWnite descriptions. Stressing ‘the’ allows us to focus the contrast in meaning between ‘the’ and ‘a’ and our judgment is supposed to be that the contrast in these cases has to do with whether the description is unique or non-unique rather than whether it is given or novel. In eVect, focus becomes a tool for probing the real diVerence in meaning in these cases and the evidence allegedly supports the thesis that the relevant diVerence in meaning is along the unique/non-unique scale. Whatever might be said about the logic of this argument, we disagree with Abbott’s judgment of what is happening in these cases. We take it that someone may very well utter (24) with stress as indicated despite having several reasons for leaving Pittsburgh. Stressing ‘the’ indicates that the reason in question was not merely one of many reasons for leaving but rather the causally determinate reason—the big reason. In our view this is typically the case when the deWnite determiner is stressed. Consider the following example. Ludlow’s third grade teacher had a husband named William Faulkner. On vacation in the South, he was asked, ‘Are you the William Faulkner?’ Presumably, the questioner was not asking if he was the unique individual named William Faulkner (at which point he was, since the author William Faulkner had long since passed on), but was asking whether this individual was the famous—i.e. given or familiar—William Faulkner. A similar incident involves Ludlow’s father, Wrst name Robert. Being a sometime resident of Naples, Florida, he was often confused with the author Robert Ludlum (who lived in Naples and is only recently deceased). Persons would typically ask ‘Are you the Robert Ludlow?’ But presumably they were not asking whether he was the only Robert Ludlow in town (which he was), but whether he was identical to the famous— i.e. given or familiar—individual who was in fact named Robert Ludlum. A Wnal note: Abbott’s paper includes a cartoon by Dick Guindon of the Detroit Free Press in which a precocious boy confronting a shopping mall Santa Claus says ‘Not the Santa Claus’. Abbott seems to think the joke is that the boy is saying this is just one Santa out of many—he’s merely a Santa Claus. We see the joke diVerently. Isn’t it actually the case that the boy is saying this isn’t the real Santa Claus? That is, this isn’t the Santa that he’s read about in books and that is reputed to bring him toys? Not the given or familiar Santa?

7. Conclusion We think the arguments marshaled in section 6 give strong support to the idea that ‘the’ and ‘a’ have the same literal meaning, yet a number of philosophers that we have spoken to Wnd the very idea to be utterly crazy. But why? Ordinarily, analytic philoso-

436

IV. Deﬁnites and Indeﬁnites in Semantic Theory

phers like to go where arguments take them, particularly if the logic of those arguments is not only familiar, but (as in this case) routinely employed. ‘The’ and ‘a’ are regarded diVerently, however. Russell’s analyses of these constructions have been at the very center of analytic philosophy for nearly 100 years, and while many have disputed Russell’s analysis of deWnite and indeWnite descriptions, few have doubted that he was correct to give them separate analyses. This tradition is not enough to save their bifurcated treatment, however. As we saw in section 6, clinging doggedly to this tradition has led to legions of philosophical puzzles—puzzles that dissolve when the traditional view is abandoned. Linguistic myopia cannot be discounted here. All too often we opt for the logical analysis that seems to Wt the surface form of the language in which we work (i.e. English). But both Russell and Frege cautioned against this kind of slavishness to surface form. In this particular case, deference to surface form not only lands us in the midst of philosophical conundrums, but it lends a procrustean quality to our analyses when we try to extend them to the other languages of the world (in this case Slavic languages, Chinese, Japanese, etc.). If we are interested in the logical form of natural language (as opposed to the logical form of English only) we need analyses that ‘travel well’. We cannot be satisWed that our analysis works for our own language if it fails as an analysis of most other languages in the world. Accordingly, we urge that further consideration be paid to the unitary analysis.17 17 One topic we have not addressed here but which would certainly play a role in future investigations is the treatment of plural deWnite descriptions. If this analysis were extended to such constructions then the idea would be that a plural description like ‘the dogs’ would have the literal meaning of ‘some dogs’, but may be used to express something about all the dogs. Prima facie there is much to recommend such an analysis for plural descriptions. If I say that the dogs are barking tonight, I certainly do not mean that all of them are barking (and it is clear that this is not a generic usage—I may be talking only about dogs in the neighborhood). I mean that some dogs are barking. Less clear is what one would say about ‘both’ (this example was brought to our attention by John Hawthorne). Clearly ‘both dogs’does not seem to mean ‘some of the two dogs’. Here we think there is a genuine numerical claim packed into the semantics (‘both dogs are barking’ must mean that two given dogs are barking) just as when we say ‘the only dog is barking’. In this respect, English ‘both’ may pattern with the Italian expression for ‘both’—‘tutte e due’ (‘all and two’). Szabo´ (2000) considers this possibility and rejects it on the grounds that no single lexical item can do what ‘all and two’ can. Once again, our point is that these and similar cases deserve further study.

PA RT V

Anaphoric Pronouns and Descriptions, IndeWnites, and Dynamic Semantics/Syntax

The chapters in this section are concerned with issues having to do with the interpretation of indeWnites and the pronouns and deWnite descriptions that are anaphoric on these indeWnite expressions. Another theme concerns the scope taking possibilities of indeWnites, and the parallels between this phenomenon and the phenomenon of anaphoric pronoun resolution. Several of the chapters presuppose some understanding of dynamic analyses of indeWnites and anaphora, such as those oVered by Discourse Representation Theory, DRT (Kamp 1981; Kamp and Reyle 1993) and File Change Semantics, FCS (Heim 1982, 1983a).1 They also presuppose some knowledge of non-dynamic E-type (Evans 1977, 1980; Heim 1990a) and D-type (Neale 1990) approaches to dealing with these phenomena. The authors of these chapters are aware that not all readers will have this specialized knowledge, and so they have sketched in some of the needed background. We will add in some more of this background here.2 As we will see, the principal issue is whether the phenomena of concern can be dealt with using standard semantic theory supplemented by a pragmatic theory of language use, or whether they require the introduction of a new semantic framework (see again n. 1). 1 Geurts (1999) argues that DRT and FCS are actually very diVerent endeavors, despite the fact these two theories are generally treated as notational variants of the same underlying dynamic semantic approach (e.g. by Chierchia 1995; Kadmon 2001). Geurts suggests that DRT is an idealized theory of language understanding/ interpretation, and not a dynamic semantic theory. In dynamic semantics processing facts are encoded in the lexicon. One consequence is that the conjunction ‘and’ is interpreted as non-commutative, unlike the ‘and’ in standard (non-dynamic) logic. Geurts argues that DRT assumes the standard interpretation of conjunction, negation, and so on. The fact that conjunctions are sometimes understood as non-reversible can be explained by the fact that interpretation is a left-to-right process in which the second conjunct is processed in a context in which the Wrst conjunct has already been processed. We will return to this issue in our discussion of Breheny’s and Dekker’s contributions to this volume. 2 For a very accessible introduction to DRT and FCS we recommend Kadmon (2001). For comparisons between DRTand E-type approaches see Chierchia (1995) and Heim (1990a). For arguments in favor of the Dtype approach as well as an accessible introduction to Evans’s E-type approach, see Neale (1990).

438

V. Anaphoric Pronouns and Dynamic Semantics

The subject of indeWnite descriptions (and indeWnite NPs more generally) has been of concern to contemporary linguists and philosophers of language at least since Geach (1962). An issue that has been of particular concern is the interaction of indeWnites with anaphoric pronouns. One puzzle arises if indeWnite NPs of the form ‘an F’ are analyzed as existential quantiWer expressions of the form ‘9xF (x)’. The puzzle concerns the fact that indeWnites can act as antecedents for pronouns beyond the boundaries of the clauses and sentences in which the indeWnites occur. Yet this reach beyond sentence and clause boundaries is not possible for quantiWer expressions more generally. For instance, compare the following: (1) (2)

The teacher gave a boy i a treat. Hei was pleased. The teacher gave every boyi a treat. Hei was pleased.3

The universally quantiWed NP ‘every boy’ in (2) cannot act as the antecedent for the pronoun ‘he’ in the subsequent sentence, whereas the indeWnite ‘a boy’ in (1) does not seem to be subject to the same restriction. IndeWnites also behave diVerently in coordinated structures, as can be seen by comparing (3) and (4) below: (3) (4) (5) (6)

Mrs Goldsmith gave a boy i a treat, and Mrs Cooke gave himi one too. Mrs Goldsmith gave every boy i a treat, and Mrs Cooke gave himi one too. The teacher gave every boy a treat. They were pleased. No new student came to the party. They didn’t know about it.

If plural pronouns are used instead of singular ones in (2) and (4), these sentences become acceptable, as is shown, for example, by (5). However, plural anaphora is tricky, and appears to introduce pragmatic factors beyond those introduced by the special behavior of indeWnites. Chierchia (1995) explicitly denies that cases of plural anaphora should be assimilated to cases such as (1) or to cases of donkey anaphora (see (7) and (9) below), where the indeWnite seems to bind the pronoun across the sentence boundary. Chierchia oVers a dynamic binding analysis for cases such as (1) and plain donkey anaphora, but he oVers a pragmatic explanation for cases of plural anaphora. Consider, for example, the plural anaphoric pronoun in (6). Here we do not want to say that the quantiWer in the Wrst sentence has the potential to bind the plural pronoun in the second sentence, for then we get an interpretation of (6) according to which no 3 Here and in the examples below we have used indices to indicate the intended anaphoric relations. These indices are for expository purposes only. It is not assumed that they are part of the syntax of the sentences in question. Note that this example is problematic only if ‘he’ is treated as anaphoric on ‘every boy’. If ‘he’ is treated as a deictic pronoun referring to some contextually salient individual, then (2) is felicitous. Sentences like (2) are sometimes used to create a humorous eVect. E.g. ‘Every Clinton supporter was at the rally. He was very lonely.’ There is also a phenomenon, known as ‘telescoping’, where pronouns can be used felicitously with a universal quantiWer as antecedent. Consider: ‘Every soldier got ready for the battle. He cleaned his riXe. He checked his water canteen. He wrote a letter home.’ The pronouns here are best given a generic interpretation. They stand for a representative soldier. See Roberts (1990a, b) for a discussion of these cases. Carminati et al. (2002) report some interesting experimental investigations into the phenomenon of telescoping.

V. Anaphoric Pronouns and Dynamic Semantics

439

new student x is such that x came to the party and x didn’t know about the party (viz. every new student who came to the party knew about it). But that is clearly not how (6) is understood. Chierchia argues instead that in (6) the use of the quantiWer phrase ‘no new student’ makes a set of entities salient, viz. the set of new students. These salient entities can then be referred to by the subsequent pronoun ‘them’.4 Another class of puzzling sentences involving indeWnites is the class of so-called donkey sentences, illustrated in (7) and (9) below: (7) (8) (9) (10) (11) (12)

If Pedro owns a donkey i he beats iti . If Pedro owns no donkey i he wants iti . Every man who owns a donkey i beats iti with a stick. Every man who owns every donkey i beats iti with a stick. Usually, if a mani owns a donkey j hei beats itj . Usually, if a mani owns every donkey j hei beats itj .

IndeWnites in the antecedents of conditionals are accessible to pronouns in the consequents of these conditionals. This is not possible for other quantiWers, as is shown in (8). Also, indeWnites but not other quantiWcational expressions can act as antecedents for pronouns outside of the relative clauses in which they occur, as shown in (9) and (10). Finally, indeWnites behave diVerently from other quantiWcational expressions in the presence of quantiWcational adverbs, as is shown in (11) and (12). The issue of indeWnites and pronominal anaphora was extensively discussed in the 1980s, and various suggestions for dealing with these phenomena emerged. One approach was the dynamic analysis oVered by Kamp (1981).5 According to Kamp’s Discourse Representation Theory (DRT), indeWnites have no quantiWcational force of their own. They are associated with free variables. In context, they will have some quantiWcational force, but this is inherited from determiners (such as ‘every’, ‘no’, ‘most’) or from quantiWcational adverbs (such as ‘always’, ‘usually’, ‘sometimes’). The determiner or Q-adverb from which an indeWnite inherits its quantiWcational force is the lowest c-commanding determiner or adverb that is not itself an indeWnite. (A c-commands B iVA does not dominate B, but the Wrst branching node that dominates A 4 As Roberts (this volume) notes, when the grammatical number of the quantiWer and the pronoun do not match, this is a sure sign that something pragmatic is going on. Since ‘no’ and ‘every’ are singular and ‘they’ is plural, the anaphoric pronouns in (5) and (6) seem to invite a pragmatic explanation. That something diVerent is going on in the case of plural anaphora is not always signaled by a clash in grammatical number. Evans (1977) discusses an example of plural anaphora involving ‘some’, which can take a plural subject. Consider ‘John has some donkeys and Harry vaccinates them.’ We do not want to analyze this as a case where the indeWnite ‘some’ binds the plural pronoun. This would give us ‘There are some donkeys x such that John owns x and Harry vaccinates x’, which is compatible with John’s having some donkeys that Harry doesn’t vaccinate, whereas the original English sentence suggests that Harry vaccinates all of John’s donkeys. 5 What follows is a presentation of DRT based on Chierchia (1995). As Bart Geurts (pers. comm.) has pointed out, what Chierchia calls DRT is really a version of FCS. For instance, Chierchia emphasizes syntactic notions, such as c-command, to a greater degree than do Kamp (1981) or Kamp and Reyle (1993). Also, the treatment of quantiWcation as unselective is no longer a feature of DRT, so that arguably some of the problems mentioned below (e.g. the proportion problem) are not problems for DRT. For an introduction to a ‘purer’ version of DRTsee Asher (1993) and Geurts (1995, 1999).

440

V. Anaphoric Pronouns and Dynamic Semantics

also dominates B. So e.g. the A node in the tree (Fig. V.1) c-commands the B node, since both are dominated by the C node. But the B node does not c-command the A node, since the Wrst branching node that dominates B (viz. D) does not dominate A.) Quantifying elements are unselective (i.e. capable of binding multiple variables), and will bind all variables that are free in their restriction clause (as is explained below). A default rule of existential closure applies to all variables that would otherwise remain free. C A

Fig. V.1

D

B

E

To see how this works, take sentence (9) above. It has the form Q [A] [B], where A is the restriction of the quantiWer Q and B is its so-called ‘nuclear scope’.6 Hence, (9) can be parsed as follows: (9a) every [man who owns a donkey] [beats it with a stick] Now we introduce a variable for each indeWnite expression and pronoun. The variables introduced by the pronouns are identiWed with the variables associated with the indeWnites on which they are anaphoric. So we get: (9b)

every [man(x) ^ donkey(y) ^ owns(x, y)][stick (z) ^ beat-with(x, y, z)]

Finally, we allow the initial universal quantiWer to bind all the variables that were introduced in its restriction clause. A default existential quantiWer binds the remaining free variable. Thus we have: (9c)

8x, y [man(x) ^ donkey(y) ^owns(x, y)] 9z [stick (z) ^ beat-with(x, y, z)]

In other words, ‘a donkey’ has inherited universal force from the determiner ‘every man’ and ‘a stick’ has existential force by default, since no other quantiWcational binder was available. In an analogous way, (7) and (11) would be analyzed respectively as: (7a) (11a)

8x [donkey(x) ^ owns(Pedro,x)][beats(Pedro,x)] mostx, y [man(x) ^ donkey(y) ^ owns(x, y)] [beats(x, y)]

Conditionals like (7) are assumed to have an implicit Q-adverb ‘always’ from which the indeWnite inherits its universal force. 6 The term ‘nuclear scope’ was introduced in Heim’s dissertation (1982: 137). Its deWnition is simply ‘the third term of a tripartite structure’. Heim reserved the label ‘scope’ simpliciter for both the second and the third terms of a tripartite structure. This terminology stuck, even when tripartite structures were conceptualized diVerently, as in Diesing (1992) or Chierchia (1992, 1995). For Chierchia’s deWnitions of the notions of the restriction and nuclear scope of a quantifying element in a tripartite structure of the form Q[A][B], see Chierchia (1995: 134–7).

V. Anaphoric Pronouns and Dynamic Semantics

441

One of the main problems facing this account is the so-called proportion problem. Consider a situation in which there are ten men. Nine of these men own one donkey each and they do not beat their donkeys. The remaining man owns twenty donkeys and he beats all his donkeys. In this situation (11) would be false, and yet (11a) would be true, since there are twenty man–donkey pairs where the man owns the donkey and beats it, and only nine man–donkey pairs where the man who owns the donkey does not beat it. A competing account for donkey anaphora was oVered by Evans (1977, 1980) and subsequently elaborated by Heim (1990a). According to this account, indeWnites are to be represented as existential quantiWers. Evans called the pronouns in (7), (9) and (11) ‘E-type pronouns’. He contrasted them with deictic pronouns on the one hand (‘He [pointing to a contextually salient male] is my friend’) and to pronouns that behave like bound variables on the other (‘Every boyi thinks hei is smart’). Evans regarded E-type pronouns as referring expressions that have their reference Wxed by a deWnite description. The reference-Wxing description is one that is recoverable from the antecedent sentence by means of a linguistic rule. Thus Heim (1990a) dubs Evans’s account the ‘linguistic E-type approach’. For example, Evans would analyze (7) and (9) as follows: (7LE ) (9LE )

If there is a donkey that Pedro owns, then Pedro beats it [¼ the unique donkey that Pedro owns]. Every man x is such that if there is a donkey that x owns then there is a stick y such that x beats it [¼ the unique donkey that x owns] with y.

As should be clear, the linguistic E-type analysis is committed to an implication of uniqueness. This is problematic, since (7) and (9) do not seem to have uniqueness implications. For example, (7) could be true even if Pedro owns multiple donkeys, so long as he beats all the donkeys he owns. Neale (1990) suggests that Evans’s account should be modiWed in two ways. First, Neale argues that instead of thinking that anaphoric pronouns in donkey sentences have their references Wxed by means of description, we should think of these pronouns as going proxy for deWnite descriptions. Secondly, the descriptions for which these pronouns go proxy must be thought of as numberless deWnite descriptions. Neale’s motivation for saying that pronouns must be thought of as going proxy for deWnite descriptions is that he thinks that these pronouns enter into scope relations (e.g. with modals). This suggests that they should be given a quantiWcational analysis. But on Evans’s E-type analysis they are analyzed as referring expressions (albeit ones whose references are Wxed by description). The motivation for the second change is to avoid the unwanted uniqueness implications. Neale dubs his account the ‘D-type approach’.7 On his account (7) and (9) would be analyzed as follows: 7 Neale (this volume: s. 8) argues that the problem of interpreting donkey pronouns is a special case of the problem of interpreting incomplete descriptions. This latter problem is extensively discussed in the chapters in Part I of this volume.

442

V. Anaphoric Pronouns and Dynamic Semantics (7D ) (9D )

If there is a donkey that Pedro owns, then Pedro beats the donkey or donkeys that Pedro owns. Every man x is such that if there is a donkey that x owns then there is a stick y such that x beats the donkey or donkeys that x owns with y.

One problem with the D-type analysis is that it cannot account for cases of donkey anaphora that have what Chierchia (1995) calls ‘9-readings’ as opposed to ‘8-readings’. (9) does seem to have a preferred 8-reading according to which every man with a donkey beats all the donkeys he owns. Treating anaphoric pronouns as D-type pronouns would account for this 8-reading, since plural deWnite descriptions carry a maximality implication—‘the donkeys’ means ‘all the donkeys’. But consider the following: (13)

Every man who had a credit card used it to pay for his dinner.

Here there is no suggestion that if a man had multiple credit cards he used them all to pay for his dinner. Rather, (13) seems to mean that every man who had a credit card used at least one of his credit cards to pay for his dinner. In other words, (13) has a preferred 9-reading.8 Heim (1990a) outlines a pragmatic version of an E-type account. According to this analysis, indeWnites are analyzed as existential quantiWers and the pronouns anaphoric on them are treated as functions from individuals to individuals. The particular function that is used to interpret an anaphoric pronoun is one that is made salient in the discourse context. So, for example, (9) is analyzed as: (9PE )

8x [[man(x) ^ 9y [donkey(y) ^ owns(x, y)]] ! 9z [stick(z) ^ beatswith(x, f (x), z)]]

Here the subject NP brings to salience a function f from men into the donkey or donkeys they own (Chierchia 1995: 116). Clearly more would need to be said about this notion of salience, and about how context is able to make a function salient. Moreover, like other accounts in the E-type family, this account faces problems concerning unwanted uniqueness implications. Heim (1990a) discusses this and other problems in detail. 8 One factor that determines the preferred reading of a donkey sentence is our world knowledge. In other words, this is a partly pragmatic aVair. E.g. ‘Every man who had a credit card kept it in his pocket’, unlike the superWcially similar (13), has a preferred 8-reading. Another factor in determining the preferred reading is whether the initial determiner is weak (e.g. some, a few, no) or universal (e.g. every). Geurts (2002) presents some experimental evidence in favor of the generalization that donkey sentences with weak determiners have only 9readings, whereas donkey sentences with universal determiners prefer 8-readings, but may have 9-readings too. Geurts notes the Wrst part of the generalization is not exceptionless. Some donkey sentences with weak determiners have preferred 8-readings. E.g. ‘No one who has an umbrella leaves it at home on a day like this.’ This would not be false if someone left an umbrella at home, provided that person took another of his umbrellas with him. On the other hand, ‘No father who has a teenage son lends him the car on weekends’ has a preferred 9reading. It is falsiWed by a single father–son pair such that the father lends the son his car on the weekends, even if that father has another son to whom he doesn’t lend his car. The latter two examples are discussed by Chierchia (1995).

V. Anaphoric Pronouns and Dynamic Semantics

443

One advantage of the pragmatic E-type approach is that it can deal with examples that present problems for DRT. Consider examples such as the following: (14) (15) (16)

Every chess set comes with a spare pawn. It is taped under the box. It is not true that John doesn’t have a car. It is parked outside. I think John has an apartment in Paris. I hope he hasn’t sold it.

According to DRT the indeWnite in (14) should not be accessible to the subsequent pronoun, because the pronoun is not c-commanded by the quantifying element that ccommands the indeWnite.9 In the case of (15) and (16) the fact that the indeWnite occurs within the scope of a negation or an attitude verb should make it inaccessible to the subsequent pronoun. But the pragmatic E-type approach can account for these cases, on the grounds that a function from individuals to individuals is made salient by the antecedent sentences. In the case of (14), it is a function from chess sets to their spare pawns, in the case of (15), it is a function from people to the cars they own that is applied to John, etc. However, the fact that the pragmatic E-type approach works for cases such as (14)– (16) does not mean that this is the correct account to use for cases of donkey anaphora. It may be that one sort of account is appropriate for donkey anaphora and another for cases of anaphora across inaccessible domains. Chierchia (1995) oVers such a hybrid account. He argues that plain donkey anaphora, unlike anaphora across inaccessible domains, is ‘governed essentially by structural factors, just like c-command anaphora’ (1995: 10).10 Cases of donkey anaphora that require 9-readings, for example, (13) ‘Every man who had a credit card used it to pay for his dinner’, are to be dealt with by his dynamic binding account. On the other hand, he argues that anaphora across inaccessible domains and 8-readings of plain donkey anaphora are best dealt with by a pragmatic E-type account (see again n. 8). Neale (1990) mentions yet another sort of approach to dealing with anaphoric pronouns. We will call this the ‘pragmatic referentialist’ (PR) account. Neale attributes this idea to Kripke (1977) and Lewis (1983b), and argues that it must be distinguished from Evans’s E-type approach and also of course from his own D-type approach. Defenders of the PR account treat indeWnites as quantiWcational expressions, as do 9 There is no implied claim here that a pronoun must be c-commanded by an indeWnite in order for the indeWnite to be accessible to the pronoun. In donkey sentences the pronoun is not c-commanded by the indeWnite, and yet in these sentences the indeWnite is accessible to the pronoun. However, in donkey sentences the pronoun is c-commanded by the quantiWcational element that c-commands the indeWnite. Since this is not the case in (14), the indeWnite here is inaccessible to the following pronoun. See Chierchia (1995: 8, 227). 10 What Chierchia means by c-command anaphora are those cases of pronoun use governed by Chomsky’s Binding Principles. For example, a pronoun c-commanded by an NP cannot co-refer with that NP. Thus in ‘John hurt him’, ‘him’ must refer to some male other than John. If one wanted the pronoun to refer to John, one would have to use the reXexive pronoun ‘himself ’, as in ‘John hurt himself ’. (Zolta´n Szabo´, p.c., notes that this is a little too strong. There are cases of so-called accidental co-reference. ‘Joe hurt him’ is not odd if the speaker is unaware that the hurt person is Joe himself.) If the pronoun occurs outside the governing category of the NP, it may or may not co-refer with the NP. For example, if the pronoun ‘he’ in ‘John gave the book to Bill. He didn’t like it’ is interpreted as a bound pronoun, it can refer either to John or to Bill. It can also be interpreted as an unbound pronoun that refers to a third, contextually salient male individual, diVerent from either John or Bill.

444

V. Anaphoric Pronouns and Dynamic Semantics

Evans and Neale. Like Evans, but unlike Neale, defenders of the PR account treat anaphoric pronouns as referring expressions. Unlike Evans, defenders of the PR account do not think that pronouns have their references Wxed by means of a deWnite description that is recoverable from the antecedent sentence by means of a linguistic rule. Nor do they appeal to the idea of a contextually salient function from individuals to individuals, as do defenders of the pragmatic E-type approach. Rather, the idea is that the use of the indeWnite (sometimes) makes some individual salient, and subsequent pronouns can refer to this contextually salient individual. Stalnaker (1998) defends something like this view, and as we will see below, the chapters by Breheny and Dekker (this volume) build on some of the ideas from Stalnaker (1998). The various approaches for dealing with indeWnites and the pronouns anaphoric on them that we have discussed can be summarized by the tree diagram. Approaches to indefinites/anaphora

Dynamic

DRT (Kamp)

Fig. V.2

FCS, Dynamic Binding (Heim, Chierchia)

Non-dynamic

Pronouns interpreted via descriptions

Linguistic E-type (Evans)

D-type (Neale)

Pronoun interpretation dependent on contextually salient material

PR-type (Stalnaker)

Pragmatic E-type (Heim)

Richard Breheny (this volume) is concerned to oVer an account of inter-sentential anaphora from a non-dynamic semantic perspective that takes seriously the pragmatic nature of anaphor resolution.11 He is not concerned with the whole array of cases that have been discussed in the literature. His focus is speciWcally on cases of discourse anaphora such as: (17)

A man walked in the park. He whistled.

11 As mentioned above, one of the principal issues in the debate about indeWnites and anaphoric pronouns is whether their treatment calls for a modiWcation of standard semantics. Advocates of DRT and FCS and semanticists such as Chierchia (1995) and Groenendijk and Stokhof (1991) have argued that we need a dynamic semantics to account for these phenomena. According to dynamic semantics, the meaning of a sentence should not be thought of (solely) in terms of its truth conditions but rather in terms of its ‘context-change potential’. As Chierchia (1992: 131) puts it: ‘the semantic contribution of a sentence is not just the presenting of a content, but also that of changing the information of the interpreter through that content’. In opposition to these theorists, Cresswell (2002) argues that the behavior of indeWnites and anaphoric pronouns merely reveals the extent to which natural language meanings are context-dependent. Moreover, this context-dependence can be adequately dealt with in a standard (non-dynamic) semantic framework, since it involves no more than the contextual assignment of values to free variables. Geurts (1999) has a slightly diVerent reason for denying that the phenomena in question call for a revision in standard semantics, as mentioned in n. 1 above.

V. Anaphoric Pronouns and Dynamic Semantics

445

Breheny begins his chapter with a helpful summary of various dynamic and nondynamic analyses that have been proposed to deal with discourse anaphora and so called ‘donkey pronouns’, and he lists some of the main criticisms that have been raised against these approaches. Breheny classiWes these rival accounts into dynamic versus static approaches. Within static approaches he considers what he calls cross-sentential binding (Geachean) accounts and E-type approaches.12 Within E-type approaches he considers linguistic and pragmatic variants. He then considers the pragmatic approach oVered by Stalnaker (which we have called the pragmatic referentialist (PR) approach). Finally he oVers his own solution. As will emerge below, Breheny’s account of pronoun interpretation is an interesting blend of the sort of descriptivist approach taken by Neale (1990, this volume) and Stalnaker’s pragmatic account. According to the pragmatic account oVered by Stalnaker (1998), the Wrst sentence in (17) is analyzed as an existential generalization. The pronoun in the second sentence is treated as a referring expression. The use of such a pronoun carries a presupposition that there is an individual uniquely available to be its referent. In many contexts, it will be reasonable to assume that the speaker used the indeWnite with some particular individual in mind. Thus the presupposition associated with the use of the subsequent pronoun can be accommodated by assigning as its referent the individual that the speaker has in mind. If instead we assume that ‘he’ refers to some other male in the context, the connection between the walking man and the whistling man is lost. It is because we assume that the speaker is trying to obey the rules of conversation and is not introducing a second male into the conversation that we accommodate the presupposition of the pronoun by assigning as its referent the individual the speaker has in mind in using the indeWnite.13 Stalnaker is not saying that indeWnites are always used with referential intentions. In cases in which the speaker cannot be construed as having an individual in mind, the only way of accommodating the pronoun’s presupposition is to assume that the individual in question is the unique individual that satisWes the descriptive condition ‘man who walked in the park’, and thus in these cases the pronoun will seem to behave as an E-type or D-type pronoun. 12 What Breheny calls the static binding or Geachean account does not have a place in the tree diagram (Fig. V.2). It is a non-dynamic account. But it treats pronouns as analogous to bound variables, rather than as referring expressions (as linguistic E-type and PR-type approaches do), or as proxies for deWnite descriptions (as the D-type approach does), or as introducing free function variables (as the pragmatic E-type account does). 13 As Dekker (2002a, b) notes, a pragmatic referentialist account is able to deal with examples such as the one from Strawson (1952): ‘Anne: A man just fell over the edge of the cliV. Bob: He didn’t fall. He was pushed.’ If Bob’s use of the pronoun ‘he’ is analyzed as a descriptive pronoun that gets its content from the prior utterance, then Bob’s utterance would turn out to be contradictory, saying in eVect that the man who fell over the cliV did not fall over the cliV. Thus examples like this present a problem for linguistic E-type and D-type approaches, as well as for DRT. Neale (1990) calls this the problem of pronominal contradiction. However, on the pragmatic referentialist account, Bob’s use of ‘he’ is interpreted as a referring expression that refers to the individual that Anne has in mind by her use of the indeWnite ‘a man’, and presumably such an individual can be identiWed independently of the descriptive content ‘man who fell over the cliV’ (e.g. the man could be the salient individual in Anne and Bob’s common perceptual environment).

446

V. Anaphoric Pronouns and Dynamic Semantics

Despite its virtues, Breheny (2002b, this volume) argues that Stalnaker’s account is unsatisfactory, since the pragmatic presupposition Stalnaker claims is associated with the pronoun could in certain circumstances be accommodated without assuming that the pronoun’s referent is the individual the speaker has in mind and yet without construing the speaker as having violated any pragmatic maxims of conversation (such as Grice’s maxim of relation: ‘Be relevant’). In such cases Stalnaker’s account yields the wrong predictions. Another problem facing such a pragmatic referentialist account is to explain how it is that an individual is made salient by the utterance of an indeWnite, and to say more precisely when an individual is or is not accessible to a subsequent pronoun. Breheny’s own proposal builds on Stalnaker’s insights. But it also relies on ideas from Sperber and Wilson’s (1986/1995) relevance theory, from Cooper’s (1996) theory of generalized quantiWers, from Barwise’s (1988) situation theory, and from psychological theories about the focus of attention. From relevance theory Breheny takes the idea that the utterance of a sentence can implicitly convey a second proposition along with the proposition that it explicitly expresses. Thus the utterance of the Wrst sentence in (17) above directly expresses an existential generalization. However, it may also implicitly convey the proposition that there is an individual who is the actual causal source of the speaker’s intention concerning her use of the indeWnite expression ‘a man’ and who is believed by the speaker to satisfy the complex predicate ‘man(x) and walked-inpark(x)’. This implicitly conveyed proposition about the speaker’s referent can contribute to what is in the audience’s focus of attention. Breheny argues that pronouns are governed by a rule according to which the pronoun’s referent/interpretation must be accessible from what is currently the focus of attention for the audience. Breheny is not oVering a referentialist account of anaphoric pronouns. Rather, he treats such pronouns descriptively, and hence they are given a quantiWcational analysis. The principal diVerence between descriptive pronouns and deWnite descriptions is that pronouns have no explicit category restriction, whereas things falling under a deWnite description of the form ‘the F ’ must belong to the category of F s. However, descriptive pronouns can be implicitly restricted. Breheny’s view is that any subset of the set of individuals in the focus of attention is potentially available to restrict the pronoun. Moreover, as already mentioned, pronouns are governed by a rule according to which their interpretation must be accessible from what is currently the focus of attention for the audience. Breheny uses ideas from situation theory and the theory of generalized quantiWers to spell out more precisely the notion of accessibility that is required. The closing sections of his chapter spell out this notion and consider several novel predictions of his account as to what aspects of the prior discourse will be accessible to anaphoric pronouns. Paul Dekker (this volume) is also concerned with the phenomenon of discourse anaphora, such as is illustrated in (17), although Dekker uses diVerent examples. The view Dekker defends is similar to the one defended by Stalnaker (1998). Dekker assumes that indeWnite NPs are to be given an existential analysis. He also assumes that

V. Anaphoric Pronouns and Dynamic Semantics

447

indeWnites can indirectly introduce discourse referents, in virtue of the fact that indeWnites are generally used with referential intentions. The pronouns that are anaphoric on these indeWnites are treated as referring expressions that introduce discourse referents, which can be identiWed with the object(s) the speaker has in mind in using the indeWnite.14 However, Dekker claims that the interpretation of pronouns will depend on pragmatic facts about the discourse, and in particular on facts such as that the pronoun has been used after an indeWnite NP has been used. The use of an indeWnite NP induces a change in the facts that are relevant to the interpretation of subsequent pronouns. Although the pronoun could be interpreted independently of the indeWnite NP (e.g. in the case of (17) this would yield the interpretation that some man walked in the park and some man whistled), it is also possible to interpret the pronoun as referring to the object the speaker had in mind in using the indeWnite NP. Presumably, conversational rules of a Gricean sort (e.g. ‘Be relevant’) will determine when such identiWcation is called for. One of the main points that Stalnaker (1998) makes is that discourse anaphora can be accounted for without resorting to the sort of dynamic semantic approach taken by DRT. Stalnaker argues that we need not complicate our semantics. A static semantics is adequate to the task once we recognize certain pragmatic facts about discourse structure. One structural fact is that in a sequence of utterances, later utterances are interpreted in a context that has already been changed by prior utterances. A prior utterance changes context in at least two ways. First, if the proposition expressed by the utterance is accepted, that proposition will be added to the common ground (viz. the set of assumptions taken for granted by the conversational participants). Second, the fact that the utterance has been made, and facts about the utterance (e.g. that it is in English, that it contains an indeWnite expression used with a certain individual in mind, etc.) can also become a part of the common ground used to interpret later utterances. Dekker agrees with Stalnaker that a non-classical, dynamic conception of meaning is not required to account for discourse anaphora (see nn. 1 and 11). However, he argues that once we focus on the sorts of pragmatic facts about discourse structure just 14 Karttunen (1976) Wrst introduced the notion of a discourse referent. This notion has been used in DRT, where it is claimed that indeWnites and the pronouns anaphoric on them introduce discourse referents. A discourse referent is an element in a discourse representation that may or may not correspond to anything in reality. Bach (this volume) criticizes this notion. He wants to reserve the term ‘referent’ for a real world referent. Bach is also critical of the assumption that indeWnites are generally used with referential intentions, whereas this is an assumption that is central to the position argued for by Dekker (2001, 2002a, b, this volume). This assumption may seem more palatable once one recognizes two things. First, Dekker himself mentions that there are many contexts (e.g. negative and modal contexts, the antecedents of conditionals, the restriction clauses of universal quantiWers, etc.) in which referential intentions are blocked, and he sets out to give a systematic explanation for when and why this should be so. Second, at least as far as Dekker is concerned, the assumption that an indeWnite is used with referential intentions amounts merely to the claim that someone who introduces an indeWnite NP into the discourse must have someone or something in mind. The speaker may not be able to identify or recognize the object in the real world that corresponds to the individual she has in mind (say e.g. if she has learnt of its existence through the testimony of a third party), but she will nevertheless assume that there is such an object to be identiWed.

448

V. Anaphoric Pronouns and Dynamic Semantics

mentioned, we should recognize the need for a dynamic conception of interpretation. He attempts to accommodate this dynamic perspective on interpretation in an extension of Wrst order predicate logic that he calls Predicate Logic with Anaphora (PLA). This is a system of interpretation intended to bring out what the dynamics of interpretation actually consists in. Although PLA relies on a classical (Tarskian) notion of satisfaction, its system of interpretation is dynamic, principally because of its dynamic conception of conjunction. What is dynamic about the PLA notion of conjunction is that it accounts for the intuition that in the conjunction of two assertions, one assertion literally precedes the other, and hence is evaluated in a pragmatic environment diVerent from that in which the second assertion is evaluated. The Wnal sections of Dekker’s chapter address the problem of why, in negative, modal, and certain other environments, the anaphoric potential of indeWnites seems to be blocked, as illustrated in: (18)

Farley doesn’t run a sushi bar. #It is down the street.

Dekker appeals to the topic-focus distinction as well as insights from game theoretical semantics to give an explanation of these blocking phenomena. IndeWnites in the background or non-focused part of an assertion do not introduce possible referents for subsequent pronouns. Such indeWnites are not used with referential intentions, and the speaker is not required to support the use of these indeWnites with evidence for the existence of the things in question. IndeWnites in focus are associated with referential intentions and a speaker is required to have some evidence to support this focused content. Thus if a speaker utters ‘Farley runs a sushi bar’ she is expected to have some evidence for the existence of such a sushi bar, whereas if a speaker utters ‘Farley doesn’t run a sushi bar’ without any special intonation or stress, she will not be required to have any such evidence. In fact she could be rejecting the idea that any such evidence exists. Craige Roberts (this volume) argues that pronouns are deWnite expressions. The proVered (i.e. asserted) content of a pronoun is simply a discourse referent. However, pronouns are associated with presuppositions of ‘weak’ familiarity, salience, and informational uniqueness. Weak familiarity requires merely that the pronoun’s referent be something whose existence is entailed by the common ground (i.e. its referent need not have been explicitly mentioned in the prior discourse). Informational uniqueness requires that the pronoun’s referent be the most salient referent in the context that satisWes the descriptive content of the pronoun (the descriptive content being the person, number, and gender information encoded by the pronoun). Informational uniqueness is a weaker notion than semantic uniqueness. The latter is uniqueness of a Russellian sort, viz. a context-independent entailment that there is one and only one thing satisfying the descriptive content associated with the deWnite expression. There may be uses of pronouns that do not satisfy weak familiarity. But such uses are handled under this theory by assuming that the presupposition will be accommodated (i.e. information will be added to the common ground so that the existence of the referent is now entailed by the common ground). There are restrictions on what can be accom-

V. Anaphoric Pronouns and Dynamic Semantics

449

modated, but generally accommodation will occur if what is to be accommodated is easily retrievable and plausible. The Wrst part of Roberts’s chapter consists in defending these claims, arguing that these presuppositions are indeed the correct ones to associate with pronouns. For example, she argues that pronouns almost never give rise to uniqueness eVects, suggesting that the weaker notion of informational uniqueness is all that is required. Consider: (19)

A wine glass was broken at the party last night. The glass had been a wedding gift. It would be diYcult to replace. Actually, several glasses broke, but only one had any value.

The fact that the Wnal sentence does not seem like a correction or contradiction (but merely an elaboration) shows that semantic uniqueness is too strong for the interpretation of the pronoun ‘it’. In fact, it also shows that not all uses of deWnite descriptions give rise to uniqueness eVects, since the same argument shows that semantic uniqueness is too strong for the interpretation of the deWnite description ‘the glass’. This is a consequence that Roberts embraces. One of Roberts’s central aims is to argue against any sort of ambiguity theory for pronouns. For example, we have already seen that Evans (1977, 1980) argues that some pronouns are deictic expressions, some are analogous to bound variables of quantiWcation, and some are E-type pronouns. Roberts denies that we need to distinguish three sorts of pronouns. All pronouns are deWnites whose proVered contents are discourse referents and whose interpretation is constrained by the presuppositions already mentioned. A full defense of this claim is given in Roberts (forthcoming). In this current chapter she focuses speciWcally on the claims of those who defend the existence of E-type or D-type pronouns, and argues against these views. Roberts concentrates especially on the views of Evans (1980), Heim (1990a), and Neale (1990).15 Having spelt out these views in a very accessible way she goes on to raise a number of problems for these views. Some of these have already been mentioned (such as the proportion problem, the problem of pronominal contradiction, the problem of unwanted uniqueness interpretations, and the problem of accounting for 9-readings of donkey sentences such as ‘Every man who had a credit card used it to pay for his dinner’). But a number of other important and original criticisms are raised. Roberts also discusses an interesting class of examples that tend to have been ignored by those who favor a descriptive analysis of pronouns. These are cases in which anaphoric pronouns must be given a generic or indeWnite interpretation. Consider: (20)

Every farmer that owns donkeys breeds them.

15 Neale (this volume) continues to defend his D-type analysis of anaphoric pronouns. He, like Roberts, oVers a unitary account of pronouns. However, unlike Roberts, who argues that pronouns should be treated as referring expressions, Neale argues that demonstrative, anaphoric, and bound pronouns are all to be given a descriptive and hence a quantiWcational analysis.

450

V. Anaphoric Pronouns and Dynamic Semantics

In (20) the pronoun ‘them’ does not mean ‘the donkeys that the farmer owns’ but rather has the generic interpretation ‘donkeys’. For (20) to be true it need not be the case that the farmers in question breed all the donkeys they own. For example, some of the donkeys they own may be too young or too old to breed. Alice ter Meulen (this volume) contrasts indeWnites that are asserted with those that are inferred, presupposed, or implicit. She notes and characterizes their diVering potential to act as antecedents for pronouns. Consider: (21) (22)

One of the ten marbles is not in the bag. It rolled under the sofa. Nine of the ten marbles are in the bag. # It/The missing marble rolled under the sofa.

The indeWnite in the Wrst sentence of (21) can be the antecedent for the pronoun in the second sentence. However, even though it may be inferred from the Wrst sentence in (22) that one of the marbles is not in the bag, this inferred indeWnite cannot be the antecedent for a subsequent pronoun. On the other hand, this inferred indeWnite can play a role in accommodating the presupposition of weak familiarity attached to the deWnite description ‘the missing marble’. Remember that weak familiarity requires only that the referent of the deWnite expression be something whose existence is entailed by the common ground. It need not have been previously explicitly mentioned. Implicit arguments of predicates behave like inferred indeWnites, in that they cannot be the antecedents of pronouns, although they are available as antecedents for deWnite descriptions, as is shown in (23) below: (23)

The car was washed. #He/the washer/the kids did a good job.

In the Wrst sentence of (23) the agent argument of the verb ‘wash’ is implicit. (Someone or something washed the car.) This implicit argument cannot serve as the antecedent for a subsequent pronoun. However, it can be the antecedent for either a singular or plural deWnite description. (This implicit argument carries no number, gender or animacy information. Hence it imposes no syntactic agreement features on its anaphors.)16 Ter Meulen assumes that these facts about implicit arguments and inferred indeWnites are to be explained within the framework of a dynamic semantics. However, she argues that DRT as formulated by Kamp and Reyle (1993) is not able to account for these facts. According to DRT, explicit indeWnites introduce discourse referents that are available for pronoun resolution and presupposition accommodation. Such dis16 Kent Bach (p.c.) oVers the following as a counter-example to this claim: ‘The car was washed in the service department. They made it sparkle.’ Here it seems that the implicit agent argument (viz. those who work in the service department) can act as an antecedent for the pronoun ‘they’. However, this once again raises the problem of plural anaphora (see n. 4). Bach’s example is not felicitous when a singular pronoun is used: ‘The car was washed in the service department. #He made it sparkle.’ Even if we happen to know that only a single individual works in the service department this is still infelicitous.

V. Anaphoric Pronouns and Dynamic Semantics

451

course referents are said to be declared. Since the discourse referents associated with inferred indeWnites and implicit arguments are not available in this way, they would have to be treated as undeclared. But in DRT undeclared discourse referents are interpreted as being bound by a default existential quantiWer having wide scope over the entire discourse. This would give the wrong results in cases such as: (24)

Every man who had his car washed paid the guy twenty dollars.

Here the implicit agent argument of the verb ‘wash’ plays a role in accommodating the presupposition of weak familiarity for the deWnite description ‘the guy’. But it need not be given a wide scope interpretation. The washer of the car could be diVerent for every man in the relevant domain. A second diYculty that ter Meulen sees is that DRT’s account of inference is inadequate, since it blurs the distinction between inferred and asserted contents. These problems lead ter Meulen to advocate the use of the alternative dynamic framework of Lambda DRT, proposed by Blackburn and Bos (1999) and Kohlhase et al. (2000). This is a compositional DRT that facilitates a richer natural deduction logic. The technical details of this theory are supplied in an appendix, but ter Meulen also works more informally through a simple example to show how it deals with the facts about indeWnites laid out in the Wrst part of her chapter. Ruth Kempson and Wilfried Meyer-Viol (this volume) point to an interesting array of examples that suggest that there are parallels between the way in which anaphoric pronouns depend on their antecedents and the scope behavior of indeWnites. They support this claim with a wealth of examples, including examples from languages other than English, such as Mandarin Chinese and German. One simple example from English that illustrates this parallelism concerns the way pronouns and indeWnites behave in double object constructions, namely constructions consisting of a verb followed by two noun-phrases (NPs). Compare the following: (25) (26) (27) (28)

I showed a student every problem case. I showed every student a problem case. I showed hisi mother every student’si room. I showed every studenti hisi room.

An indeWnite occurring in the Wrst NP of a double object construction cannot be construed as being within the scope of a quantiWer expression occurring in the second NP. So (25) does not have an inverted scope reading according to which ‘every problem case’ has wide scope over ‘a student’. In contrast, (26) has an inverted scope reading in which the indeWnite ‘a problem case’ has wide scope over ‘every student’. There is a parallel constraint on pronoun–antecedent relations when the pronoun occurs in the Wrst NP of a double object construction. Pronominal dependency cannot be established from the second NP. Thus in (27) the pronoun his cannot be bound by ‘every student’. These particular restrictions on pronouns and indeWnites are tied to the double object construction. IndeWnites in other situations are rather free in their scope possi-

452

V. Anaphoric Pronouns and Dynamic Semantics

bilities. For example, (29) below does have an inverted scope reading in which ‘every patient’ has wide scope.17 And reversed dependency relations are possible for pronouns in certain environments, as is shown in (30): (29) (30)

A nurse interviewed every patient. Theiri job prospects preoccupy most Wnal-year studentsi.

We saw in examples (1)–(12) above that indeWnites diVer from other quantiWed expressions in their potential to act as antecedents for pronouns. Kempson and MeyerViol point out that indeWnites also diVer in their scope-taking potential from other quantiWed expressions. For example, only indeWnites freely allow interpretations in which they have a scope wider than the clause in which they are contained. Compare: (31) (32)

Everyone agreed that two examiners marked most particularly good scripts. Everyone agreed that most examiners marked two particularly good scripts.

In the case of (31), ‘most good scripts’ can be construed as taking wide scope over ‘two examiners’, but it cannot take wide scope over everyone. However, two good scripts in (32) can take wide scope both over ‘most examiners’ and over ‘everyone’. Moreover, the wide scope-taking potential of indeWnites is not subject to any of the familiar syntactic restrictions, such as construal across a relative clause boundary. Compare: (33) (34)

Which professor do few students succeed at the problem sets that are assigned by? Few students succeed at the problem sets that are assigned by one professor.

WH-movement is not possible across relative clause boundaries, as is shown in (33). Yet the indeWnite in (34) can be construed as taking wide scope over ‘few students’, yielding the interpretation that there is a professor such that few students succeed at his problem sets.18 17 However, readings that are possible in English are not possible in every language. Kempson and MeyerViol present evidence that such inverted scope readings are not possible in Mandarin Chinese, which allows inverted scope readings only when an indeWnite follows another quantiWer, but not when it precedes another quantiWer. Thus in Chinese, ‘Every man read a book’ has an inverted scope reading, whereas ‘A man read every book’does not. In English of course, both sentences have inverted scope readings. 18 In certain syntactic frameworks, such as Government and Binding (GB), the scope behavior of quantiWers is explained by a process called quantiWer-raising. See May (1985). This is supposed to be a movement phenomenon similar to WH-movement, except that it is covert rather than overt (i.e. it is not reXected in surface structure, but only at the level of logical form (LF), which is the level of syntactic representation that interfaces with the conceptual–intentional system and is subject to semantic interpretation). If quantiWer scope is explained by such covert movement, it ought to be subject to syntactic constraints similar to those facing WHmovement. Thus the fact that indeWnites have scope possibilities that violate these syntactic constraints is a puzzle for theories such as GB. As noted in the text, some have responded to these puzzles by arguing for an ambiguity account of indeWnites.

V. Anaphoric Pronouns and Dynamic Semantics

453

Fodor and Sag (1982) have argued that the exceptional scope behavior of indeWnites is best explained by positing an ambiguity between two sorts of indeWnites, referential and quantiWcational. Referential indeWnites are able to escape the clauses in which they are contained and cross clause boundaries. This would explain the possibility of the wide scope readings in (32) and (34). However, Kempson and Meyer-Viol reject the idea of a referential indeWnite. Instead, they wish to provide a univocal characterization of indeWnites that under-speciWes their various uses, and that also highlights parallels with the way in which pronouns are assigned an interpretation.19 The account they give is framed in terms of a model that they call Dynamic Syntax (DS), which is a ‘formal model of language which purports to be at one and the same time a model of an incremental left–right process of interpretation for natural language strings, yet also a basis for explaining structural properties of language, and hence a grammar formalism’ (Kempson et al. 2001: 253). On the assumption that utterance interpretation is a left to right inferential processing task, Kempson and Meyer-Viol proceed to oVer an account of the choice of the scope of indeWnites that sees it as a pragmatic choice, analogous to the process of anaphoric pronoun resolution. Their chapter in this volume does not contain an account of pronoun resolution. However, an account of the E-type behavior of pronouns from the perspective of DS is given in Kempson et al. (2001: ch. 7). DS is a representational theory. In other words, it does not assign interpretations directly to natural language strings, but posits a representational level of logical form (LF) that is the primary object of semantic interpretation. In DS an LF representation is built up incrementally. This process is represented as one of building a tree, the nodes of which are decorated with logical formulae. These formulae are expressed as terms in a typed lambda calculus. Nodes contain information about how formulae are to be built up and combined with those at other nodes. Nodes also specify a set of requirements on what is needed to complete the LF representation at that node, and may specify restrictions on other nodes. Pre-Wnal trees may include nodes whose position in the tree is left unspeciWed. Pragmatic information is then necessary to arrive at a complete representation of the string being interpreted. In DS, a quantiWed NP is associated with a scope statement. This is a proposition of the form x < y, where x and y are individual variables. It states that the quantiWer that binds x has scope over the quantiWer that binds y. After parsing a string, the scope relations appear as atomic propositions involving the bound variables on the root node of the tree generated by the parse. The scope propositions are used to algorithmically 19 Schwarzschild (2002) also opposes the idea of positing a referential indeWnite in order to account for the exceptional scope-taking behavior of indeWnites. However, Schwarzschild wants to explain away rather than to explain this exceptional behavior. He suggests that indeWnites that appear to have unusual wide-scope readings, or to take scope outside of the clauses in which they are contained, are singleton indeWnites. A singleton indeWnite is an existential whose domain is (either explicitly or implicitly) restricted to a single individual. Schwarzschild’s claim is that ‘apparent unexpected scope-taking by indeWnites is due to their being singleton and hence having their scope neutralized’ (2002: 296). This accounts for the ‘air of referentiality’ that attaches to some uses of indeWnites (2002: 294).

454

V. Anaphoric Pronouns and Dynamic Semantics

update the tree description into a complete and well-formed representation. In the case of indeWnites there is a scope statement of the form Ui < x, where x is the variable projected by the indeWnite, and Ui is a metavariable whose value has to be chosen in context. Interpreting the indeWnite involves substitution, replacing the metavariable with a concrete term. If the value selected for Ui is some variable y bound by a quantiWed NP, then the indeWnite has narrow scope with respect to that NP. If the value selected for Ui denotes the index of evaluation for the sentence as a whole, then the indeWnite has wide scope over any other quantiWed NPs in the sentence. It is this process of choice that makes scope assignment to indeWnites parallel to anaphoric pronoun resolution. In both cases, the input determines an intermediate representation, which must then be completed by appeal to pragmatic information. However, scope assignments must also respect any locality or other constraints. In Chinese, for example, there would be a restriction on the variable x that it must take wider scope than any subsequently introduced variable. There are also restrictions associated with predicate terms. In English there would be a restriction concerning the variable introduced by the Wrst NP in a double object construction. The Wve chapters in this section represent some exciting new directions in the analysis of phenomena involving indeWnites and anaphoric pronouns. Although work in this area was originally inspired by the work of philosophers such as Geach (1962), Lewis (1975), Evans (1977), and Stalnaker (1978), much of the current work in this area is dominated by linguists. While there have always been philosophers whose work has kept abreast of developments in linguistics regarding pronouns and indeWnites (such as Neale 1990, this volume, Larson and Segal 1995; Ludlow and Segal this volume; and Berger 2002), mainstream Anglo-American analytic philosophy of language has at times seemed to be drifting away from concerns with indeWnites, and to be focused mainly on deWnite descriptions. We hope that the work represented in this section will reignite the interest of mainstream philosophers of language in these issues.

13 IndeWnites and Anaphoric Dependence: A Case for Dynamic Semantics or Pragmatics? Richard Breheny

1. Introduction This chapter concerns itself with discourses such as those of the form, ‘An F Gs. It Hs.’—exempliWed in (1). We focus on the anaphoric relation between the pronoun in the second sentence and indeWnite in the Wrst. (1)

A man walked in the park. He whistled.

We ask how this kind of discourse is understood and how this understanding relates to the form of the sentences used. In particular, we want to know what it is about the way in which the Wrst sentence is understood that makes subsequent anaphoric reference possible. In asking such questions we note that, as always in the study of natural language understanding, they are somewhat vague or ambiguous. We could, for instance, be asking about how language users typically view their interlocutors’ actions. Answers to this type of question would be along the lines of, ‘We understand the discourse as . . . ’, where what follows could include some kind of speech-act descriptions making reference to propositional content. Alternatively, we could be asking a more psychological question. A typical answer to this kind of question would be along the lines of, ‘We understand the discourse by. . . ’, where what follows is a description of some kind of process. We can classify approaches to (1) as to whether they focus on the ‘as’ question or the ‘by’ question. Dynamic approaches to meaning adopt the latter stance, playing down the signiWcance of the nature of the act and hence of pragmatics. Non-dynamic approaches have tended to ignore processing issues, sticking with the tools of traditional syntactic and semantic analysis.

Thanks to Anne Bezuidenhout, Robyn Carston, and Deirdre Wilson for feedback on earlier versions of this chapter. Thanks for useful feedback also to audiences at various conferences where elements of this chapter have been presented, including: Emma Borg, Paul Dekker, Hans Kamp, Manfred Kupfer, James Prior, Robert van Rooy, Jason Stanley, and Robert Stalnaker.

456

V. Anaphoric Pronouns and Dynamic Semantics

In the next two sections, I will consider some well-known dynamic and non-dynamic proposals about such anaphoric relations as well as their well-known shortcomings. In subsequent sections, I will argue that a ‘pragmatic’ approach would be far more preferable to the dynamic analysis if the problem of accessibility could be overcome for that approach. I suggest that the shortcomings of the dynamic approach stem from its fundamental assumptions about the explication of semantic facts. I go on to compare two candidate pragmatic approaches, that which employs an E-type analysis of the pronoun in the second sentence and that proposed by Stalnaker which employs the pragmatic strategy of diagonalization. I argue that the account based on diagonalization is in an interesting way too weak and so the only viable alternative is the E-type pragmatic account. Finally, I make some new proposals based on ideas from situation theory and relevance theory which will enable us to overcome the accessibility problem for the pragmatic E-type account and which will enable us to formally deWne a static notion of accessibility for pronouns.

2. Static Approaches: Cross-Sentential Binding and E-Type Analyses Among non-dynamic approaches to (1) we can distinguish between those which assume that the pronoun in the second sentence is bound by the indeWnite in the Wrst and those which do not. Among the latter, E-type accounts assume that the pronoun in the second sentence goes proxy for a deWnite description. E-type approaches in turn can be distinguished according to whether they treat the relation between indeWnite and pronoun as mediated by linguistic rule.

Static Binding In terms of the traditional framework of semantic description, the only analysis according to which anaphoric pronouns such as in (1) are to be treated as variables has thus far been to assume that they are bound cross-sententially. The binding approach, promoted in Geach (1962), is motivated by the generally accepted intuition that (1) is understood according to the gloss, ‘A man who walked in the park whistled’ or according to the translation 9x[man0 (x) ^ walked in the park 0 (x) ^ whistled 0 (x)]. It should be emphasized that it is not thought that there would necessarily be a uniqueness implication in such cases according to which just one man walked in the park. The static binding approach is further motivated by the apparent fact that pronouns which derive their interpretation from previous discourse are judged to be inappropriate where there is no proper antecedent. The infamous marble discourse, due to Barbara Partee, typically illustrates the point. (2a) below is judged infelicitous in spite of the fact that the pronoun in the Wnal sentence is clearly meant to be referring to the

Breheny, IndeWnites and Anaphoric Dependence

457

missing marble. Given that the antecedent sentences in (2a) and (2b) make available the same information, the contrast in acceptability suggests that there is something about the manner in which the information is presented which is the source of the unacceptability. Given the ancillary assumption that deictic pronouns are only properly used in the presence of their referent, there is a straightforward account of this given the binding approach: there is nothing to bind the pronoun in question, so it cannot receive an interpretation. (2)

a.

I had ten marbles but dropped them. I found nine. ??It had rolled under the sofa.1 b. I had ten marbles but dropped them. I found all but one of them. It had rolled under the sofa.

Without going into formal details, it is not diYcult to be convinced, however, that the binding approach suVers from a lack of generality. If the indeWnite is treated as some kind of quantiWcational expression and the pronoun as a variable-like element, then one would assume that if cross-sentential binding can occur in the case of (1), it should occur in the case of other quantiWcational expressions. But this is not so. Consider that (3a) cannot be understood as (3b) as we would expect if cross-sentential binding were a general phenomenon: (3)

a. Every boy left school early. He went to the beach. b. Every boy left school early and went to the beach.

A more telling consideration involves certain quantiWcational antecedents as in (4). Here, the binding account predicts (4a) to be equivalent to (4b) and misses the fact that (4a) entails that just one boy left school early: (4)

a. Exactly one boy left school early. He went to the beach. b. Exactly one boy left school early and went to the beach.

The E-type approach assumes that the anaphoric relation in (1) results from the pronoun being understood as if it were a deWnite description. As Evans (1977) points out, the E-type approach correctly predicts our intuitions about (4) and (3), assuming that binding is only intra-sentential.

Linguistic Versus Pragmatic E-Type Approaches Linguistic E-type approaches assume that the description in question is recovered by linguistic rule. With some proposals, the rule makes reference to the actual linguistic material in the antecedent sentence (see Evans 1977; Heim 1990) with others it makes reference to the semantic interpretation of the antecedent expressions (see Neale 1990; 1 Throughout this chapter, examples which are judged infelicitous or less than acceptable will be indicated as such by the use of question marks. ‘?’ is for less strongly infelicitous and ‘??’ is for more strongly infelicitous.

458

V. Anaphoric Pronouns and Dynamic Semantics

van der Does 1996). The alternative, proposed in Cooper (1979) is to have no rule constraining which description the pronoun goes proxy for, leaving this matter to pragmatics—that is, extra-linguistic principles which govern discourse and facts about the context of utterance. Both approaches face problems.

Problems with Uniqueness, Contradictions, and Accessibility One serious problem for the linguistic E-type approaches has to do with unwanted uniqueness implications. The kind of syntactic rule for recovering the E-type interpretation which Evans proposes predicts that the discourse in (1) entails that just one man walked in the park. Similar problems arise for the semantic rules proposed by Neale (1990). Consider Neale’s rule P5, shown here as (5): (5)

If x is a pronoun that is anaphoric on, but not c-commanded by ‘[Dx: Fx]’ that occurs in an antecedent clause ‘[Dx: Fx](Gx)’, then x is interpreted as the most ‘impoverished’ deWnite description directly recoverable from the antecedent clause that denotes everything that is both F and G. (Neale 1990: 182)

An analysis of (1) based on this rule is given in (6) where (6b) is the logical form of (6a). The analysis incorrectly predicts that (1) implies just one man walked in the park. (6)

a. He whistled. b. [thesing x: man0 (x) ^ walked0 (x)](whistled0 (x) ) c. [thesing x: F (x)](G(x) ) is true iV jF-Gj ¼ 0 & jFj ¼ 1

The diagnosis of this problem is that, in these cases where there is no uniqueness implication, a speaker’s referent of some sort is being introduced in the Wrst part of the discourse and subsequently being made reference to with the second part. One way to resolve this problem might be to suppose that the indeWnite in the Wrst sentence is implicitly contextually restricted in such a way that this speaker’s referent can be picked up in some way by the E-type description in the second. For example, suppose that sgu expresses the property of being the individual which a speaker who makes an utterance u of an indeWnite has in mind. We can think of a speaker’s referent as instantiating this property. What having an individual in mind amounts to in the general case is perhaps slightly problematic, but for the purposes of this chapter we will assume that in factual utterances sgu expresses the property of being the actual causal source of the intention underlying the speaker’s utterance. Note also, we assume that it is in the nature of sgu that it be uniquely instantiated, if at all. The proposal would then be that the Wrst sentence in (1) is understood as in (7): (7)

a. [an x: man0 (x) ^ sgu (x)](walked0 (x) ) b. [an x: A(x)](B(x) ) , jA \ Bj 6¼ 0

Breheny, IndeWnites and Anaphoric Dependence

459

As such, subsequent E-type pronouns could be used to make reference to this speaker’s referent. This would be possible on Neale’s semantic approach to recovering the E-type interpretation, if not on Evans’s syntactic approach. While it would solve the uniqueness problem, this proposal is not viable since it makes the truth of the proposition expressed by the utterance of the Wrst sentence dependent on how things stand with this speaker’s referent regarding being a man and walking. It is well known that this is intuitively the incorrect analysis, for even if the speaker mistakes a woman for a man and thinks of that person as a man walking in the park at the time in question, then the proposition expressed by their utterance of the Wrst sentence would still be true so long as there were men walking in the park at the time. A related problem for the linguistic E-type approaches has to do with contradictions using pronouns. Let us say that the speaker of (1) mistakes walking for roller-blading and that an interlocutor knows this, then she can legitimately say, ‘He was not walking. He was roller-blading.’ If we recover the E-type description from the antecedent linguistic material the retort should appear contradictory, but it is not. Of course, we could suppose that when a speaker utters the Wrst sentence in (1), some assumption about the speaker’s ground is communicated implicitly—as in (8)— and that E-type pronouns can exploit the implicit information. (8)

What is said: [an x: man0 (x)](walked0 (x)) Implicitly communicated: [an x: sgu (x)](Bel(sp, man0 (x) ^ walked0 (x)))

Such a move is not possible for linguistic E-type approaches, but on the pragmatic analysis suggested by Cooper such information can be exploited, resulting in an understanding of the second sentence in (1) as in (9): (9)

9x[sgu (x) ^ whistled0 (x)]

As we will see presently, (9) is in fact a fair representation of our understanding of what the speaker expresses with the second sentence in (1a). It is also clear that the cases where there is contradiction would not be problematic for the pragmatic approach as the description according to which we understand the pronoun would just involve sgu . While this proposal would resolve the uniqueness problem and the contradiction problem could be overcome, the free pragmatic approach would be obliged to give an account for why the appropriate description can be recovered from implicit or pragmatically communicated information in the case of (1) but not in the marble discourse. This problem is quite severe since, if E-type pronouns are just like deWnite descriptions, then one would expect their use to be acceptable in all cases where implicit information has to be exploited. That is, one would expect them to be understood via so-called bridging cross-reference—just as the descriptions are in (10):

460

V. Anaphoric Pronouns and Dynamic Semantics (10)

a.

I had ten marbles but dropped them. I found nine. The missing one had rolled under the sofa. b. Mary checked the picnic supplies and found that the beer was warm.

It is natural to think that there is a pragmatic presupposition concerning salience or accessibility which attaches to pronouns due to the attenuated nominal content and that there is something about the use of the indeWnite that makes this speaker’s referent suitably salient or the relevant information about the speaker’s grounds suitably accessible. However, up until now, almost nothing has been oVered in the way of an adequate deWnition of what is or is not salient or accessible that would relieve the pragmatic account of this problem. In later sections, it will be proposed that a principled distinction can be drawn between the type of implicit information in the indeWnites case and that in the marble example. In the meantime, we will consider how dynamic semantics treats anaphoric relations between indeWnites and pronouns. We will see that, although dynamic semantics treats these relations at the level of semantics and not pragmatics, the treatment requires a whole new way of thinking about meaning.

3. Dynamic Semantic Approaches To account for the fact that we can communicate information with language, the methodology which lies behind static approaches to meaning is to consider utterances as actions, such as ‘saying that’ or ‘asserting that’ and so on. The object of such an act is a proposition and it is traditionally assumed by conservative static theories that, to a good Wrst approximation, this object is determined by what the meaning of the uttered sentence is. We can move away from traditional static methodology by considering that, although folk notions of speech acts are useful for us to talk and think about language use, there is no reason to think that, cognitively speaking, the principles underlying utterance processing necessarily make reference to terms such as these. There is an analogy in this line of thought with Chomsky’s thoughts on natural language syntax. Chomskyan methodology does not presume that the principles of grammar which we more or less unconsciously employ in language use make reference to notions employed in folk discourse about grammar. See Chomsky (2000) for a recent statement along these lines. Following this line, it is fair to argue that there is no reason to think that the tools of traditional semantic description—based on model theoretic semantics for Wrst-order logical languages, say—would necessarily be suitable for description within the context of more cognitively oriented semantic theories. A main claim of dynamic semantics is that semantic facts derive ultimately from the manner in which utterances are pro-

Breheny, IndeWnites and Anaphoric Dependence

461

cessed and so from the principles according to which semantic processing operates. These principles do not make reference to folk pragmatic notions. Taking the canonical use of language to involve information exchange, dynamic semantics further assumes that language processing brings about a change from an initial information state to an output information state. One often hears the dynamic slogan that the meaning of a sentence is its context update potential and that the meaning of a sentential constituent is its contribution to that. This is meant to suggest that the properties of expressions which are ultimately responsible for our semantic data are those which have a bearing on this updating process. In addition to this fundamental shift in the view of meaning, dynamic semantics comes with some technical innovations—mostly concerning the structure of the input and output states. The traditional view would be simply that the states are sets of propositions (or sets of possible worlds) and that updating is a matter adding a proposition (or reducing the set of possible worlds). Dynamic approaches impute more structure than this. Correspondingly, updating involves a more complex manipulation of such states. As a result, the propositional content of an utterance comes out only as a derivative, secondary notion. The original dynamic proposals concerning the structure of input and output information states (Kamp 1981; Heim 1982) involved objects known as discourse referents. Discourse referents are neither linguistic symbols nor referents in the conventional sense. Kamp has pursued the idea that information states are mental representations of a sort and that discourse referents are mental symbols, but there are alternative, nonmentalistic versions (see, for a recent discussion, Groenendijk and Stokhof 1991; Dekker 1993; Zimmermann 1999). An important selling point of dynamic semantics lies in its ability to characterize semantic properties of expressions which do not bear directly on truth-conditional content. The paradigm case for dynamic treatment is the indeWnite–anaphor relationship as found in (1) as well as that found in so-called donkey sentences: (11)

a. If a farmer owns a donkey he beats it. b. Every farmer who owns a donkey beats it.

The traditional treatment of indeWnites is as expressions used to make general statements—analysed as expressions of existential quantiWcation. This treatment is motivated by facts about our intuitions concerning the truth conditions of an utterance of the Wrst sentence in (1) above. More importantly, it is motivated by considerations of other uses of indeWnites which are not ‘speciWc’ in any way. These are exempliWed in (12a)–(12d ), each of which has a ‘nonspeciWc’ reading: (12)

a. John wants to buy a house. b. John didn’t eat a mushroom from this Weld. c. Did you eat a mushroom from this Weld?

462

V. Anaphoric Pronouns and Dynamic Semantics d. I predict a woman will be the next prime minister. e. If a student fails the test, John will lose his job.

The idea is that, assuming a unitary analysis of the linguistic meaning of indeWnites is to be preferred, the existential analysis would account for both the nonspeciWc cases and the speciWc cases, so long as we allow for scope interactions as well as pragmatic supplementation of meaning as discussed above. The dynamic meaning for indeWnites is associated with a rule for introducing discourse referents into the current information state structure. For instance, the Wrst sentence in (1) would augment an input context by introducing a discourse referent as well as conditions on that imposed by the rest of the sentence. In Kamp’s version, the information states are called Discourse Representations Structures (DRSs). We can think of these as pairs of sets of discourse referents and conditions, where conditions can be formed by combining either predicates and the appropriate number of discourse referents or certain logical connectives and DRSs. If we start with a null information state, < , >, then processing (1) would Wrst result in the pair, < {x}, {man(x), walked(x)} >. Pronouns also introduce discourse referents but impose a condition that they be identiWed with an accessible referent. Accessibility is a formal notion designed to ensure the interpretability of the resulting structure. I will return to this shortly. The second sentence in (1) would result in the structure, < {x}, {man(x), walked(x), whistled(x)} >. We can speak of an embedding, f, of such structures in a model just in case f assigns values to the discourse referents such that the conditions come out true. We say that a DRS is true just in case there is such a verifying embedding. In this way, Geachean intuitions about the truth-conditional content of (1) are accounted for. Note also that the facts about the marble discourses in (2) are accounted for along the lines of the binding account as there is no suitable discourse referent to be identiWed with that for the pronoun in (2a) while there is in (2b). The moral here is meant to be that the dynamic account better captures the semantic eVects of the preceding discourse than the traditional approach which records the same propositional content in each case. The dynamic approach proposes a unitary treatment of indeWnites and anaphora across the diVerent cases which we classiWed speciWc and nonspeciWc. Depending on the context in which indeWnites appear, they will result not only in diVerent kinds of interpretation but also, correspondingly, diVerent accessibility relations. We will consider two types of linguistic context brieXy: (sentential) negation and conditional sentences. A sentence of the form [s neg[s A]] augments an input context by introducing a condition, :K —where K is itself a DRS constructed by rule from [s A]. A condition :K is true just in case there is no embedding verifying K . This universal quantiWcation over assignments to discourse referents contained in K not only gives the correct quantiWcational force for examples such as (12b), it also eVectively disbars subsequent identiWcation of discourse referents contained therein with those introduced by subse-

Breheny, IndeWnites and Anaphoric Dependence

463

quent discourse. Thus it is predicted that discourses such as (13) cannot be understood with the indeWnite in the scope of negation and the pronoun anaphoric on that. (13)

John didn’t eat a mushroom from this Weld. It was poisonous.

For technical reasons, in Kamp’s DRT this accessibility condition is codiWed in a rule which eVectively says that a pronoun’s discourse referent can only be identiWed with discourse referents introduced at the same level or above. If we consider the structure for the Wrst sentence in (13) on the syntactically narrow scope construal, < , : < {x}, {mushroom(x), john ate(x)} >>, we see that subsequent discourse containing pronouns would be uninterpretable because there are no discourse referents in the structure at that level. Conditional sentences are similarly embedded by universally quantifying over assignments, but they have more elaborate accessibility structure. It is worth considering them also as they have a bearing on donkey sentences. The example in (11a) would be analysed using a conditional condition, < , { < {x, y}, {farmer(x), donkey( y), owns(x, y)} >!<, {beats(x, y)} > } >. Note that discourse referents introduced by pronouns into the consequent DRS can be identiWed with discourse referents in the antecedent as well as in any superordinate structure. The conditional condition is true in a model iV for every function, f, which is a verifying embedding for the antecedent DRS, there is a function, g, which diVers from f exactly in assigning values to any new discourse referents introduced into the consequent DRS and which veriWes the consequent DRS. Thus it is predicted that (11a) is understood as 8x8y[farmer(x) ^donkey(x) ^ owns(x, y) ! beats(x, y)] which is the agreed understanding. This is a marked improvement on traditional static binding approaches which cannot compositionally derive the unselective binding. It is also an improvement over those E-type approaches which incorrectly imply that quantiWcation is over farmers who own just one donkey. To summarize, dynamic semantics introduces innovations in three ways. First, it is proposed that we should look to properties of expressions which bear on processing for the explanation of semantic data. Secondly, it is proposed that information states have more structure than is traditionally assumed. Thirdly, it is proposed that discourse referents are key components in this structure—implying an analysis of descriptions which cuts across traditional typologies. In evaluating dynamic semantics, we ultimately have to consider it as a program of research, much as we consider generative syntax as such. On one level, we can scrutinize the current best theory to see how it stands up empirically. On another level, we need to Wnd more general considerations for choosing dynamic semantics over alternatives. In the next subsection, I will review a number of empirical considerations which weigh against the treatment of anaphoric relations based on the analysis of indeWnites using discourse referents. I will then consider some general arguments against the dynamic semantic program.

464

V. Anaphoric Pronouns and Dynamic Semantics

4. Problems for Dynamic Semantics It is now widely acknowledged that the original dynamic proposals for dealing with donkey sentences are problematic in many ways. For instance, they fail to predict certain readings of donkey sentences involving adverbial quantiWers such as ‘usually’ (see Heim 1990 for discussion). There is also a problem with variable secondary quantiWcation in donkey sentences (see Chierchia 1992). In addition, the accessibility conditions which come with the dynamic analysis are too strong. We have seen that negation, quantiWcation, etc. should be ‘plugs’—that is, discourse referents introduced in the scope of these should be inaccessible to subsequent pronouns. But there are cases where this does not seem to be so (see Geurts 1998 for discussion): (14)

a.

It is not true that John did not bring an umbrella. It is purple and it stood in the hallway. b. It is ludicrous to pretend that this house doesn’t have a bathroom. You showed it to me, remember?

It is interesting to contrast the predictions of the E-type approach regarding the examples in (14) and (13) with those of dynamic semantics. About (13), the E-type approach simply says that, in this case, there is nothing which could be the referent of the pronoun. In cases where indeWnites fall within the scope of negation but where it can be inferred that a referent for an E-type description recoverable from an antecedent sentence exists, then the E-type account predicts that there should be no problem—as attested in (14a, b). There is also a contrast in predictions with (15) below. On the dynamic account, (15a, b) should be equally bad. On the E-type account, (15a) is bad because universal noun-phrases strongly imply there is more than one individual being quantiWed over while the singular pronoun entails there is only one—thus we have something like a manner maxim violation and a judgment of unacceptability unless there is some kind of rhetorical oVset, as in the case of the joke in (15b). (15)

a. Every boy left school early. He wanted to go swimming. b. Every Mandelson supporter was at the reception; but he was pretty lonely.

Turning to plural pronouns with quantiWcational antecedents, as in (16), Kamp and Reyle (1993) have to resort to a very ad hoc arrangement to achieve the correct anaphoric dependence (see van der Does 1996): (16)

a. Few congressmen admire Clinton and they are junior. b. John bought some donkeys and the vet vaccinated them.

This is a sample of the problems faced by the dynamic approach to anaphoric dependencies. I have elsewhere argued that an appropriately constructed E-type account can

Breheny, IndeWnites and Anaphoric Dependence

465

deal with all kinds of donkey sentences as well as the other data which is problematic for DS. But as we are focusing on the discourse of the type in (1) in this chapter, I leave that issue here.2

5. Dynamic Semantics or Pragmatics? Turning to (1), it was suggested above that, modulo the accessibility problem, the best analysis from a non-dynamic perspective is in terms of a speaker’s referent being introduced pragmatically. As Kamp (1990) notes, such pragmatic considerations do not Wgure in a dynamic account of the semantics of indeWnites and anaphora, irrespective of whether participants think about the speaker’s ground. Also, it is proposed by all dynamic accounts that indeWnites introduce a discourse referent which can be identiWed with subsequent referents, so long as the indeWnite’s referent is accessible for these. The pragmatic E-type account says that where a discourse of the form, ‘An F Gs. It Hs.’ is understood as 9x[Fx ^ Gx ^ Hx] then this is due to the presence of a kind of implicature through which the speaker’s referent is made available for the pronoun to refer to. There are a number of consequences that follow from this. First, we can ask what happens if, for some reason, there is no such implicit assumption? In that case, the pragmatic account predicts that, as happens in (15b) above, the discourse would be understood to imply that just one F G-ed since this interpretive option is still available via what is explicitly expressed via the utterance of the antecedent sentence. The dynamic account predicts no diVerence since these pragmatic considerations do not bear on semantic processing and since indeWnites introduce a discourse referent as a matter of its ‘dynamic meaning’. To test these predictions, we need to consider cases where an indeWnite is used but where we cannot reasonably assume that the speaker is implicitly communicating the relevant assumption about his grounds for using the indeWnite. Two kinds of case come to mind. One is where the speaker just does not have speciWc grounds for an utterance involving an indeWnite. The other is where, although the speaker may have speciWc grounds for what she says, this fact is not relevant or, more generally, not part of what the speaker can reasonably be assumed to be intending to communicate with the use of an indeWnite. The latter kind of case can be illustrated with an adversarial discourse where, it is assumed, the speaker only gives away as much as is necessary (cf. Green 1995). In (17) below, we have a discussion between A and B about speed limits on motorways (i.e. British freeways). We see that B’s use of the indeWnite does not imply he is thinking about any particular accident. 2 See Breheny (1999, 2000) for a unitary treatment of the quantiWcational variability of plural and numberless descriptions in general and E-type pronouns in particular as well as a slightly diVerent E-type analysis of donkey anaphora than that Wrst proposed in Neale (1990).

466

V. Anaphoric Pronouns and Dynamic Semantics (17)

A: B:

If you’d ever witnessed a high-speed motorway accident, you wouldn’t oppose the introduction of a speed limit. I’ve spent half of my working life driving on motorways, so, in fact, I have witnessed a high-speed motorway accident. But I still think that one should be allowed to drive as fast as one wants.

If B were to follow up the general claim in this context with a statement using an anaphoric pronoun, then the pragmatic approach predicts there to be a uniqueness implication (that he had witnessed just one such accident) since there is no implicature introducing a speaker’s referent and so no option for the audience to understand the pronoun as referring to some particular accident the speaker has in mind. Considering (18), we Wnd that this is the understanding we get: (18)

B:

I’ve spent half of my working life driving on motorways, so, in fact, I have witnessed a high-speed motorway accident. It was fatal. But I still think that one should be allowed to drive as fast as one wants.

There is no explanation for this in the dynamic framework which predicts a reading, ‘I witnessed a high speed accident which was fatal’ (where the relative clause is restrictive). It is worth emphasizing that one would suppose that if someone spends a lot of time on motorways, then they would most likely have witnessed a good number of these accidents. So the reading predicted by dynamic semantics would be favored on the grounds of world knowledge were it to be available. This suggests that one is not free to interpret such discourses as if the interpretive options were a matter of some kind of linguistic ambiguity (as van Rooy 2001 seems to suggest). It is the rhetorical properties of the preceding discourse which determine whether a speaker’s referent is available. Further examples of this are already available in the literature. (19) is adapted from Geurts (1997). Here the Wrst part of the discourse contradicts some stance of the audience. What kind of grounds the speaker has for this act is not all that relevant to this purpose and so it is unlikely that there is any implicature involving grounds. So, when a pronoun is used in the next assertion, we once again are bound to use only the general quantiWcational information as a resource for constructing an interpretation for the pronoun—resulting in a surprising uniqueness implication: (19)

It is ludicrous to pretend that there has never been an accident on this motorway. We both witnessed it, remember?

Let us now turn to the second kind of case where the speaker just does not have grounds involving a particular (type of ) individual. One kind of example (used in Stalnaker 1998) involves predictions. When someone makes a prediction involving an indeWnite, we often understand the speaker to have no speciWc grounds. In such cases we predict a uniqueness implication to attend subsequent accessible anaphoric usage. Stalnaker oVers the example in (20a) to illustrate the point, but intuitions may be a

Breheny, IndeWnites and Anaphoric Dependence

467

little sharper with (20b) which suggests that the speaker thinks just one woman will Wnish in the top Wfty:3 (20)

a.

I predict that a woman will be nominated for president in 2004. I also predict she will win. b. I predict a woman will Wnish in the top Wfty in this year’s marathon. But I predict she won’t win it.

Again these diVerences in interpretation are not predicted by dynamic treatments of the indeWnite–anaphoric pronoun relation. The same problems that arise for linguistic E-type approaches also arise for dynamic semantics in the case of contradiction discussed above. Recall that one can coherently contradict the second part of (1) with, ‘He wasn’t walking. He was roller-blading.’ Of the two options open to a dynamic approach, neither deals adequately with our intuitions. Either the resulting state would be contradictory ( < {x}, {man(x), walked(x), whistled(x), : < , {walked(x)} > } > ) or too weak to account for the denial ( < {x}, {man(x), whistled(x), : < , {walked(x)} > } > ). The problem, as has been previously observed (see van der Does 1996; van Rooy 1997), is that the dynamic treatment of anaphora only imputes existential force to pronouns, and fails to capture the fact that they are, in some sense, deWnite. Possibly the most damaging fact for dynamic semantics is that it does not really give an adequate account of the truth conditions of discourses of the form, ‘An F Gs. It Hs’ even when they are understood in the Geachean manner (i.e according to the gloss, 9x[Fx ^ Gx ^ Hx]). Recall that, for dynamic semantics, truth conditions are a derivative notion, determined by the embedding of the information structures in a model— but where examples like (1) are concerned, dynamic semantics predicts the truth conditions are as Geach claimed. By contrast, according to the pragmatic E-type approach, the Wrst sentence uttered expresses the general proposition involving the coinstantiation of F-ness and G-ness. While the claim about the second sentence is that it makes reference to the individual the speaker has in mind and says of it that it Hs. If that is the case, then the prominent, Geachean understanding would have to be in some sense implicit. Can that be correct? To demonstrate that it is, consider the scenario in (21a), adapted from Stalnaker (1998). Suppose also that, in reporting on the events, John utters (21b): (21)

a.

John is politically naive. A practical joking host introduces a tabloid journalist to him as a cabinet minister and a real cabinet minister to

3 Owing to space considerations I will be unable to consider the interesting phenomenon of modal subordination (as in ‘A thief might get in. He would steal the silver.’) of which the examples in (20) are a species. Nor can I consider intentional identity statements (‘Hob thinks a witch killed Bob’s mare. Nob thinks she killed Cob’s cow.’). Both phenomena have posed problems for both E-type and dynamic approaches. SuYce it to say that the most appealing solutions proposed to date (based on Roberts 1990b and Geurts 1995, in the former case, and van Rooy 1999 in the latter) can be made to work for both approaches.

468

V. Anaphoric Pronouns and Dynamic Semantics him as a journalist. In the ensuing (sincere) conversation, the real cabinet minister comes across as pro-Europe while the fake minister comes across as anti-Europe. b. Last night I met a member of the cabinet. He was anti-Europe.

While it would be appropriate for us to respond with (22a) below, we clearly could not respond with (22b) or (22c). So while it is clear that John is unwittingly misleading us into thinking that he met a member of the cabinet who was anti-Europe—nothing he actually says can be denied. (22)

a. He wasn’t a member of the cabinet. b. You didn’t meet a member of the cabinet last night. c. He wasn’t anti-Europe.

It is important to note about (22a) and (22c) that it is the speaker’s intentions in introducing the referent into the discourse and not those of whoever uses the pronoun that determines the referent of subsequent pronouns. This point is illustrated in Stalnaker (1998) with (23). Here we Wnd that A can coherently contradict what B says. If B were able to determine the referent of the pronoun, this would not have been possible: (23)

A: A man jumped oV the cliV. B: He didn’t jump, he was pushed. A: No not that guy, I know he was pushed. I was talking about another guy.

Thus we Wnd that the pragmatic approach correctly predicts that discourses such as in (1) could be true in circumstances which would falsify the Geachean proposition (and the corresponding DRS). To sum up, we have been considering dynamic approaches to the discourses under discussion at three levels. As things stand, it would seem that the original analysis of indeWnites and anaphoric pronouns using discourse referents is empirically problematic in many ways. At another level, if we are to pursue the idea that information states involved in discourse have more structure than is traditionally assumed then we should abandon the discourse referent as the key component. Recent proposals cast in the spirit of dynamic semantics have addressed some of these problems (van Rooy 2001; Dekker 2002a). However, it would be fair to say that the proposed solutions do little more than recast in dynamic terms non-dynamic analyses (in particular, that of Stalnaker 1998, discussed below) involving a separation of semantic and pragmatic facts. By their own admission, these neo-dynamic accounts are more concerned with making claims about the nature of information states and of inference involving these (see Dekker, this volume). The issue surrounding the best general framework for representing the eVect of language use on subsequent discourse is an interesting one, and will be taken up below. Regarding the value of dynamic approaches to inference, I have elsewhere argued that the analysis of conjunction as dynamic (involving the non-

Breheny, IndeWnites and Anaphoric Dependence

469

equivalence of A ^ B and B ^ A) does not fare any better than classical accounts in dealing with the relevant logical properties of discourse (see Breheny 2002a,c). More generally, it would seem that we do not gain a whole lot by considering meaning or information content in these dynamic terms. On the contrary, it seems more plausible to think that conceptualization of language use as action plays a key role in determining semantic facts.4 In the rest of this chapter, I will consider two approaches to discourse which take pragmatics more seriously but which nevertheless incorporate consideration of the process involved in language use.

6. A Non-E-Type, Pragmatic Account Thus far, reference has been made to the pragmatic E-type approach as ‘the pragmatic approach’, but there is available a proposed account of discourses of the form ‘An F Gs. It Hs’ which is pragmatic in the relevant sense but not E-type. On the E-type approach, the understanding of the second sentence in (1) is something like, ‘the individual the speaker has in mind whistled’ (see (9) above). This understanding is assumed to be the proposition which the utterance of the sentence expresses, as a matter of its meaning since the E-type pronoun is analysed as a description. In the case of examples discussed in the last section, such as (18), we similarly have a descriptive understanding, ‘the motorway accident B witnessed was fatal’. In terms of traditional semantic description, we have two analytical options, either these pronouns are implicitly quantiWcational (the option I will adopt below) or they denote functions from possibilities to individuals. Stalnaker’s (1998) analysis of indeWnites and anaphoric pronouns holds out the possibility that we could make do without the E-type analysis if we consider how the discourse would be understood within his speaker presupposition framework, employing sets of possibilities to represent both context and content.5 The suggestion is that in this framework, it would suYce to assume that the pronoun in (1) is understood simply as a variable term of direct reference. Clearly, if the pronoun in (1) were simply a variable, then one does not generally recover what is said (a singular proposition) in these cases. Stalnaker’s proposal is that our understanding of (1) corresponds to a (contextually restricted) diagonal proposition. If this proposal were viable, then we would have a genuine alternative to the E-type analysis, at least if we were prepared to work within the two-dimensional pragmatic framework. However, it turns out that Stalnaker’s account is in an interesting way too weak. The reasons for this have to do more with the pragmatic strategy of diagonalization than the semantic analysis of the anaphoric pronouns. As such, I will scant on a full comparison of the two pragmatic 4 In Breheny (2002c) it is pointed out that this fundamental assumption of dynamics that semantic facts are to be explained at the level of processing leads to more serious fundamental consequences. 5 It is important to note that Stalnaker’s account of (1) being considered in this section faces the same problem with the marble examples as the pragmatic E-type account. Stalnaker (1998) recognizes this problem, noting that its solution has to do with salience but that he can shed no new light on it.

470

V. Anaphoric Pronouns and Dynamic Semantics

approaches, which is taken up in Breheny (2002b), and just give the outline of the argument. In Stalnaker’s two-dimensional pragmatic framework both the context of utterance and the content of an utterance are represented as sets of possibilities. The general assumption is that the context set of possibilities represents what the speaker presupposes (takes to be mutually assumed for the purposes of the discourse). Stalnaker’s account of (1) is just like the pragmatic E-type account in that it is assumed there that the speaker’s referent is introduced via an implicit assumption about the speaker’s grounds. His account of (20) follows from his proposal about the presupposition which attaches to the use of pronouns. It is also assumed that the meaning of indeWnites is just existential. An outline of the account of (1) is as follows. The meaning of the Wrst sentence uttered determines that the speaker has expressed just the existential proposition as part of an assertion. So, in each possibility in the context set, the speaker is asserting this proposition. If accepted, then in each possibility, a man walked in the park. Stalnaker assumes, quite reasonably, that there is a presupposition which attaches to the use of a pronoun which is that there is an individual uniquely available to be the referent of the pronoun. As such, in order to accommodate this presupposition, it needs to be further presupposed between the utterance of the second sentence and its acceptance that, in each possibility in the context, there is an individual available to be the referent of the pronoun. Note that this presupposition is quite weak. As such, simply accommodating it would lead to an understanding of the second sentence along the lines of, ‘Some male whistled’. However, as is reasonable, Stalnaker further assumes that this presupposition gets strengthened by contextual factors to being that the individual the speaker had in mind is, in each possibility in the context, the one available for reference. The reasoning behind this latter assumption, one assumes, is that in possibilities in which the intended referent of the pronoun is someone other than this speaker’s referent, then in those possibilities, the speaker would be coming into conXict with more general Gricean maxims, particularly that enjoining relevance. In the case of examples like (20) where it is not presupposed that the speaker has someone in mind, again, relevance or some other principle would dictate that the individual available in each live possibility would be that which uniquely instantiates being F and G. At least, that is how the account is meant to go. In summary, Stalnaker’s two-dimensional framework for semantic and pragmatic description oVers the analytical possibility that these anaphoric pronouns are just terms of direct reference. The simpliWcation of the semantic analysis comes at the price of more complicated pragmatic reasoning since in these discourses, what is literally said in uttering the second sentence is something which is not recovered as part of the interpretation. What is literally said would be a singular proposition. What we recover is a descriptive proposition involving a function from possibilities in the context to some individual in those possibilities. This is achieved because the framework always leaves open the possibility of diagonalizing.

Breheny, IndeWnites and Anaphoric Dependence

471

The problem for this account is that it relies on the strategy of diagonalization to get the descriptive understanding and this makes the account too weak. If the diagonalization strategy were generally available, it should be the case that we get an existential interpretation of pronouns (‘some male’, etc.) so long as such an understanding is consistent with the presupposition that the speaker is observing the maxims or principles which regulate discourse. To illustrate that this is not what actually happens in discourse, we need to consider cases where there is more than one potential referent but where it would be consistent with pragmatic principles if we understood the speaker to mean ‘one of them’. Examples of this sort can be found in (24): (24)

a.

Mary’s Hollywood dream was slowly turning into a nightmare of drugs and prostitution. She discussed her problems with Father Smith and Father Jones. ??He wisely advised her to go back to her family’s farm in Iowa and that’s what she did. b. Two boys were playing cricket next door and ??he hit a shot which smashed my window.

The problem is that one should be able to get a perfectly acceptable understanding of the pronouns in (24) as ‘one of them’ since, before the Wnal sentence is asserted, in order to satisfy our expectations of relevance as well as the presupposition which attaches to the pronoun, all we need to do is to reduce the context set so that in some possibilities Father Jones is available for reference and in others Father Smith. It may seem possible that the problem can be overcome if the presupposition attaching to pronouns can be strengthened. This possibility is considered and rejected in Breheny (2002b). As such, it would seem that the only way to account for the descriptive interpretation of these anaphoric pronouns is to treat them as descriptions. But this leaves the accessibility problem—to which we now turn.

7. A Situation Theoretic–Relevance Theoretic Alternative In this section, a version of the pragmatic E-type approach will be proposed, drawing on ideas from relevance theoretic pragmatics and situation theory. The aim will be to motivate a suitable notion of non-dynamic accessibility which will enable us to account for the marble discourse and related examples mentioned above.

Communication and Relevance Theory We can begin with a leading idea from Sperber and Wilson’s theory which is that an utterance is an act on the part of one agent directed toward other agents with the aim of drawing their attention toward something. Recognizing an act directed toward oneself as ostensive raises in one’s mind the question, ‘What is my attention being drawn to?’

472

V. Anaphoric Pronouns and Dynamic Semantics

Sperber and Wilson contend that the resolution of this issue is constrained by the search for relevance, which is deWned partly in terms of a kind of cognitive nutrition and partly in terms of the ease with which this is extracted. We could spell out Sperber and Wilson’s leading ideas as follows (see Sperber and Wilson 1986/1995; Sperber 1994): recognizing an act as ostensive not only raises the issue about what is being indicated, but it also raises an expectation that information recovered from what is being indicated will at least provide enough nutritional value to justify the eVort expended in recovering it. There is a further expectation that the agent has acted in a most relevant way modulo her own means and goals. Given this expectation of (optimal) relevance, one is justiWed in resolving the issue which an ostensive act raises by seeking a source of relevant information in the direction indicated. Moreover one can expect that having found a source of information which meets one’s (circumstantially relative) expectations of relevance, one would be justiWed in supposing that that was what the speaker was indicating.

Pragmatics and Situation Theory It is interesting to pursue the idea that the objects of such acts are situations—in the sense of Barwise and Perry’s situation theory.6 When attending to a situation one can obtain information from it in two ways: from the type of situation it is and via constraints which relate situations of that type to situations of other types. For instance, when attending to a situation of a type where John has a cold, one can get the information that John has a cold and, relative to a certain constraint, the information that John cannot go swimming. We will assume that, for each sentence in a language, the semantics will determine, relative to a context, a type of situation. To illustrate, we might suppose that an utterance of the sentence, ‘John has a cold’, determines a type of situation based on the infon < has old , john, l > ^ < now, l > where now is a constant property of spatio-temporal locations Wxed by the context. In the spirit of Sperber and Wilson (if not also Barwise and Perry) we might approximately characterize the role that language plays in communication in the following way:7 For simple assertive utterances of declarative sentences, the normal assumption is that such an utterance forms part of an ostensive act. Then, the type of the 6 Situation theory and situation semantics is something of a moving target. In this chapter, I will be drawing on a middle period version. In particular, I will be drawing on Barwise (1988) and Israel and Perry (1990). I appreciate that many interesting advances have taken place subsequent to this period, but I believe that the basic ideas of situation theory have remained suYciently constant throughout. As I am drawing on basic ideas of the theory, I believe that what I say here would not be incomprehensible to current versions. 7 The idea of ‘strength of a type of situation’ derives ultimately from the idea that states of aVairs (and indirectly, infons) form something like a Boolean algebra. It should also be noted that a more careful statement of the speech convention would make reference to a notion like ‘atomic clauses’ taking into account the fact that sentences have sentential constituents.

Breheny, IndeWnites and Anaphoric Dependence

473

situation to which the ostensive act directly draws our attention is constrained by speech conventions to be at least as strong as the type determined by the sentence uttered. The reason we do not say that the type of situation indicated is fully speciWed by the type determined by the sentence uttered has to do with facts of pragmatic enrichment or ‘unarticulated constituents’ (see, for instance, Carston 1998, Perry 2001). In accounting for the discourse in (1) we come across a kind of case where the type of situation indicated goes beyond the type that would be determined by the meaning of the sentence in the context but where this is not a case of enrichment at the level of the proposition expressed. To account for cases such as (1), we will need to consider a distinction between direct and indirect content.

Indirect Content It is interesting to consider the fact that, when people perform ostensive acts, they often simultaneously display how they relate to what they indicate. To illustrate, suppose a friend is telling you that his pet bird has died. He may betray his feelings of sadness about this fact by the manner in which he behaves. Alternatively, he may draw your attention to certain feelings in a more ostensive way. He can do this either by a kind of mugging of the typical behaviour appropriate to the feeling or explicitly by using a parenthetical—as in, ‘My pet bird, sadly, has died’. It would seem natural to suggest that when people engage in ostensive acts, we conceive of these actions as involving a situation directly indicated and (optionally) situations indirectly indicated. So, what is indirectly indicated in the current example is a situation in which the speaker feels sad about what he is indicating directly. While communicative acts can involve indirectly as well as directly indicated situations, I would suggest that our intuitions about the truth-conditional or propositional content of an utterance derive from intuitions about what is directly indicated only. In the current case, this is borne out by the often made observation that parentheticals do not generally have any eVect on the proposition expressed (see Wilson and Sperber 1993). To illustrate, consider that (25a) is judged to be truth-conditionally equivalent to (25b). That is, although we Wnd out from an utterance of (25a) that the speaker would Wnd it sad if Mary did not get into Cambridge, this information does not aVect how we evaluate the conditional statement. (25)

a.

If, sadly, Mary has not got into Cambridge, she will have to go to Hull. b. If Mary has not got into Cambridge, she will have to go to Hull.

Other cases of indirect content involve speech-act parentheticals such as ‘allegedly’ and parenthetical uses of connectives such as ‘therefore’:

474

V. Anaphoric Pronouns and Dynamic Semantics (26)

a. John is an Englishman. He is, therefore, brave. b. John is seeing another woman, allegedly.

(26a) is discussed by Grice (1975) where he proposes that the sequence involves two basic speech acts (saying that John is English and that he is brave) plus a kind of indirect indication that the grounds for the second act include the Wrst mentioned fact. In (26b), the hearsay parenthetical also indicates something of the grounds for the utterance. In terms of the current proposal, we could say about (26a) that the speaker is indirectly indicating a situation in which he is related to a constraint according to which the Wrst situation indicated carries the information obtainable from the second. A similar kind of story could be told about (26b) involving other speech situations. The proposal about (1) builds on this idea. That is, where a speaker makes an utterance using an indeWnite, she can indirectly indicate that she has speciWc grounds for that use of the indeWnite. What is more, it is not unreasonable to suggest that considerations of relevance motivate the supposition that, where it can be assumed that the speaker has an individual in mind, she is indirectly indicating so. The reason for this has to do not only with the often observed fact that speakers of such utterances normally do have such grounds (rather than general circumstantial grounds), but that to suppose there is a particular individual at issue would make it easier in many ways to conduct the discourse overall (because it is easier to coordinate actions about a particular individual and so on). Given such considerations of eVort, the presumption of optimal relevance implies that one is justiWed in assuming that the speaker intended such indirect information to be introduced, unless there are reasons for thinking otherwise (see Sperber and Wilson 1986/1995 for a discussion of these issues). Having introduced the ideas of directly and indirectly indicated situations, we are in a position to discuss the pragmatic presuppositions which attach to pronouns.

Pronouns and the Focus of Attention If we think about the prototypical deictic use of a pronoun, we think about it being accompanied by some kind of pointing. There is a natural sense in which the pointing precedes the utterance of the pronoun. This can be thought about in terms of the pointing drawing one’s attention to the individual’s presence. Thinking along these lines, we might want to motivate a pragmatic presupposition for pronouns as follows:8 given that pronouns contain no nominal content, it would be unreasonable for a speaker to use a pronoun to make reference to an individual that was not already, in some sense, contained in the situation to which the audience is attending. Otherwise the search space for the referent would be unreasonably large. As an alternative or additional motivation, we could say that it is the normal practice among speakers to 8 I leave out of discussion here pronouns which are dependent on quantiWcational elements (such as so-called bound-variable pronouns and donkey pronouns)—these can be thought of as having more stringent accessibility conditions dictated by linguistic considerations.

Breheny, IndeWnites and Anaphoric Dependence

475

use pronouns only when making reference to things currently in the focus of attention. Given these considerations we can make the following proposal: (27)

The use of a pronoun presupposes that its interpretation is accessible from what is currently the focus of attention for the audience.

I will come shortly to some formalization of the notion of accessibility of an interpretation, but some more general comments need to be made about (27). First, (27) makes mention only of the audience’s attention and not something like joint attention. Clearly the speaker will be attending to what she is drawing other agents’ attention to in virtually all cases and so there will eVectively be joint attention. However, the formulation in (27) is meant to include cases where, through some electronic or other means, there is no speaker or the speaker can have no idea of the situations actually indicated, but we still have intuitions about the felicity or otherwise of pronouns. Secondly, mention is made of the focus of attention. This is a notion about which I can say some general things only at this stage. The notion of focus of attention is something which is closely related to ideas found in the psychological literature and in the computational literature on discourse interpretation (see Garrod and Sanford 1994; Grosz and Sidner 1986, etc.). It seems that, over a stretch of discourse, what is attended to can be added to cumulatively, it can change completely or shift temporarily. There are very many diVerent kinds of moves that can be made in a discourse and these have diVerent eVects on what is attended to. I will mention some fairly basic cases here. These will motivate the subsequent treatment of the marble problem. The marble discourse (in (2) above), like (1), is a case of what we might call continuous planned monologue. In such discourses, the speaker draws the audience’s attention to one situation after another. In very many cases, the situation indicated by a given utterance extends the situation which was already being attended to. That is, the utterance extends the current focus of attention to a situation which includes previously attended to situations. Examples would include many stretches of narrative where a sequence of events involving a single protagonist is described, and they might also include scenario descriptions. There are diVerent cases though where situations are introduced by constituent utterances of planned monologue but where they are put to various supportive purposes such as elaboration. In these cases, the focus of attention may not be augmented but merely temporarily shifted. The examples in (28) illustrate how, in such cases, the subsequent use of pronouns may be inappropriate. (28)

a.

John had a great time last night. He ate a sumptuous meal. The meal came with a bottle of champagne. Then he danced for hours with several beautiful women. ?It was Dom Perignon and had made him feel like Fred Astaire. b. Sears delivered a new siding to my neighbors with the bull mastiV. ?It was the same dog that bit Mary Ben last summer. (Gundel et al. 1993)

476

V. Anaphoric Pronouns and Dynamic Semantics

When it comes to face-to-face dialogue, the focus of attention can be altered by means other than utterances. Consider the following felicitous variation on the marble discourse: (29)

When John came into the room, he found Mary holding a bag of marbles and staring intently at the Xoor. ‘What’s up?’, asked John. ‘I had ten marbles in this bag, but I dropped them’, replied Mary, lifting up the rug. ‘How many have you found?’ ‘Nine.’ ‘Bummer.’ Now both John and Mary began searching the nooks and crannies of the room. After half an hour’s searching, John turned to Mary, ‘Do you think it could have rolled into the next room?’

But note that something has to happen to focus participants’ attention on the presence of an individual to make such anaphoric reference acceptable: (30)

John (manning a cake stall at a church fete and standing over a lone cake, addresses Mary): I baked six cakes and have already sold Wve of them. Mary (holding John’s gaze, not looking down): ??I saw it./It looks delicious.

It seems clear that something like the focus of attention needs to be invoked to account for all the relevant facts to do with acceptable pronominal reference. The marble examples which are generally paraded by proponents of some kind of binding approach to cross-sentential anaphora illegitimately focus on just one subtype of discourse. In the original marble case, we shall say that the missing marble is inaccessible because, although the current focus of attention may carry information about it, one does not get information about the marble from this situation in virtue of the type of situation it is. In the cases such as (1) where an indeWnite is used speciWcally, we are saying that the speaker’s ground for using the indeWnite is indicated indirectly. Although indirectly introduced information does not Wgure in the proposition expressed by utterances containing these uses of indeWnites, the indirectly indicated situation does Wgure in what is indicated overall and so it can augment the focus of attention. So, this way of looking at language use motivates the right kind of distinctions for dealing with the accessibility problem. The accessibility problem arose because we could not distinguish between types of implicit information. In this situation-theoretic pragmatic framework, we have implicit information obtained from situations indirectly indicated and we have the more familiar implicit information obtained in virtue of the information which indicated situations carry in virtue of constraints which relate them to situations of other types. The former information can be exploited in Etype interpretation and the latter cannot. But I now need to say some more about accessibility.

Breheny, IndeWnites and Anaphoric Dependence

477

Accessibility The proposed analysis of the pronoun in (1) is that it is E-type. I will assume without discussion that E-type pronouns, like descriptions, are quantiWcational (although nothing here hinges on this particular choice). Generally speaking, I will assume that unless an expression refers directly to an individual (as perhaps do names and certain indexicals) the types of situation involved in communication will be ‘general’ and will be here analysed quantiWcationally. So, for instance, I shall also be analysing the type of indirectly indicated situations involving speakers’ grounds quantiWcationally. I will explicate what this means shortly. Note that, since the situations indicated with utterances most often involve quantiWcation, what I say about quantiWcation here will bear on both the analysis of E-type pronouns and on what is accessible. I will follow current approaches to the semantics of quantiWcational noun-phrases (labelled DP in syntactic theories) which are based on Westersta˚hl (1985a). EVectively, the Westersta˚hl analysis says that quantiWcational DPs of the form [DP Det[NP [N0 A]]] have a meaning (character) which is a function from contexts to generalized quantiWers restricted by the set containing individuals satisfying the condition which the speaker intended to associate with the use of the noun-phrase. This condition is in turn constrained by the compositional meaning of [N0 A] (via a meaning postulate). Adapting and simplifying9 Cooper’s (1996) treatment of generalized quantiWers in situation theory, I will recast the Westerstahl analysis in situation-theoretic terms along the following lines: (31)

A sentence of the form [S [DP D[NP [N0 A]]][XP B]] determines a type of situation conditioned by the infon < qD , R, S > where qD is the generalized quantiWer relation interpreting D and R is the restrictor constrained by the meaning of the NP constituent and S is the scope determined by the meaning of the XP. b. The restrictor, R, is the set of individuals satisfying the condition which [NP [N0 A]] expresses in the context: R ¼ {x: P(x)} where P is constrained by the meaning postulate, 8u[P(u) ! A(u)] and A is the compositional meaning of [N0 A]. a.

I adopt the standard VP-internal subject hypothesis analysis for VP structure which just states that there is a structural subject position within the VP but that this is occupied by an empty category which is co-indexed with the subject DP phrase. In addition, I adopt the common hypothesis that at the level of syntactic representation that is interpreted semantically, the core VP is evacuated of all overt DPs and it 9 Apart from superWcial diVerences, the main diVerence between the standard situation-theoretic treatment of descriptions and quantiWcation is that we do not exploit resource situations.

478

V. Anaphoric Pronouns and Dynamic Semantics

contains only empty categories for arguments. (See (33) below for an illustration of the semantically relevant syntactic structure (LF) for ‘John ate some sweets’.) As such, I can give the more general rule in (32) where I is the interpretation algorithm taking LFs to infons. Note that the letters in bold in (32) represent parameters and that f is a variable over anchors—partial functions from parameters to appropriate objects. (32)

I ([YP [DP D[NP A]]i [XP ei ]]) ¼ < qD , R, fx: 9s, f s I ([XP . . . ei . . . ])[f [ei =x]]g > I ([VP e1 V e2 . . . ek ]) ¼< rV , e1 , . . . ek , l > (ignoring tense)

To illustrate, consider the sentence in (33a). It will have an LF structure as in (33b). Without going into too many details, the XP will be given an interpretation as in (33c). (33)

a. John ate some sweets. b. [s [DP 9 [some sweets]]2 [XP [DP john]1 [VP e1 ate e2 ]]] c. S ¼ {x: 9s, f s < eat, john, e2 , l > ^ < past, l > [f [e2 =x]]}

I observe that GQ relations which interpret natural language determiners satisfy conservativity, extension, and are quantitive.10 Thus the ‘evaluation’ of an utterance of a quantiWcational sentence need involve only the sets R and R \ S (see van Benthem 1987). I claim that, in the case of utterances containing quantiWed noun-phrases, the situation indicated is the minimal situation which supports < q, R, R \ S > (i.e. required for determining the individuals in R and R \ S).11 When a sentence contains one quantiWed noun-phrase [D[NP]]k in the scope of another [D[NP]]i, the situation j j j indicated will contain component situations required for determining Rk and Rk \ Sk relative to each member, j, of Ri \ Si . Recall that, with the discourse in (1), we assume that the situation indirectly indicated contributes to the focus of attention before the second utterance is made. In that case, the relevant infon for the focus of attention would be something along the lines of (34): (34)

< q9 , Man, W_in_the_P > ^ < q9 , Rsgi , lx.Bel(sp, Man(x) ^ W_in_the_P(x) ) > b. Rsgi denotes the unit set containing the actual causal source of the speaker’s intention concerning her utterance, i, of the indeWnite noun-phrase. c. q9 (A)(B) , jA \ Bj 6¼ 0 a.

10 There is some controversy about this claim due to certain readings of ‘many’ and other cases. I also note, however, that, given an appropriate treatment of the context-dependence of the determiner ‘many’ itself, the assumption of these nice properties of determiners can be maintained (see Westersta˚hl 1985b). 11 The notion of minimal situation required in this discussion can be deWned along the following lines: min(s,s) iV s s & 8s 0 [s 0 s ! s s 0 ]. Also, given that we can have conjunctive states of aVairs (s1 ^ s2 ) and given s s1 ^ s2 iV s s1 and s s2 , we need to think of the minimal situations needed to support infons involving sets as ‘discontinuous’ in an intuitive sense.

Breheny, IndeWnites and Anaphoric Dependence

479

Turning now to E-type pronouns, I will sketch, without argument, an analysis based on Breheny (2000). The analysis diVers from that found in Neale (1990) or van der Does (1996). These latter both treat the deWnite article as a determiner building existentiality and exhaustivity into its meaning. By contrast, I propose that descriptions in general are headed by a phonologically null determiner, Ø9 , which is interpreted by the existential generalized quantiWer relation, q9 . So for ‘the man’ we have [DP Ø9 [NP [art the][N0 man]]]. I assume that the article contributes nothing to the compositional interpretation of the phrase but merely signals deWniteness.12 As with all quantiWed noun-phrases, I assume that the restrictor set for descriptions contains the members of the category the speaker intended to associate with the phrase and this category is constrained by a meaning postulate determined by the compositional interpretation of the NP constituent of the DP. This of course makes both deWnite and indeWnite descriptions merely existential. What distinguishes the two kinds of noun-phrase is that the deWniteness of deWnite descriptions signals to the audience that there is an individual (or collection in the case of plurals) at issue (either because the speaker has an individual or collection in mind or for some other reasons in certain contexts—e.g. modal ones) and that the audience has the means to discern an identifying condition for that individual (or collection). The notion of an identifying condition belongs to a kind of folk semantics. That is, it is presupposed that we keep track of individuals (or collections) by associating identifying conditions with them. When it comes to keeping track of single individuals, it is in the nature of these conditions that necessarily they are uniquely instantiated if at all. It is important to note that in proposing that the deWniteness marker signals that an individual is at issue, we are suggesting that this signaling is part of the act of directly indicating. Thus, for deWnites, the inference would be that the intended category for the restrictor is to be determined by an identifying condition. This means that for singular deWnites, the restrictor set will be a singleton. To illustrate, suppose someone utters, ‘The man whistled.’ Then the semantic conventions associated with the sentence give us two types of information. First, we know that the speaker is indicating a situation the type of which is determined by a quantiWcational infon < q9 , Rman , Swhistled > where the restrictor set is constrained to contain only members who are men. Secondly, we are informed that the situation directly indicated involves an individual that instantiates an identifying condition which we can come up with. The only diVerence between deWnite descriptions and descriptive, E-type pronouns is that in the latter case there is no associated meaning postulate constraining what the intended category is.13 But in addition, pronouns come with the pragmatic condition set out in (27) above. 12 This analysis is motivated in Breheny (2000) for reasons having mostly to do with plurals but it also uniWes the treatment of descriptions in languages which lack markers of deWniteness with those that do. 13 I wish to sidestep, at this stage, the issue of whether any semantic interpretation is given to the gender information contained in English pronouns. I tend to favor the idea that English is much like other languages in

480

V. Anaphoric Pronouns and Dynamic Semantics

We can now spell out in more detail what we mean by an accessible interpretation for a pronoun. Suppose that the current focus of attention is the minimal situation supporting a set of infons, C. C can include quantiWcational infons as well as nonquantiWcational ones (i.e. infons just involving individuals, properties, and relations). For any quantiWcational infon, gi in C , the focus of attention will contain the members of the restrictor set Ri and the intersective set Ri \ Si . For any gk embedded in a gi in C , the focus of attention will also contain the members of the dependent restrictor and j j j intersective sets, Rk and Rk \ Sk for each member, j, of Ri \ Si . And so on for quantiWcational infons embedded in embedded quantiWcational infons. The focus of attention will also contain the individuals in the non-quantiWcational infons. We can then say that any subset of the set of individuals contained in the focus of attention is potentially available to be the restrictor of E-type pronouns. However, as these pronouns are deWnite, they presuppose that, whatever set is intended as the restrictor, this set can be described according to an identifying condition. This means that, in the case of singular pronouns, the individual contained in the unit set in question can be identiWed according to some condition which, it is assumed, the speaker assumes the hearer can discern. So, for an interpretation of a singular E-type pronoun to be accessible, the individual which falls under the description has to already be in the audience’s focus of attention and that individual has to be identiWable according to a condition which, it can be mutually assumed, the hearer can come up with.

Consequences and Comparisons At this stage, it will be useful to consider some examples in order to make some comments and comparisons. These will help to clarify what is going in the proposed framework. First, let us consider the cases where an anaphoric pronoun forced a uniqueness implication (see (18), (19), and (20), as well as (15b) above). I assume that, unless there is reason to think otherwise, in these joined-up monologues it is presupposed that the focus of attention only will be some subpart of what has been indicated with previous utterances. So, in cases where quantiWcational antecedents are used and where it cannot be presupposed that the speaker has an individual in mind, the only potential identifying properties available would be those which describe the restrictor and intersective sets of the previous utterance. By contrast, in the case of (1), what is indirectly indicated about the speaker’s ground augments the focus of attention (as in (34)) and makes available the appropriate utterance-linked identifying condition. that gender features do not get interpreted semantically; but this may be hard to defend as English language users do not seem to have as fully developed a concept of grammatical gender as other language users. If gender does not provide any conceptual constraints on the interpretation of pronouns, then this leaves one free to understand the speaker to be coherently employing the restricting condition lx:x ¼ a (where a is a rigid term) thus allowing such ‘descriptive’ pronouns to be used to express so-called singular propositions. If maleness were a descriptive condition then such an understanding would be incoherent in so far as individuals do not have their gender necessarily.

Breheny, IndeWnites and Anaphoric Dependence

481

It may have been noted that, if the focus of attention is as characterized by (34), then the deWnition of accessibility predicts that, in cases where the speaker has an individual in mind, one can nevertheless use a pronoun to make reference to the members of the restrictor or intersective set of the previous utterance of the sentence containing the indeWnite. This is, I believe, possible. Consider that the pronoun in (35a) can be felicitously used to make reference to the riot policemen in the context, while in (35b) it picks up the intersective set (i.e the riot policemen who cracked protesters’ skulls for no reason): (35)

a.

At the Seattle demonstration, I saw a riot policeman crack a protester’s skull for absolutely no reason. They all seemed to be under orders to club people at will. b. At the Seattle demonstration, I saw a riot policeman crack a protester’s skull for absolutely no reason. They should have been prosecuted for doing that.

In order to accommodate these examples, dynamic approaches would have to apply the (largely ad hoc) treatment of quantiWcational antecedents to anaphoric pronouns as sketched in Kamp and Reyle (1993) (mentioned regarding (16) above). But this treatment presupposes that the antecedents of these plural anaphoric pronouns are quantiWcational. Thus, to deal with (35), the current dynamic treatment would have to suppose that singular indeWnites can be understood quantiWcationally after all. This would have to be some kind of ambiguity account. But there is worse for the dynamic approach since, as the pragmatic approach predicts, it is possible that subsequent to the use of an indeWnite where a speaker’s referent is introduced, one can make reference both to the intersective set and the speaker’s referent. Consider: (36)

At the Seattle demonstration, I saw a riot policeman crack a protestor’s skull. He just did it for no reason! They seemed to be under orders to club people at will./They should have been prosecuted for that.

Within the dynamic framework, it would be incoherent to suggest that the indeWnite here is understood both ways at once, but there does not seem to be any other way to deal with this kind of case. Although examples in (35) and (36) show that all of what is indicated with use of indeWnites in principle can contribute to the focus of attention, it seems that it is in the nature of the focus of attention that, even in joined-up monologue, not all situations previously indicated remain in focus as discourse proceeds. This was suggested by the examples in (28) involving elaborative utterances. Indeed, intuition suggests that, generally, in cases where indeWnites are used to introduce an individual into the discourse—as in (1)—the focus of attention changes implicitly so that the resulting situation attended to is the one in which the speaker’s referent is a man who walks in the park and who whistles. We saw, by considering (21), that though this information

482

V. Anaphoric Pronouns and Dynamic Semantics

is what we normally take away from such a discourse, it is not explicitly communicated by the discourse. Let us now reconsider the claim that, where a quantiWcational infon is involved, the minimal situation in question is one in which we can determine the individuals in the restrictor and intersective sets. This claim is based on the observation of van Benthem that we need only consider these sets to determine a generalized quantiWer’s logical properties. But it is well known that, independently of this consideration, it is virtually impossible to use a plural pronoun to make reference to, say, the individuals in the set R–S. To illustrate, consider that the pronoun in the second sentence in (37) cannot but refer to the Democrats who supported Clinton, even though it is pragmatically more plausible that it should pick up the Democrats who did not support Clinton: (37)

During the Lewinsky aVair, most Democrats in Congress still publicly supported Clinton. Of course they represented more fundamentalist electorates.

In cases where one noun-phrase falls within the scope of another, the role which j j j dependent situations and the dependent sets Rk and Rk \ Sk play as sources of interpretation can be illustrated with (38) where the pronoun in the second sentence is used to make reference to the uninvited friends of the guests who brought uninvited friends. (38)

Many guests brought along some uninvited friends. But they were turned away at the door.

For reasons of space, I have left out of this exposition how quantiWed noun-phrases which fall within the scope of intensional operators are to be treated. An account of such cases could be developed out of suggestions in Barwise (1988) that infons can be used to classify possible situations as well as actual ones. In cases where noun-phrases fall within the scope of modals and other intensional operators we will generate ‘split’ contexts containing sets of possible situations supporting the quantiWcational infons. It is well known from the modal subordination literature that continuations of modal discourse involving anaphora are only felicitous under certain conditions. Contrast (39a) with (39b) below. In the felicitous (39b), the second sentence augments each of the possible situations in the focus of attention so that the students who fail the exam complain to the Dean. Note also that (39c) is somewhat odd, due, I would claim, to the presence of a uniqueness implication attendant on the use of the singular pronoun in the second modal sentence. Again, dynamic treatments of modal subordination (see n. 2) would predict no such uniqueness implication. (39)

a.

It may be that quite a few students failed the exam. ?They didn’t study hard enough. b. It may be that quite a few students fail the exam, and they might complain to the Dean that it was too diYcult. c. A student might fail. He may complain to the Dean.

Breheny, IndeWnites and Anaphoric Dependence

483

8. Conclusion The signiWcantly new idea behind dynamic semantics is that semantic facts derive ultimately from the eVect expressions in a language have on utterance processing. On this view, although folk may talk and think about discourse in terms of the actions of agents, such pragmatic notions do not Wnd their way into this cognitive account of meaning. The anaphoric relation between indeWnites and pronouns has always provided one of the major empirical supports for dynamic semantics. In this chapter, I have tried to give something of the current box score with regards to whether the dynamic approach or a more traditional E-type approach provides better empirical coverage. I have also tried to provide motivation for the more telling point against the dynamic paradigm which is that our conceptualization of utterances as grounded action does have a bearing on semantic facts. On the positive side, I have tried to address the accessibility problem for pragmatic E-type accounts. This was motivated by a situation-theoretic/relevance-theoretic approach to pragmatics. This account adds a processing dimension to the non-dynamic notion of accessibility by having relevance (and hence cognitive processing) considerations determine the focus of attention in cases where utterances provide the context for other utterances.

14 Grounding Dynamic Semantics Paul Dekker

Wenn jemand heute dasselbe sagen will, was er gestern das Wort ‘heute’ gebrauchend ausgedru¨ckt hat, so wird er dieses Wort durch ‘gestern’ ersetzen. (Frege 1918: 64)

1. Introduction In this chapter I present and motivate a formal elaboration of the treatment of anaphoric relationships suggested in (Stalnaker 1998) which is close in spirit and scope to that of Heim (1982) and Groenendijk and Stokhof (1991). I agree with Stalnaker that an adequate account of the relevant facts does not require a dynamic semantic notion of meaning. Yet, I think, the data do provide motivation for such systems of dynamic interpretation. A systematic treatment of anaphoric relationships along the lines suggested by Stalnaker requires us to take into account the referential intentions which can be associated with the use of referring terms, as well as the dynamics of assertion. The dynamics of interpretation then can be seen to reside in a dynamic notion of conjunction, which is a form of intersection, basically, but infected by the generally available pragmatic information that, in most conjunctions actually used, one conjunct literally precedes the other.

2. Assertions in Contexts The treatment of structural semantic relationships in discourse has given rise to several, sometimes deemed major, deviations from classical semantic paradigms, as it has raised systems like discourse representation theory and dynamic semantics, among many others. However, in Stalnaker (1998) it is argued that the phenomena, in particular

Dekker, Grounding Dynamic Semantics

485

inter-sentential anaphoric relationships, are not inconsistent with a classical conception of meaning, if only one pays due attention to the pragmatics of interpretation. In this chapter I elaborate Stalnaker’s view on the issue, and argue that, although indeed a systematic account of the facts along the lines suggested does not presuppose a dynamic conception of meaning, it does motivate a dynamic notion of conjunction. I present a formal semantic interpretation of a language of Wrst-order logic which can be seen to model the compositional interpretation of anaphoric relationships between indeWnite noun-phrases and anaphoric pronouns in natural language. The semantics is spelt out as a classical satisfaction relation, which is extended so as to account for two, arguably systematic, pragmatic principles. First, it pays due attention to the fact that indeWnite noun-phrases are generally used with referential intentions and therefore enable subsequent anaphoric co-reference. Secondly, it takes the fact to heart that, in the conjunction of two assertions, one literally precedes the other. I want to show that by taking these two pragmatic principles seriously, we have a convenient handle on the data, and I thus hope to reconcile the linguistics and philosophy of anaphoric binding. The empirically promising theories of anaphora which have emerged from the discourse representation and dynamic paradigms can be coherently conceived of as naturally emerging from a philosophically motivated, standard, formal notion of meaning. Surprisingly, it appears that it is not so much the dynamics of (existential) quantiWcation and conjunction that stands in need of explanation, but, rather, the blocking eVects which other connectives and quantiWers are generally assumed to have upon this. In the Wnal section of this chapter I therefore oVer a further explanation of these blocking eVects, this, again, in terms of pragmatic principles governing information exchange. I start, however, with an overview of the points made in Stalnaker (1978, 1998). In his pioneering ‘Assertion’ (Stalnaker 1978), Stalnaker points out that the interpretation of indicative sentences may crucially rely upon pragmatic facts related to their use. In order to see what proposition is expressed by means of an utterance of ‘You could be a mega-star’, we have to know the identity of the addressee of the utterance. The content of the utterance, thus, depends on a non-linguistic (non-syntactic and non-semantic) fact, one which I dub a ‘pragmatic fact’ in what follows.1 Pragmatic facts aVect interpretation in more than one way. When someone makes an assertion, and no interfering conditions hamper communication, it is normally assumed that the interlocutors acknowledge that the assertion has been made, and that they accept the assertion if they do not explicitly indicate disagreement with its contents.2 If this is indeed a systematic fact about the use of indicative sentences in our 1 Referring, exclusively, to the established linguistic trichotomy between syntax, semantics, and pragmatics. 2 This is a vast simpliWcation of what ordinarily goes on in actual situations of information exchange, and it also neglects diVerences in the culture of the relevant information society. In some (sub)cultures, what certain persons say is accepted by rule; in other (sub)cultures, this depends on the absence of explicit disagreement of the interlocutors; in yet other (sub)cultures, acceptance requires explicit agreement among all parties.

486

V. Anaphoric Pronouns and Dynamic Semantics

culture, then other utterances may rely on it. For instance, to abuse a worn-out example, we can imagine that someone says: (1)

France has one and only one king.

and, if nobody objects, that she continues with: (2)

And the king of France has dreadlocks.

Linguistic agents may object to this sequence of utterances for various reasons. Semantic reasons include that, for example, France does not have one and only one king, or that, if France has one, the king of France is bald. But it seems that it is inappropriate to admit the utterance of (1), and then to say that the subsequent utterance of (2) is pragmatically deviant, because France does not have a (unique) king. The utterance of (1), and the subsequent absence of protest, thus precludes a rejection of (2) on grounds of presupposition failure. It is a platitude that things change when people do something, but things also may change when people do nothing. If agents carry out an action, and also if they refrain from doing so, the world changes: after the action, the action has been carried out, and the eVects are probably notable; but also if no action has been carried out, then the world follows its scheduled course of events and afterwards it is true that the action has not been carried out. In short, we may observe that, after an assertion has been made, and not before, it is true that that assertion has been made and that after no objection has been made, it is true that the content of the assertion is accepted, for the time being at least. The preceding observations I consciously deem to be platitudes, though platitudes which may have non-trivial implications for the theory of interpretation. For systems of discourse representation theory and dynamics semantics, which arguably focus on the systematic eVects which certain utterances may have on the current state of a dialogue, the implications of this have been taken to be far-reaching. Hans Kamp, for instance, has taken them to support a non-compositional semantics,3 and Groenendijk, Stokhof, and Veltman have taken them to motivate a dynamic semantic notion of meaning.4 In either case the analysis of the dynamics of discourse has been seen to imply a signiWcant departure from the received semantic paradigms, in which meaning is spelt in terms of truth or reference conditions. In what follows I take my cue from Stalnaker (1998) and argue that, eventually, such radical shifts of our concept of meaning are not at all required to account for the data which the mentioned theories set out to give an account of in the Wrst place—anaphoric relations in discourse.5 However, I will only follow Stalnaker in maintaining that the data are consistent with a classical outlook upon meaning, like the one presup3 It must be admitted that these claims are not attested any longer. 4 Although it seems such has never been said so explicitly in print. 5 Thus, this chapter says nothing about other issues people might bring up in favor of a representational or dynamic theory of meaning.

Dekker, Grounding Dynamic Semantics

487

posed in, for example, Stalnaker (1978). Apart from Stalnaker, however, I argue that a fair and systematic account of the data does require some dynamic notion of interpretation and that discourse representation theory and dynamic semantics can be said to oVer just this. Frameworks like these two arguably oVer a systematic elaboration of the analysis of anaphoric relations suggested by Stalnaker himself. I will show this by presenting a toy system of interpretation (predicate logic with anaphora, PLA), which gives a truly dynamic account of the facts about anaphoric relationships on the basis of a static, classical, notion of meaning. I present intuitive, semantic, and pragmatic motivation for all features of this system, including both its dynamic notion of conjunction and its static notion of negation.

3. IndeWnites and Anaphoric Pronouns Let us Wrst inspect some of the relevant data. Suppose someone (Liz) says at some time t: (3)

A magistrate from Gotham city yesterday confessed having blackmailed young women.

Various, more and less trivial, observations can be made about such an assertion. In the Wrst place, Liz has produced certain sounds at t, most probably a string of English words, together making up a sentence. Assuming this to be indeed the case, we can also observe that a substring consisting of the Wrst Wve words constitutes an indeWnite nounphrase. So it also holds true that Liz uttered an indeWnite noun-phrase at t. Other facts about the utterance are of a semantic and pragmatic nature, for instance facts about the truth or truth conditions of the utterance, and facts about its felicity. Our interest here is mainly in semantic/pragmatic facts about the indeWnite nounphrase ‘A magistrate from Gotham city’. It may Wrst be observed that the utterance says nothing about the identity of the magistrate. For the utterance to be true, there must be some magistrate from Gotham city who made such a confession, but as far as we know this can be any magistrate, and it may concern diVerent magistrates in diVerent contexts of evaluation. However, it is often assumed or argued that indeWnites are generally used with what we call referential intentions.6 Thus, where I just now talked about ‘the identity of the magistrate’, I was really supposing that Liz, in the hypothetical situation, would have referred or would have intended to refer to a certain individual. Let us inspect this referential intention in a little more detail. First, it seems obvious to me that Liz’s utterance by itself does not enable anyone to identify the intended referent.7 Secondly, even the speaker (Liz) may fail to know, in a certain sense, to whom 6 Related ideas can be found in the more classical and philosophically oriented literature such as e.g. Chastain (1975); Donnellan (1978); Kripke (1977); Evans (1982). More recent, and, arguably, more linguistically oriented motivation can be found in Kadmon (1987); Kamp (1990); van Rooy (1997); Zimmerman (1999); Dekker (2002a). 7 It has been suggested in the literature on E-type pronouns that the intended referent be the unique individual satisfying the description ‘is a magistrate from Gotham city who confessed, yesterday, having black-

488

V. Anaphoric Pronouns and Dynamic Semantics

she is referring.8 For she just might have heard, herself, what she reported with her utterance of (3), or she might have read it in the tabloids. Nevertheless, thirdly, I think this still does not preclude that Liz used the indeWnite with referential intentions. For her it may very well be the individual, whoever it is, that she was told about when she was told what she now reports with (3), or, alternatively the individual, whoever it is, which she read about in the tabloids.9 Fourthly, and Wnally, I take it that the intended reference is of a pragmatic nature and not a semantic one. Although the intended reference of an indeWnite description is qualiWed as a pragmatic aVair, it may have a semantic role to play. As I said above, a pragmatic fact about an utterance situation may determine the semantic reference of a personal pronoun like ‘you’ or ‘I’, and, likewise, the intended referents of indeWnite noun-phrases may constitute the referents of subsequent anaphoric pronouns. Thus, after uttering (3), Liz might continue with (4): (4)

He had found out they had been bewitching cows before.

In the described situation, the pronoun can be taken to refer to whoever was the intended referent of the preceding indeWnite. That is, the interpretation of a pronoun may draw from a fact about the utterance situation, that it is used in a context where just before an indeWnite noun-phrase has been used, and be interpreted as referring to the intended referent of precisely that noun-phrase.10 So, as Stalnaker himself observes, an account of anaphoric relationships between indeWnite nounphrases and pronouns does not seem to require anything beyond a truth-conditional notion of meaning, if we pay due attention to the pragmatic principles governing the use of indeWnite noun-phrases and pronouns. Meaning and pragmatics together suYce to give a characterization of the content of utterances like that of (3)–(4). However, unlike Stalnaker, it seems, we do not think these observations constitute any evidence against systems of dynamic interpretation, or point at any (conceptual) redundancy of these. For it is one thing to observe that meaning and pragmatics might, in principle, be able to account for a set of data; it is another one to actually provide that account. Consider again the sequence of utterances (3)–(4). For a characterization of the content of the two utterances, one will have to take into account the fact that, after Liz’s utterance of (3), and not before, an indeWnite noun-phrase has been used, and that this explains why the pronoun in (4) can be resolved. (It literally has an antecedent. Notice that this would not be true if example (4) was uttered before (3) and that that mailed young women’, but this is obviously not what we are after here. For Liz may be wrong about the blackmailing, or she may mistakenly think that the individual which she has in mind is a magistrate. In either case the description will not apply to the individual she actually has in mind. 8 That is, if her interlocutor asks which magistrate she is talking about, she may reply ‘I don’t have the faintest idea’, without this disqualifying her as a cooperative speaker. 9 Surely, anyone who would accept what Liz has said with (3) may report the same thing with the intention to refer to the individual which Liz intended to refer to when uttering (3). 10 Thus, the situation is entirely similar to the ones involving deWnite descriptions, as they are discussed in Donnellan (1978).

Dekker, Grounding Dynamic Semantics

489

order of utterances, out of the blue, would sound odd.) It thus appears that the temporal order of utterances is relevant to interpretation, and also that this appears to be quite a systematic fact.11 So, even if we submit that a dynamic notion of meaning need not be called for, still, if we are after a systematic account of the relevant facts of interpretation, we are bound to acknowledge these aspects of interpretation, which may very well be called dynamic. And as a matter of fact, one can hold as well that what systems of dynamic semantics provide is precisely such an account. In order to account for anaphoric relationships in discourse, we somehow have to get a hold on the fact that the use of, for example, indeWnite noun-phrases induces a change in the facts relevant to the interpretation of (subsequent) pronouns. Systems of discourse representation and dynamic semantics can be seen to present two of the many possible means of coming to grips with these pragmatic facts in a systematic fashion. In order to substantiate the last points, I set out, in the next section, to present an, admittedly stylized, analysis of anaphoric relations which I claim to be fully in line with all of Stalnaker’s observations. The proposed analysis is based upon a classical notion of meaning, which is enriched just in order to accommodate the pragmatic observations above. I acknowledge that, as the reader will see, this is somewhat of a formal exercise, but it does help to clarify two points. First, it serves to substantiate Stalnaker’s own point that the relevant data are not at odds with a classical notion of meaning. Secondly, however, it also serves to bring out what the dynamics of interpretation actually consists in. In sections 4 and 5 I will argue in further detail that each of the clauses making up our, arguably dynamic, system of interpretation indeed are given by independently motivated semantic and pragmatic principles. In the remainder of this section I sketch how I plan to carry out a Stalnakerian analysis of simple anaphoric relationships. I substantially simplify matters by only looking at possible anaphoric relationships between indeWnite noun-phrases and anaphoric pronouns, and I furthermore assume that: (i) surface indeWnite noun-phrases are generally used with referential intentions;12 (ii) anaphoric pronouns pick up possibly intended referents of preceding indeWnites;13 (iii) we can neglect information structure, so that the descriptions under which the possible referents of indeWnites are introduced carry the same cognitive or communicative value as those next attributed to them.14 11 Anaphora is a much more widespread phenomenon than the relatively marginal phenomenon of kataphora. 12 Non-surface indeWnite noun-phrases are not generally associated with referential intentions. In various contexts these intentions are blocked, for instance in the context of a negation, of a question, or in the clauses making up the restriction of quantiWers. I will come back to this point. 13 It is only for ease of exposition not to include other possible antecedents. 14 This is not simply a technical simpliWcation, but one which has to be given up in due course. From the formal, systematic, perspective which I adopt in this chapter, however, this is only a technical one.

490

V. Anaphoric Pronouns and Dynamic Semantics

As I said, although indeWnite noun-phrases and pronouns are used with referential intentions, it is normally not clear from their semantics alone which individuals as a matter of fact satisfy these intentions. Their informational contribution is therefore conceived of as functional upon their possible referents. Thus, the referential intentions associated with the use of indeWnite noun-phrases are modeled, below, in terms of their possible referents. And since, in a piece of discourse, the various indeWnites come in a particular order, the whole sum of referential intentions are modeled by means of sequences of possible referents, that is, by means of sequences of individuals that are possible referents of the indeWnites in the discourse. As we will see below it turns out to be convenient to model these sequences in reversed order, so that a sequence d1 , d2 , . . . is said to satisfy a certain discourse, if d1 is a possible referent of the indeWnite used last, d2 of the indeWnite used last but one, etc.15 The above assumptions are made explicit in my statement of the interpretation of a simple language of Wrst-order logic, in which existentially quantiWed formulae model the interpretation of indeWnite noun-phrases and in which an additional category of anaphoric pronouns ( p1 , p2 , . . .) pick up the possible referents of preceding indeWnites. Pronouns are assumed to look back in the discourse for preceding antecedents. More in particular, a pronoun pi is interpreted so that it requires there to be at least i preceding antecedents, and when this requirement is met it is co-referential with the ith indeWnite found when going back in the discourse from the place where the pronoun occurs. Pronouns, thus, are truly indexical.

4. Predicate Logic with Anaphora The system of interpretation which I present in this section is referred to as PLA, (static) predicate logic with anaphora.16 The language of PLA is like that of Wrst-order predicate logic except for the fact that it also contains a category of pronouns P ¼ { p1 , p2 , . . . }. For ease of exposition, I focus on a minimal language, without names and identity, which is built up from variables, pronouns, and n-ary relation expressions, by means of negation :, existential quantiWcation 9x, and conjunction ^. As is usual, I use existentially quantiWed expressions to model indeWnite nounphrases. Conditional sentences can be modeled using implication !, deWned by (f ! c) :(f ^ :c). 15 The relevant order of indeWnites does not always correspond to their linear surface order though. It is determined by syntactic structure, and an indeWnite in the scope of another indeWnite is taken to precede it. In formal terms, 9y comes before 9x in 9x( . . . 9y( . . . ) ), although, of course, it comes after 9x in 9x( . . . ) ^ 9y( . . . ). 16 In an earlier paper (Dekker 1994) a closely related system was presented as an update semantics. Although the system in that paper was called Predicate Logic with Anaphora as well, I prefer to call the system presented here PLA, and refer to the system in Dekker (1994) as DPLA, Dynamic Predicate Logic with Anaphora.

Dekker, Grounding Dynamic Semantics

491

Before I turn to the semantics of PLA, I Wrst have to deWne what I call the ‘length’ of a sentence, the number of indeWnites in it which are or can be used with referential intentions.17 Within the formal system of PLA the length n(f) of a formula f corresponds to the number of existential quantiWers not in the scope of a negation: (5)

n(Rt1 . . . tm ) ¼ 0 n(9xf) ¼ n(f) þ 1 n(:f) ¼ 0 n(f ^ c) ¼ n(f) þ n(c)

Since I only count the contribution made by indeWnite noun-phrases in this chapter (cf. the assumptions at the end of the previous section), only the corresponding existentials are taken to contribute an element. The length of 9xf thus is that of f plus one. Obviously, the number of existentials in a conjunction has to be the sum of the numbers of existentials in the two conjuncts. And, as I said, negations are assumed to block referential intentions, and, hence, anaphoric potential, so that n(:f) ¼ 0.18 Using the notion of n(f) we can already explain how anaphoric pronouns are going to be resolved. As I said, a pronoun pi is indexically analyzed, in that it refers back to the i-th indeWnite noun-phrase which it Wnds if it looks back in the discourse from the location where the pronoun occurs. Thus, if an atomic formula c with a pronoun pi is conjoined with a preceding formula f, and if i n(f), then the pronoun is coreferential with a term in f (the n(f) i-th one), and the pronoun is called ‘resolved’ in the conjunction f ^ c. If, on the other hand, n(f) < i, then the pronoun is not resolved, and it will be taken to refer back to the (i n(f))-th potential antecedent before the conjunction f ^ c. Furthermore, a pronoun cannot be resolved by a quantiWer that has the pronoun in its syntactic scope. So if a pronoun in a formula c presupposes j antecedent existentials before c, then it will also presuppose j antecedents before 9xf (and before :f, for that matter). I now turn to the semantics of PLA itself. The semantics of PLA is speciWed, basically, as a Tarskian satisfaction relation between sequences of individuals and formulae.19 This relation holds of a sequence and a formula if the formula is judged true when its open places (existential and pronominal) are Wlled with the corresponding individuals in the sequence. Notice that such sequences this time do not determine the values of the free variables of a formula, as in, for example, Tarski (1956).20 The PLA-interpretation of terms is deWned relative to a variable assignment and a sequence of individuals. Variables are assigned the value which the current variable assignment assigns to them, and a pronoun pi selects the i-th individual from the 17 The fact that a sentence or discourse has length n can be conceived of as a ‘fact about the discourse’ or ‘discourse information’. Although this type of information is not part of the content of a discourse, it is relevant for the determination of its content, i.e. for its interpretation. (Cf. also Groenendijk et al. 1996; Stalnaker 1998). 18 Notice that this is the default option chosen in systems of discourse representation and dynamic semantics. Cf. s. 6 for more discussion. 19 It is notationally convenient to assume these to be inWnite sequences, although nothing hinges upon this assumption. 20 Variables and variable binding are dealt with by means of variable assignments. The open places which I am concerned with here are those corresponding to indeWnite noun-phrases and anaphoric pronouns.

492

V. Anaphoric Pronouns and Dynamic Semantics

sequence, thus indicating that it is co-referential with the i-th potential antecedent in the preceding discourse:21 (6)

[x]g,~e ¼ g(x)

[ pi ]g,~e ¼ ~ ei

Satisfaction is deWned relative to a Wrst-order model M , a variable assignment g and a sequence of individuals~ e . A model M ¼ hD, Ei consists of a domain of individuals D and an interpretation E for the non-logical constants. If a sequence~ e satisWes a formula relative to a model M and an assignment g I will write ~ e M , g f. Satisfaction is deWned as follows: (7) ~ e M, g Rt1 . . . tm iV h[t1 ]g,~e , . . . , [tm ]g,~e i 2 E(R) ~ e M, g 9xf

iV ~ e 1 M, g[x=~e1 ] f

~ e M, g :f

iV :9~ c 2 Dn(f) : ce c~e M, g f

~ e M, g f ^ c

iV ~ e M, g c and ~ e n(c) M, g f

emþ2 , . . . where ~ e m is the sequence ~ emþ1 , ~ (8)

f is true with respect to M , g and ~ e iV 9~ c 2 Dn(f) : ce c~e M, g f

Given a model-theoretic interpretation of relational constants, atomic formulae are evaluated relative to variable assignments, as in ordinary predicate logic, and relative to sequences of individuals (as in e.g. Tarski 1956). Any pronoun pi in such a formula puts constraints on the i-th individual in such a sequence, and—by our dynamic notion of conjunction—this implies that it puts constraints on the possible referent of the i-last existential. An existentially quantiWed formula 9xf behaves like an ordinary quantiWer as it governs free occurrences of the variable x in its scope. However, it also behaves like a free variable itself. Possible witnesses for the quantiWed statement come from the Wrst individual in the sequence relative to which the quantiWed formula is evaluated. So if ~ e is a sequence satisfying f under an assignment of d to x, then d~ e satisWes 9xf. It is precisely because these sequences thus keep track of the possible witnesses d of x, that they can be (re)addressed by subsequent pronouns.22 The negation of a formula f tells us that f is simply false. It states that there is no way to Wll f’s open places with a sequence ~ c of n(f) individuals. A negation thus closes the ‘existential holes’ of the formula f in its scope, so that, for example, :9xFx, as 21 If~ e is a sequence of individuals d1 , d2 , . . ., then~ ei is di . 22 In the, arguably somewhat scholastic, literature on indeWnites the Lewis/Kamp/Heim approach has been characterized as one oVering an outlook upon indeWnites as free variables, whereas the dynamic approaches of e.g. Barwise and Groenendijk and Stokhof stick to the analysis of indeWnites as (existentially) bound variables. This picture has to be adjusted signiWcantly. On the one hand, the Lewis/Kamp/Heim approach favors an (existential) closure of all indeWnites at some level of analysis. On the other, existentially bound variables in systems of dynamic semantics are still free in some sense. As I have shown in Dekker (1993), such dynamic existentially bound variables can be (re-)bound by other quantiWers using a technique called ‘existential disclosure’. For the interested reader: if x is bound by a dynamic existential quantiWer in c, then it turns out to be universally quantiWed in 8y(c ^ x ¼ y). Thus, if c 9xFx, then 8y(c ^ x ¼ y) is (fully) equivalent to 8yFy.

Dekker, Grounding Dynamic Semantics

493

usual, means that no x is F . As a consequence, existential quantiWers (corresponding to indeWnites) in f cannot serve as antecedents for subsequent pronouns. If we evaluate a conjunction f ^ c relative to a sequence ~ e , we evaluate the Wrst conjunct f relative to ~ e n(c), which is ~ e with the contribution of c stripped oV. Intuitively, this says that f is evaluated before c has contributed its discourse referents. Probably it is easier to read it in a constructive way. If ~ e satisWes f, and ce c~e satisWes c, where ~ c Wts the indeWnites contributed by c, so that the length of ~ c is n(c), then ce c~e satisWes f ^ c as well. The reader may observe that this is a truly dynamic notion of conjunction. The dynamics resides in the fact that the interpretation of the Wrst conjunct f is updated. Sequences satisfying f are updated with the information that n(c) more terms have occurred when we evaluate f’s conjunction with c. Let us brieXy see how PLA handles key-note examples of systems of dynamic interpretation: (9)

A diver found a pearl: [9x(Dx ^ 9y(Py ^ Fxy))]

The length of formula (9) is 2, and straightforward calculations show that: (10)

cd~ e M, g (9) iV c 2 E(D), d 2 E(P) and hc, d i 2 E(F )

that is, if, and only if, c is a diver who found pearl d . Next consider: (11)

She sold it to a tourist: [9z(Tz ^ Sp1 p2 z)]

The length of this formula is 1, but it imposes restrictions, not only on the Wrst element which it mentions, but also on two preceding subjects. The formula is satisWed by a sequence bcd~ e: (12)

bcd~ e M, g (11) iV b 2 E(T ) and hc, d , bi 2 E(S)

that is, if and only if b is a tourist which c sold d to. In the conjunction of (9) and (11), the two pronouns are resolved: (13)

A diver found a pearl: She sold it to a tourist: [9 ^ 11]

The length of the conjunction is 3 and since it is resolved, it only places constraints on these Wrst three elements of a satisfying sequence: (14)

bcd~ e M, g (13) iV bcd~ e M, g (11) and cd~ e M, g (9) iV c, a diver, sold d , a pearl c found, to tourist b

Since a speaker, by uttering (9), is assumed to refer to a diver and a pearl she found, the two subsequent pronouns can be identiWed with the intended referents of the two terms. Thus, in particular, both sentences impose their own constraints on the second and third element of the sequences satisfying the conjunction. Since pronouns may address existentials in precedent conjuncts, and since implication (f ! c) is deWned as :(f ^ :c) in terms of negation and conjunction, pronouns in c may relate to existentials in f. The semantic eVect is the usual one in

494

V. Anaphoric Pronouns and Dynamic Semantics

systems of dynamic semantics. For if we work through the deWnition of ! we Wnd that: (15) ~ e M, g f ! c iV 8~ c 2 Dn(f) : if ce c~e M, g f n(c) then 9~ a 2 D : a~ c e M, g c Thus, a worn-out example like the famous ‘donkey-sentence’: (16)

If a farmer owns a donkey he beats it. [9x(Fx ^ 9y(Dy ^ Oxy)) ! Bp1 p2 ]

turns out to be satisWed in PLA (by any sequence ~ e ) provided that every farmer beats every donkey he owns: (17) ~ e M , g [16a ! 16b] iV e M, g [16a] then cd~ e M, g [16b] iV 8cd 2 D2 : if cd~ 8cd 2 D2 : if c 2 E(F ), d 2 E(D) and hc, d i 2 E(O) then hc, d i 2 E(B) i.e. iV ((E(F ) E(D)) \ E(O)) E(B) Now we have deWned and illustrated the system of PLA,23 we may turn back to the issues discussed in the previous sections. What does this system teach us about the dynamics of interpreting (inter-sentential) anaphoric relationships?

5. Motivation for PLA In the presentation and explanation of the PLA-interpretation procedure I have alluded, every now and then, to observations about the use of indeWnites and pronouns in natural language interpretation. Earlier, in section 3, I claimed that the system of PLA Xeshes out the type of observations or intuitions called upon by Stalnaker. So let us now try and see, in a systematic fashion, how much of the system is actually motivated by such intuitions. This section inspects the intuitions behind all but one of the clauses by means of which the semantics of PLA is spelt out, that is, the deWnition of the length of a formula and three of the four clauses deWning the PLA satisfaction relation. Motivation for the remaining clause, the one dealing with negation, is the subject of section 6. First consider the length n(f) of f, the number of open existentials in a formula. As it is stated it is an entirely syntactic notion, which cannot (always) be derived from the set of sequences satisfying a formula. However, as I indicated, this notion should not be understood to characterize a syntactic property of formulae, but a pragmatic property of the use of these formulae. Properly conceived, n(f) tells us that if someone has 23 The interested reader is referred to Dekker (1994; 2002a: s. 4) for a further exposition of this system, a normalization procedure, and a systematic comparison with Wrst-order predicate logic.

Dekker, Grounding Dynamic Semantics

495

uttered a (sequence of ) sentences corresponding to f, then she has uttered n(f) indeWnites with referential intentions (cf. the assumptions made at the end of section 3). Thus I claim n(f) to characterize a pragmatic fact about the utterance of sentences corresponding to f, which I, like Stalnaker, take to be relevant for the interpretation of subsequent utterances. For each of these indeWnites might have been used with referential intentions which steer the interpretation of subsequent pronouns. Let us now take another look at the interpretation of terms, which are dealt with in the clauses deWning the satisfaction of atomic and existentially quantiWed formulae. Satisfaction of an atomic formula At is deWned in a standard way, except for the fact that the formula may house anaphoric pronouns, the interpretation of which is determined by satisfying sequences of individuals. As the system is set up, a pronoun pi in At is coreferential with the i-th existential or indeWnite used before At. I submit that this rigid use of indices is somewhat artiWcial, but actually it is nothing more than a technical simpliWcation.24 The rationale behind this interpretation of pronouns is like that indicated by Stalnaker. The pronoun is intended to refer to what can be called an ‘intentionally present’ individual, viz. one that the target indeWnite was intended to refer to before. Existentials are treated in the traditional fashion, except for the fact that a sequence that satisWes an existentially quantiWed formula is headed by a witness for the variable quantiWed over. As I have already indicated, this witness counts as a possibly intended referent, given that, as I assumed at the end of section 3, we do not distinguish between the descriptive material which a speaker uses to (indeWnitely) refer to an individual on a certain occasion, and what she asserts of that individual on that occasion.25 The analysis of existentials and pronouns thus is motivated by the idea that both types of terms are used with referential intentions, even though the actual referents are not deWnitely determined. A sequence of individuals satisWes a formula if a corresponding sentence can be asserted with the successive individuals as the possible referents of the successive terms in that sentence. Before we take a further look at the intuition behind the PLA notion of conjunction, it may be emphasized that the semantics of PLA is stated in a static fashion. Sequences of individuals are employed as they are in Tarski’s deWnition of the semantics of Wrstorder logic (Tarski 1956), the mere diVerence residing in the sort of terms which these sequences supply values for. In a Tarski-style semantics, the sequences supply possible values of (sequences of ) open positions which are eventually quantiWed; in my system they relate to the (sequences of ) open positions induced by pronouns and indeWnites. Technically, the diVerence is immaterial, while, pragmatically, it is given by the kinds of intuitions put forward by Stalnaker. 24 Notice that such a disambiguating simpliWcation is adopted by virtually all formal systems of interpretation. 25 I thus neglect, deliberately, the interesting issues addressed in Donnellan (1978) and Kripke (1977), among others. Dropping the mentioned assumption is of course not impossible, but it would complicate matters more than needed for the purposes of this chapter. Cf. e.g. Dekker (2002a) for some further discussion.

496

V. Anaphoric Pronouns and Dynamic Semantics

Although the PLA satisfaction relation can be conceived to be classical in nature, its system of interpretation nevertheless is dynamic, and this is due to its arguably dynamic notion of conjunction. The satisfaction of a conjunction f ^ c by a sequence ~ e is stated in terms of the satisfaction of c by~ e and that of f by~ e n(c). This deWnition should be understood in the following manner. If a sequence a~ c e supplies the possibly intended referents ~ a of c plus those ce c~e that might have been introduced before c, and if ce c~e supplies the possibly intended referents ~ c of f plus those ~ e that might have been ~ of f ^ c plus introduced before f, then a~ c e supplies the possibly intended referents ac those~ e that might have been introduced before f ^ c. The PLA notion of conjunction is meant to account for the fact that when we extend our evaluation of an utterance of f with that of a subsequent utterance of c, we take into account that n(c) more terms with referential intentions have been used after the conjunction of f with c. The PLA satisfaction of a conjunction f ^ c thus is stated in terms of the satisfaction of both f and c, while it bears witness to the fact that satisfaction of f is evaluated n(f) terms before that of c is assessed. This notion of conjunction clearly shows the indexical nature of PLA’s system of interpretation. The satisfaction of f ^ c is stated in terms of the satisfaction of c, and the satisfaction of f before all of the indeWnites of c have been used, and this is why pronouns in c may pick up intended referents mentioned before. Thus conceived, all that is dynamic about the PLA notion of conjunction—and, hence, about the PLA notion of interpretation—is that it accounts for the intuition that, in a succession of two assertions, one assertion literally precedes the other.26 So far, satisfaction of a PLA formula f can be seen to be deWned as a classical satisfaction relation which is extended so as to accommodate certain pragmatic facts about the use of sentences corresponding to f. In order to deal with the interpretation of anaphoric pronouns which can possibly be used later, the system of interpretation systematically takes into account what are possibly intended referents of the indeWnites used, neatly (linearly) ordered following the order of occurrence of the indeWnites at issue. This leaves me to Wnd motivation for the last clause which I have not dealt with so far, the one concerned with negation. As I think this clause deserves special attention, in particular attention from a more general perspective, it is dealt with in a separate section.

6. On Blocked Referential Intentions Under the Stalnakerian perspective which I have sought to elaborate in this chapter, it does not seem to be a mystery any longer why indeWnites and pronouns interact in the 26 It is this intuition which we also read in Frege’s ‘donkey-imperative’, which Wgures as a motto above this chapter. If someone asserted ‘Es regnet heute’ yesterday, and asserts ‘Es regnet heute nicht’ today, then the conjunction of the two assertions today is given by the assertion that it rained yesterday, and that it isn’t raining

Dekker, Grounding Dynamic Semantics

497

dynamic way they do. Like deWnite noun-phrases, indeWnite noun-phrases are used with referential intentions, and anaphoric pronouns simply pick up the referents intentionally related to their antecedents. However, another mystery now rears its head: why do indeWnites seem to stop behaving dynamically when they appear in the scope of a negation? As I said, in PLA, as in most other systems of dynamic interpretation, the negation :f of a formula f disables pronominal co-reference of subsequent pronouns with existentials in f. Derived notions like that of implication, universal quantiWcation, etc., do likewise, as do questions, suppositions, and imperatives. From the pragmatic perspective advocated in this chapter, this raises an intriguing question. If, as I think, anaphoric potential derives from referential intentions, then why do these referential intentions vanish in these negative (and other) contexts?27 I think the blocking eVects of negations and other constructions can be understood well when we take into account their typical role in actual dialogues and the thematic structure of such dialogues. A negation Not S may serve to answer the issue—raised explicitly or implicitly—whether S is true. For instance, consider an utterance of (18): (18) Farly doesn’t run a sushi bar. Typically, such an utterance does not serve to state of Farly and of some sushi bar that the Wrst does not run the last. It is much more likely that such an utterance serves to state—possibly in answer to the question whether Farly runs a sushi bar—that he does not, that is, that there is no such bar which Farly runs.28 Generally, then, a speaker need not have a particular sushi bar in mind when uttering (18), and the reason may be that, intuitively, the existence of such a sushi bar is not part of what the speaker claims to have evidence for. Rather, the existence of such a bar appears to be part of the issue which the speaker addresses—negatively in example (18)—or even part of what the hearer might have claimed just before. So actually, when somebody utters (18), she is normally not coming up with a sushi bar herself, but she is claiming to have evidence against the existence of such a bar, were anybody else thinking of the possibility of there being one, or even of thinking of claiming there actually to be one.29 today. The indexical proposition asserted yesterday has to be updated, today, with the pragmatic fact that it is one day later now. 27 Surely this ‘generalization’ requires some qualiWcation. IndeWnite noun-phrases can be used ‘speciWcally’ in all of the mentioned contexts—under a negation, in the restriction or scope of adnominal and adverbial quantiWers, etc. However, I think that the usual interpretation of indeWnites in these contexts is not ‘speciWc’ in this sense. Since my aim is to understand the more systematic patterns of interpretation in the Wrst place, I would like to understand why indeWnites in these contexts generally fail the anaphoric potential which they normally have outside of them. 28 One has to be careful with this kind of qualiWcation, because alternative interpretations are easily made available by emphasizing, e.g. Farly or run. I here assume the utterance to carry what may be called a ‘neutral’ intonation. 29 In PLA, as in most other systems of dynamic interpretation, negation really acts as a denial of truth, for :f is satisWed by ~ e if and only if f is not true relative to ~ e . The reason is that our truth-deWnition also invokes an existential closure of the open places related to indeWnites, and this type of closure we claim to be intuitively motivated. For, as I said, the referential intentions associated with indeWnites are of a pragmatic nature, and these ought not to aVect basic semantic notions like that of truth (and falsity, for that matter). So, although I hold that

498

V. Anaphoric Pronouns and Dynamic Semantics

Our sketch of the pragmatics of negation is reminiscent of the one adopted in systems of game-theoretical semantics (GTS, cf. e.g. Hintikka and Sandu 1997, for a recent overview). Borrowing GTS terminology, negation can be considered a ‘role switching’ device. In GTS, the truth of f is deWned in terms of the existence of a winning strategy for a ‘veriWer’. A veriWer is supposed to come up with evidence for f and be able to supply witnesses for (indeWnite) terms in f. However, when it comes to a negation :f, the veriWer gets the role of the ‘falsiWer’ of f, who has to refute any attempt to verify f by somebody else. Under this GTS outlook upon negation it may be clear why indeWnites under a negation are not any longer associated with referential intentions. When a speaker asserts :f, the indeWnites in f do not belong to her (pragmatic) jurisdiction so to speak, but to that of any potential antagonist who could come by so as to assert that f. By stating :f the speaker thus rejects any use of f with whatever referential intentions.30 The conception of negation as a role switcher has a further pay-oV when we consider the assertion of conditional sentences f ! c, which are deWned in terms of negation (and conjunction). For instance, the evidence which a speaker may bring to bear upon her assertion of a conditional sentence can be seen to consist in her evidence for the consequent of the conditional, were she to accept anybody else’s evidence for the antecedent.31 Not only does this provide motivation for the often attested existential closure over the indeWnites in the antecedent of a conditional, but it also suggests the speaker’s evidence for the consequent clause to be functionally dependent on the possible witnesses for these indeWnites. As we will see below, the functional type of support which the speaker can be required to have for indeWnites in the consequent clause can be cashed out by subsequent pronouns, if these are also read functionally. As we saw, the typical impact of negations (and implications) can be spelt out, in an intuitively appealing way, in terms of a GTS game between a speaker (or veriWer) and a (virtual, or abstract) respondent (or falsiWer). However, I think, the pragmatic division of labour at issue is more general than the one accounted for in the framework of gametheoretical semantics. Coherent discourse and dialogues generally consist of assertions which have an (explicit or implicit) ‘background’ or ‘topic’ part, and an (explicit) ‘focus’.32 Typicthe referential intentions associated with indeWnites aVect the interpretation of subsequent pronouns, I do not want to bring them to bear upon the evaluation of the assertions in which they are used, whether as true or false. 30 Interestingly, the idea of negation as a role-switcher is also fruitful when we are to specify both what a speaker’s support for an assertion may consist in and what may be the hearer’s update of information when he accepts the speaker’s utterance. It then turns out that speaker’s support for :f is most adequately deWned in terms of the absurdity of her update with f. Cf. Dekker (2002a) for details and further discussion. 31 Recall that f ! c is deWned as :(f ^ :c). If we use this deWnition of ! in the support and update calculus of Dekker (2002a, cf. the previous footnote), a speaker’s support for f ! c turns out to consist in her support for c after her update with f. 32 Cf. e.g. JackendoV (1972: ch. 6); Karttunen and Peters (1979); Stechow (1991); Rooth (1992); for a number of formally quite diVerent implementations of this distinction, which I, however, think are really close in spirit to the one I have in mind. Notice that, when I use the term ‘focus’, I do not mean ‘contrastive focus’ here.

Dekker, Grounding Dynamic Semantics

499

ally—that is, if context or intonation have no interfering eVects—one could say that, for example, the contents of negated sentences, the antecedents of implications, and the restrictions on quantiWers constitute a background or topic, which the speaker is not automatically supposed to support, but which she is supposed to react to. It is the focus part of her utterance which she can be required to have support for, possibly in functional dependence on such a background. Given this, it is no mystery that, by default, indeWnites in the background part of an assertion do not introduce possible referents for pronouns used in subsequent utterances. Since the speaker need not be required to support that part of her utterance, these indeWnites fall beyond her ‘pragmatic jurisdiction’, so to speak, and they are not assumed to be used with referential intentions. IndeWnites in focus, however, do require speaker’s support, and generally are associated with referential intentions. Since the focus may be functional on a background, the referential intentions associated with indeWnites in focus may be functional, too. Interestingly, this view upon the matter suggests or even predicts that anaphoric pronouns may refer back to such functional witnesses. And this ‘prediction’ is indeed borne out. Consider the following examples, known from the literature: (19) (20)

If a book is printed with Kluwer it has an index. It can always be found at the end. (After Heim) Harvey courts a girl at every convention. She usually comes to the banquet with him. (After Karttunen)

Support for an utterance of the Wrst sentence of (19) may consist of a witness function f , assigning indexes to books printed with Kluwer. If the speaker has such a function in mind, then she may refer back to it with a pronoun when she subsequently utters the second sentence. The second utterance then is assumed to be about books printed with Kluwer, too, and expresses that, always, if b is a book printed with Kluwer, then f (b) can be found at the end of b.33 Something essentially similar holds of example (20). Assuming that an utterance of the Wrst sentence of this example is supported by a function g, which associates girls which Harvey courts with the conventions he visits, an utterance of the second sentence refers back to it and states that for most c, if c is a convention which Harvey visits, then g(c) accompanies Harvey to the banquet of c. I note that these readings can be derived compositionally by combining the techniques from Jacobson (1999) with the account of anaphoric relationships sketched in this chapter.34 So, summing up the Wndings of this section, we may observe that the blocking eVects coming from, among others, negations, are indeed intuitively motiv-

33 The term ‘the end’ thus is supposed to be functional, too. 34 It may be needless to say that a functional reading of a pronoun is technically more complicated and that it imposes very speciWc demands on possible contexts of use. It is, thus, to be expected that they are much more marginal than non-functional ones. The same goes, mutatis mutandis, for the functional readings of (in)deWnites and Wh-items discussed by Jacobson, among many others (cf., the references in Jacobson 1999).

500

V. Anaphoric Pronouns and Dynamic Semantics

ated, and that, besides, the empirically attested leaks from such blocking contacts are indeed predicted.35

7. Conclusion In this chapter I have shown that indeed, as argued in Stalnaker (1998), cross-sentential anaphoric relationships can be accounted for on the basis of a classical, static, notion of meaning. A systematic elaboration of the observations made in that paper, however, forces us to take into account the dynamics of assertion, which, basically, consists in the fact that assertions are made, one by one, in a linear order. I have shown that the dynamics of interpretation can be located in the dynamics of conjunction. In the minimal system of PLA the dynamics of interpretation is modeled by a notion of conjunction which explicitly accounts for the fact that one conjunct literally precedes the other. Other systems of dynamic interpretation (Heim 1982; Groenendijk and Stokhof 1991; among many others) can be argued to account for the same phenomena in a diVerent manner, but with the same intuitive, and if you want philosophical, motivation. Interestingly, it is not so much the dynamics of interpretation that is puzzling, but, rather, the lack of it, in certain contexts. I have argued, like others before me, that deWnite and indeWnite terms generally are, or ought to be, used with referential intentions. IndeWnite terms, however, are typically not used this way when they Wgure in the scope of a negation, in the antecedent of a conditional sentence, or in the restriction of a quantiWer (adnominal or adverbial). So far, it has remained unclear in the literature why this is so. In the last section of this chapter I argued that the ‘blocking eVects’ of these constructions—which are certainly not unavoidable—relate to the typical role of these constructions in discourse and dialogue. Negations can be used to reject a proposition which has been raised in a discourse, explicitly, because somebody else asserted it, or implicitly, because it is deemed to be likely, or relevant; a conditional sentence can be used to expand upon the consequences of a possibility, stated by the antecedent of the conditional, which is deemed relevant; and quantiWers are typically used to assert something of their (presupposed) domain of quantiWcation. In either case the speaker can be held responsible for what she asserts about what is given, but not for what constitutes the background of her assertion. For this reason the speaker cannot (in

35 Notice that I have neglected, here, how the restriction of the quantiWers always and usually is accommodated in the examples (19) and (20). Arguably, this involves another instance of anaphoric resolution, for which there are several analyses on the market. Notice, too, that in whatever way we account for this kind of accommodation, it will also enable us to refer back to indeWnites in the antecedent of, for instance, the Wrst sentences of (19), by employing the identity function.

Dekker, Grounding Dynamic Semantics

501

general) be required to be able to supply witnesses for the indeWnites used in characterizing the background.36 Let me give a Wnal word to utterances other than assertions, viz. questions and orders. It seems that, typically, indeWnites in interrogatives and imperatives are used without referential intentions too. This blocking eVect can be given the same pragmatic explanation as the blocking eVect of backgrounds. For if someone asks: (21)

Is there is a hammer in the toolkit?

she is of course not required to have evidence that there is one; rather, she can be required not to know there to be one. Similarly, if she orders you: (22)

Put a red block on a blue one.

she is not supposed to have evidence supporting the fact that there is a red block on a blue one. Rather, for such an order to make sense, at least one red block should not be on a blue one. With questions and orders, there is, however, further reason for not using indeWnites with referential intentions. For if one asks: (23)

Is there a man in the house?

with a very particular man in mind, the speaker does not enable the hearer to answer appropriately. Given that, in such a counter-intuitive, or non-cooperative, case, the speaker does not specify whom she is asking a question about, the hearer cannot Wgure out which question the speaker really wants an answer to, and the question would be odd, at least. Similarly, if you order me to bring you a speciWc book from the table by uttering: (24)

Bring me a book from the table.

and if there are quite a few of them, I cannot carry out the desired request if I fail to know which book you want. Also in these cases we see that pragmatic principles, this time those concerning questions and requests, conspire so as to undo the eVects of principles governing the use of indeWnite noun-phrases. To conclude, let me recapitulate what I have done in this chapter. Systems of discourse representation and dynamic semantics have been developed to account for, among other things, inter-sentential anaphoric connections between indeWnite nounphrases and anaphoric pronouns. Apparently, the success of these systems derives from the fact that they adopt an essentially richer notion of meaning, a representational and a dynamic one, respectively. Following Stalnaker (1998), I have argued that the phenomena do not, however, require such an extended notion of meaning. Pronouns can 36 Of course, this is diVerent in the case of deWnite terms, like pronouns, proper names, and deWnite descriptions. DeWnite terms are genuinely presuppositional, and cannot be used without referential intentions. Even under a Russellian analysis of their semantics, these terms ought to be used to refer, if not to deWnite individuals, than at least to deWnite individual concepts.

502

V. Anaphoric Pronouns and Dynamic Semantics

be taken to be ordinary referring devices, which can refer anaphorically to individuals which have been made ‘intentionally present’ by previously used indeWnite nounphrases. However, in contrast with Stalnaker, I next argued that a systematic account of the interpretation of discourse after all does provide motivation for a dynamic notion of interpretation. If indeWnites indeed can be used so as to provide, in an indeWnite way, possible referents for subsequent pronouns, and if our system of interpretation aims to account for that, then the possible referents must be taken up in some way, and be dealt with in an orderly fashion. Systems of discourse representation and dynamic semantics then can be seen to do precisely this. By means of the dynamic system of interpretation of PLA, I have shown that a systematic and dynamic account of the data indeed can be seen to derive from a classical semantics, extended so as to take (the eVects of ) certain pragmatic principles into account. Classical semantic properties of sentences are dealt with in PLA in tandem with an account of systematic pragmatic facts related to their use. As we have seen, all non-standard features of the PLA satisfaction relation have been traced back to pragmatic principles related to the use of its formulae, or, rather, of the natural language sentences whose contents these formulae can be taken to model. By doing so, I have also shown that the dynamics of interpretation can be attributed, entirely, to the dynamics of conjoining assertions, one after the other.

15 Pronouns as DeWnites Craige Roberts

1. Introduction Pronouns have very little descriptive content, yet they are deWnite. Because they are deWnite, they participate in a web of anaphoric relations in discourse, relations mediated by the structured information which competent interlocutors control as they converse. An anaphoric expression harkens back not to another expression, such as a noun-phrase (NP) in prior discourse, but to a discourse referent, an element of that structured information. If a pronoun seems to refer deictically, or to be bound by a quantiWcational operator, or to borrow the descriptive content of another NP, this is because of the relationship between the corresponding discourse referent and the entity picked out deictically, the quantiWcational operator, or the other NP. In this chapter, I will Wrst spell out in more detail the view just sketched, which emerges from the theory of deWnite noun-phrases proposed in Roberts (2003).1 Then I will compare it to a prominent class of proposals in which it is claimed that pronouns are ambiguous, and that one interpretation of a pronoun is that of a disguised deWnite description, with the interpretation of Russell (1905). I will argue that such an interpretation is neither necessary to explain the relevant uses of pronouns, nor suYcient to capture certain restrictions on when a pronoun may apparently have such a use. Further, I will argue that the uniqueness associated with the Russellian interpretation is too strong for pronouns, as well as for deWnites more generally. With the simpler theory of pronouns as anaphoric on discourse referents, we have improved empirical coverage with greater simplicity, while preserving the valid intuitions underlying the proposal that pronouns may be interpreted as deWnite descriptions.

Thanks to Anne Bezuidenhout for suggesting that I write this chapter, and for her excellent comments and editorial assistance. Thanks as well to an anonymous reviewer for additional helpful suggestions, and to Ilse Lehiste and Sigrun Svavarsdottir for their encouragement. 1 The reader familiar with Roberts (2003) can skip section 2.

504

V. Anaphoric Pronouns and Dynamic Semantics

2. DeWnites in Information Structure The idea that semantic interpretation is mediated by the structure of discourse is a relatively new one, which can be traced in the formal semantics literature to Stalnaker (1978), Lewis (1983b), Gazdar (1979), Kamp (1981), and Heim (1982). The basic idea is that the interpretation of an utterance is not directly a proposition, but a function from the context of utterance to an updated context, which latter in turn serves as the argument of the next utterance. Informally, we can think of the content of an expression of a natural language like English as the additional information it contributes to the growing common ground of information that the interlocutors share. In the simplest version of such theories, known as dynamic theories of interpretation because of the way they model the growing context, the information structure of discourse is characterized formally as a set of propositions, each proposition a set of worlds or situations. But recent theories of discourse information structure (Heim 1982; Kamp and Reyle 1993; Groenendijk and Stokhof 1990; Chierchia 1995; Roberts 1996b; Asher and Lascarides 1998a, b) include other kinds of information besides the propositional. For example, Roberts (1996b) assumes that the fundamental organizing principle of discourse is the set of questions under discussion, with the underlying intentions they realize; information structure is a tuple consisting of the structured set of questions under discussion (a push-down store), the propositions in the common ground, the familiar discourse referents, and other types of information. In all these theories, the truth-conditional, propositional content of a set of utterances can be retrieved straightforwardly, but meaning cannot be reduced to propositional content. From this perspective, the conventionally given semantic content of any given expression has two main facets: its presupposed content and its proVered content. The latter corresponds roughly to what we think of as its contribution to the truth-conditional, propositional content of an utterance in which it occurs. The presupposed content constitutes the conventionally given constraints that utterance of the expression places on context. In order for an utterance containing the expression to be felicitous in a given context, these requirements on the context must be met. What is the semantic content of a pronoun, then—say English he, it, she, or they? These NPs are abbreviated ways of saying something about some entity that is already familiar and salient to the interlocutors. In actual discourse, we determine what a competent speaker intends to refer to in using a pronoun by considering the context. Within the formal theory of semantic interpretation in information structure, we use a fairly simple mechanism to model how this works. The proVered content of a pronoun is just a variable whose semantic value in a particular context is always given in the same way, via a contextually speciWed function assigning values to variables. In a modeltheoretic interpretation, these values will be entities in a model for the language. As in Kamp (1981) and Heim (1982), if it appears that a pronoun is sometimes bound by a

Roberts, Pronouns as Deﬁnites

505

quantiWcational operator, while at other times it is demonstrative or is co-referential with a proper name or other referring expression, this is because there are many ways in which the assignment functions that we use for the interpretation of variables can be constrained by the discourse context. One of these parallels Tarski’s use of assignment functions for the interpretation of quantiWcational expressions, yielding a bound variable interpretation. But unlike their use in classical standard Tarskian semantics, assignment functions may be constrained in other ways, and accordingly, there are other ways to constrain the values assigned to the variables which are the proVered content of a pronoun besides varying over the domain of an operator. One of the most striking features of theories of dynamic interpretation is the uses of assignment functions to encode the interlocutors’ discourse referents. Karttunen (1976) pointed out that we keep track of information in discourse about a number of individuals other than those we take to be actual entities in the world: for example, hypothetical individuals, or instantiations of quantiWcational statements. And we may keep track of information about someone we believe to be real, not knowing whether this individual is the same as some other with whom we are familiar, perhaps later merging information about the two. To distinguish these bundles of information from actual entities in the world, he called the former discourse referents. Heim (1982) and Kamp (1981) gave technical expression to this notion by modeling discourse referents as constraints on the assignment functions interlocutors can use to interpret variables at a given point in discourse. We assign an ‘address’ to each bundle of information we take to be about a single (actual or hypothetical) individual, the address being a particular variable, say x28. Then we require that any assignment function we use to interpret utterances at that point in the discourse be such that whatever it assigns as the value of x28 be an individual which satisWes all the information we have about that (actual or hypothetical) individual discourse referent. Hence, instead of assignment functions being arbitrarily chosen, they are used to encode the information about discourse referents, permitting a continuity of reference. Across discourse, these discourse referents, in the technical form of the set of permissible assignment functions, are managed dynamically to reXect the way that information changes. If we add more information about a particular discourse referent, the permissible assignment functions will be further constrained. If the discourse referent is merely an instantiation of a generalization or, say, hypothetical, triggered by information under the scope of a modal or other operator, then once the hypothetical mode has been closed oV, outside the scope of that operator, the discourse referent is no longer available for reference, and accordingly, the relevant constraints on assignment functions no longer hold. This is the key to interpretation of donkey pronouns in these frameworks, as in (1): (1)

Every man that owns a donkey beats it.

The discourse referent whose introduction into the information is triggered by utterance of a donkey in (1) will be available only under the wider scope of the universally

506

V. Anaphoric Pronouns and Dynamic Semantics

quantiWed subject. Hence it can bind the pronoun it (often called the ‘donkey pronoun’) in the predicate, but normally is not available for reference after that. For example, we cannot felicitously continue (1) with I thought it looked miserable yesterday.2 Since the proVered content of a pronoun is a variable, the value assigned will be constrained by any information associated with that variable by the interlocutors at that point in the discourse. Moreover, utterance of a pronoun presupposes that there is already a discourse referent corresponding to that variable in the information structure. This is what Heim (1982) called the familiarity presupposition associated with a pronoun. A speaker cannot just walk in out of the blue and say She wants a new car. To be felicitous, this would have to be the continuation of some conversation we were having about some particular female individual. This is in keeping with our fundamental intuition that a pronoun is anaphoric: it carries us back to information presupposed to be already in discourse, and its presupposed content puts constraints on where to look for this information, thus giving clues to what the speaker intends. In this sense, the true antecedent of a pronoun is the corresponding discourse referent, an element in the information structure of the discourse. The pronoun’s person, number and gender put additional constraints on where we should look for an antecedent (Dowty and Jacobson 1989). If the speaker uses, say, a third-person feminine singular pronoun, the addressee would look for a discourse referent not associated with the speaker or addressee, and about which there is information in the common ground that is compatible with using the feminine grammatical gender and singular grammatical number. This compatibility does not require that the entity referred to be sexually female or semantically singular. We may refer to a boat or plane with she, a dog of unknown gender with it, and we often use they as a gender-neutral alternative to he or she, etc. And even apart from gender-neutral plurals, the rules for using number are notoriously rather complex, as will be brieXy discussed below. However, the person, number, and gender of a pronoun impose constraints which help the addressee to identify the familiar discourse referent to which the speaker intends to refer. These constraints constitute the complete descriptive content of a pronoun. As argued in Heim (1982), Kadmon (1987, 1990), and Roberts (2003), this content, like the descriptive content of a deWnite description, is presupposed, not proVered. But there is more to the presupposed content of a pronoun. It has often been argued in the semantic literature that pronouns may have the interpretation of a disguised deWnite description, with the existence and uniqueness argued for deWnite descriptions 2 Note that the above does not assume that quantiWers are unselective, as was the case in Kamp (1981) and Heim (1982). This facet of these early theories of dynamic interpretation has been discredited as leading, in the general case, to the wrong interpretation, though there still may be a sense in which it is appropriate in the cases for which it was originally proposed in Lewis (1975). See Kadmon (1987, 1990), Chierchia (1995), and references cited therein, and the discussion of the proportion problem below. As Groenendijk and Stokhof (1990) and Chierchia (1995) illustrate, dynamic theories are perfectly compatible with more classical, selective binding by operators.

Roberts, Pronouns as Deﬁnites

507

in Russell (1905). And in a sense that is true. Roberts (2003) argues that all deWnites, including deWnite descriptions, personal pronouns, and demonstratives, have both familiarity presuppositions and uniqueness presuppositions. The familiarity presupposition of a deWnite corresponds to the existence clause of a Russellian deWnite description, except that the ‘existence’ in question is only that of a discourse referent in the information structure of the discourse. As with Russellian deWnites, this may or may not turn out to be a claim about existence in the actual world, depending on the force and scope of other operators in the discourse, intensionality, and other factors. The uniqueness presupposition is informational as well, unlike the Russellian requirement of semantic uniqueness, and hence it too is weaker than Russell’s. Russell’s uniqueness is semantic in that it requires that the intended referent be unique in the actual world in bearing the property in question. In contrast, informational uniqueness is a requirement that there be no more than one discourse referent in the interlocutors’ common ground which satisWes the deWnite’s descriptive content. But of course, though there is only one such discourse referent in the common ground, this still leaves open the possibility that there is more than one entity of the relevant sort in the world that the interlocutors just are not (mutually) familiar with. So informational uniqueness does not guarantee semantic uniqueness. Here is the informal statement of the presuppositions associated with deWnite NPs in general, from Roberts (2003):3 (2)

Informational Existence and Uniqueness of DeWnite NPs Given a context C , use of a deWnite NPi presupposes that it has as antecedent a discourse referent xi which is: a. weakly familiar in C , and b. unique among discourse referents in C in being contextually entailed to satisfy the descriptive content of NPi .

There is more to say about the presupposed content of pronouns. But Wrst, let us consider in more detail the evidence that they are deWnite, so that they carry the presuppositions in (2). In the next subsection I will consider the claim in (2a) that deWnites are weakly familiar, deWning that notion and showing that it applies to pronouns as well as to deWnite descriptions. Next I will consider (2b), exploring how the uniqueness required diVers from the uniqueness clause in the Russellian truth conditions for deWnite descriptions. Then I will turn to the respects in which pronouns diVer from deWnite descriptions, motivating additional presuppositions for pronouns.

Weak Familiarity (2a) requires that a deWnite NP be only weakly familiar. What this means is that the relevant discourse referent need not have been explicitly introduced into the common 3 Please see that paper for a formal statement, and a great deal more detailed argument for the proposed familiarity and uniqueness presuppositions than space permits here.

508

V. Anaphoric Pronouns and Dynamic Semantics

ground through the utterance of an NP. Instead, it need only be the case that the common ground entails existence. We see some examples in the following: (3) (4) (5) (6) (7) (8) (9) (10)

(To a stranger at a bus stop) The sun is really hot today! (To a stranger at an auction, when a new drawing has just been brought up to the block) It’s really unusual for a Rubens, don’t you think? (At a coat check, as the clerk hands an umbrella to the preceding person in line) That umbrella belongs to me! (Same situation, as the clerk prepares to hand a laptop computer case to the preceding person) That belongs to me! Joan was murdered. The murderer was probably an acquaintance. Each apartment has a large bay window in the living room. In this one (pointing to a unit on a blueprint), it looks out over the golf course. Suppose Joan had been murdered. If the murderer had taken her Rolex, he would probably have pawned it. (Architects, brainstorming about a proposed project) It would be great if every apartment had a large bay window in the living room. In some, it would look out over the golf course, in others, over the more intimate garden courtyard.

As illustrated by (3) and (4), even interlocutors who are strangers to one another can generally assume that they have certain information in common, including basic information about the world they inhabit (there is a sun) and information about the immediate situation in which they Wnd themselves (there is a drawing on the auction block). The utterances in (5) and (6) may or may not be accompanied by pointing, but in any case pertain to an object which is made maximally salient in virtue of the type of situation and the intentions of those present.4 In (7) and (8), presumably the murderer or the bay window in the particular apartment in question had not been previously mentioned, nor would the addressee have had reason to believe that they existed prior to the Wrst utterance. However, the context, that Wrst utterance, entails their existence: There is no murder without a murderer; we know that each apartment has a bay window. And (9) and (10) similarly illustrate how use of a deWnite can be licensed by entailment. In these cases, however, we Wnd these entailments in irrealis discourse; they are cases of modal subordination.5 The proposition that the conditional in (9) is generally taken to express requires that the modal would be restricted to range over worlds or situations in which Joan was murdered (the irrealis assumption urged in the Wrst sentence); since being murdered entails the existence of a murderer, we can assume

4 See Roberts (2002) for detailed discussion of demonstratives, and arguments that they are deWnite in the sense described here. 5 See Roberts (1989, 1990, 1996a) for analysis of this general phenomenon and arguments that it involves accommodating a relevant domain restriction for the modal (or other intensional operator) in the second sentence.

Roberts, Pronouns as Deﬁnites

509

that there is such an individual in each of the relevant worlds/situations, making the murderer weakly familiar. In examples like (8) and (10), there is clearly a preceding NP—a large bay window— that licenses the use of the pronoun. However, this NP is not an antecedent for the pronoun in the usual technical sense, requiring that the two NPs are co-referential or are both bound by the same operator. Co-reference or co-binding would require that any operator with scope over the antecedent take scope over the pronoun as well; but this is not the case here, where every apartment has scope over the indeWnite but not the pronoun. In examples like (8), let us call the preceding indeWnite NP whose descriptive content helps to determine the interpretation of the pronoun the licensing NP. Basing weak familiarity on entailment permits us to explain examples involving double negation, like (11): (11)

It is not (the case) that Mary doesn’t own a car. It is red and it is parked in front of her house.

And entailment based on world knowledge may be combined, for example, with deixis, to yield examples like (12): (12)

A: Can you help me get to the Central Station fast? B: (pointing to a car) The keys are in the ignition. Help yourself.

Here, A can be expected to know that the car made familiar by deixis has both an ignition and a set of keys, since that is the default case with cars in the contemporary world. The way that various kinds of information, including discourse referents, are introduced and managed in the information structure of discourse (Heim 1982, Kamp and Reyle 1993, Roberts 1989, Chierchia 1995) guarantees that the following principles hold, with an exception to be discussed below: (13)

Scope Principles a. The scope of an operator is conWned to (at most) the sentence in which it is uttered, i.e. operator scope is syntactically restricted. b. Information introduced within the scope of an operator is only accessible within that scope.

(14) exempliWes how the principles in (13) generally constrain felicity in discourse: (14)

Mary is considering getting her Wrst dog. #But she regrets becoming a dog-owner.

Here, the second sentence is infelicitous in this context because the presupposition of factive regret, that Mary has become a dog-owner, is not satisWed in the context. In fact, it is implicated to be false, due to the fact that a precondition for becoming a dogowner, getting a dog, occurs under the scope of a future-oriented intensional predicate in the Wrst sentence.

510

V. Anaphoric Pronouns and Dynamic Semantics

The important exception noted above is to principle (13a): this holds for most operators, but within theories based on Dynamic Montague Grammar (Groenendijk and Stokhof 1990), like that of Chierchia (1995), the existential operator introduced by indeWnite NPs may essentially take scope across discourse. The other theories noted do not make this exception, but instead make indeWnites trigger the introduction of discourse referents whose familiarity endures across discourse. This diVerence should be kept in mind in what follows, but the exception in the DMG theories does not undermine the general validity of (13a).6 From the general principle (13b) follows the more speciWc instance in (15), reXected in (16) and (17): (15)

(16) (17)

Accessibility of Discourse Referents A discourse referent which is introduced under the scope of an operator is inaccessible to serve as an antecedent beyond the scope of that operator. A: Suppose Joan had been murdered. B: Do you think the murderer took her Rolex? (Architects, brainstorming about a proposed project) It would be great if every apartment had a large bay window in the living room. #It looks out on a golf course.

Roberts (1989) argues that in felicitous cases like (9) and (10) above, there is an extension of the irrealis context in the Wrst utterance via a modal (would in these cases). This extension, which she calls the hypothetical common ground, restricts the domain of the modal having scope over the pronoun, and because the hypothetical murderer or bay window is familiar in this hypothetical common ground, this licenses the use of the deWnite the murderer or the pronoun it. (16) and (17) show that the weak familiarity in the irrealis cases involving modal subordination is not global, but only holds in the hypothetical common ground. So use of the murderer by B in (16) is infelicitous unless we are prepared to accommodate that the supposition proposed by A was, in fact, true. And the familiarity presupposition of the italicized pronoun in (17) fails because the only plausible antecedent, a large bay window, is under the scope of a modal, while the pronoun is in a realis context. All this is not to deny that there are cases where a pronoun is felicitous despite the fact that familiarity fails. There are cases where the speaker requires that their addressee straightforwardly accommodate the relevant discourse referent into the common ground. In this respect, familiarity presuppositions behave like other kinds of presuppositions, as we see illustrated here:

6 This principle has been called into question, e.g. in the technically ingenious work of Dekker (1993). However, it is generally regarded as valid, and I will assume it here without further argument. If we do not assume such a principle, with the possible exception of the existential indeWnites, we will be very hard-pressed to account for most semantic judgments involving scope in discourse.

Roberts, Pronouns as Deﬁnites (18)

(19)

511

(A and B are waiting in a crowded grocery line; B is distractedly looking at the tabloids on a rack while the small child in his cart kicks A repeatedly.) A: Please ask your child to stop kicking me. (The speaker is on the phone with a bank oYcer who she’s never talked with before) There’s a problem with my husband ’s checking account.

Though B in (18) apparently does not realize that his child is kicking A, if he is cooperative he will accommodate the factive presupposition of the verb stop (i.e. that the complement is true) and interpret A’s utterance accordingly. In (19), though the bank oYcer presumably did not know beforehand that the stranger A had a husband, he will undoubtedly conclude that she knows whether she has one, and will accommodate the familiarity presupposition of my husband, by adding the appropriate discourse referent to the common ground up to that point and treating the result as the context of utterance. There are conditions on when we may accommodate a presupposition: (20)

Necessary Conditions on Presupposition Accommodation a. Retrievability: what the hearer is to accommodate is easily inferable, so that it is perfectly clear what is presupposed, and it is both salient and Relevant to the immediate context, and b. Plausibility: the accommodated material leads to an interpretation that is reasonable and unobjectionable in the context.

The sense of Relevance here is a technical implementation of the Relevance of Grice (1975) proposed in Roberts (1996b). BrieXy, to be Relevant is to address the immediate question under discussion in the information structure of the discourse. An assertion addresses a question if it contextually entails at least a partial answer to the question, while a question q1 addresses another question q2 if the complete answer to q1 contextually entails at least a partial answer to q2 . Generally, so long as a presupposition is both retrievable and plausible, the cooperative addressee will accommodate without mentioning it. But under the proposed account of familiarity, where explicit mention is not required, accommodation of familiarity presuppositions is both less common and more restricted than has often been assumed by those who have argued against familiarity presuppositions by deWnites.7 See Roberts (1995, 1996a) for extended discussion of the ways that accommodation is restricted in these cases. Only note here that accommodation must leave the common ground consistent. For example, consider the following: (21) (22)

a. An earthquake might hit San Francisco. b. That would upset me. It would be frightening. a. An earthquake hit San Francisco in 1989. b. That upset me. It was frightening. 7 See Fraurud (1990), Birner and Ward (1994), Poesio and Vieira (1997), inter alia.

512

V. Anaphoric Pronouns and Dynamic Semantics (23)

a. An earthquake might hit San Francisco. b. That upsets me. It is frightening.

In (21a), the utterance is non-factual in mood, talking about an event which is as yet only a possibility. The subject of (21b), under the scope of would, seems to refer to an earthquake hitting San Francisco, that is, a nominalization of the proposition denoted by the nuclear scope of the modal might in (21a). In (22a), the utterance is factual, describing a real event. The subjects of the factual sentences in (22b) refer to an earthquake hitting San Francisco in 1989, the fact corresponding to that event. Notice that in (22b) it is acceptable to say ‘That fact upset me’, whereas in (21b), one cannot say ‘That fact would upset me.’ In (23), after the same nonfactual utterance as in (21a), one can utter the same sentences as in (22b), and even substitute ‘That fact upsets me’, but the meaning is not the same. (23b) means that the possibility of an earthquake hitting San Francisco (where a possibility is a kind of fact) upsets the speaker and is terrible, not the fact of an earthquake hitting San Francisco. In (21a) and (23a), the utterance is irrealis, so that the only salient event is a possibility, implicating that it is not yet reality. Then talking about it as if it were a reality in (b) would sound contradictory; hence, the understood antecedent of that and it in (b) is constrained to be either in an irrealis context, as in (21b) under the scope of would, or refer to a possibility, not a fact, as in (23b).

Informational Uniqueness and Semantic Uniqueness (2b) requires that the discourse referent which satisWes the familiarity presupposition associated with the use of a deWnite be unique among discourse referents in the common ground in being contextually entailed to satisfy the descriptive content of the deWnite. This is, of course, a weaker requirement than the semantic uniqueness of the Russellian interpretation of deWnites. Is this strong enough? With respect to semantic uniqueness, Kadmon (1990) oVers a convincing argument that uniqueness (for her, semantic uniqueness) is presupposed, as opposed to proVered, based on examples like (24). (The parenthetical material is included to motivate the use of the deWnite description in the conditional antecedent, instead of a pronoun; without that material, we could substitute he.) (24)

A strange man lives here (and another one lives down the block). If the strange man who lives here sees a cat, he screams.

She points out that, if we take the meaning of the deWnite in the antecedent of the conditional to involve entailed uniqueness, then we arrive at an interpretation which is too weak: ‘if there is a unique strange man who lives here and sees a cat, he screams’, where the antecedent of the conditional is only true if it happens that there is no more than one strange man living there. Instead, the use of the deWnite in the conditional appears to presuppose that the male referred to is unique in satisfying

Roberts, Pronouns as Deﬁnites

513

the description of the antecedent and its properties, and the conditional asserts that, if this person sees a cat, he screams. So the appropriate comparison here is between two sorts of presuppositions, those of semantic uniqueness and those of informational uniqueness. The following, after examples in Heim (1982), argue that semantic uniqueness is too strong: (25)

(26)

Mary is quite upset. A wine glass was broken at the party last night. The glass had been a wedding gift, and it will be diYcult to replace. Actually, several others broke as well, but she only knows about the one. Everyone who bought a sage plant or a rosemary planted the sage plant with extra bonemeal or the rosemary in a well-limed soil. And if it was a sage plant, they bought a second one along with it.

The underlined deWnite descriptions and pronouns in these examples fail to satisfy semantic uniqueness (even under quantiWcation, in (26)), yet there is no sense that they are thereby false or infelicitous. They do satisfy informational uniqueness at the time of utterance: before the Wnal sentence in (25) is uttered, there is only one broken wine glass under discussion, and we only know of one sage plant per person before uttering the Wnal sentence of (26). There are, of course, what Roberts (2003) calls uniqueness eVects, where use of a deWnite, especially a deWnite description, conveys semantic uniqueness. The most famous examples, including Russell’s (27), involve titles: (27)

The king of France is bald.

Roberts (2003) argues that uniqueness eVects arise with titles precisely because they are designed with a view to being able to uniquely pick out the bearer of the relevant social role, whatever the common ground of possible interlocutors might be at any given time. In the satisfaction-based theory of presupposition assumed here (Karttunen 1973; Stalnaker 1974; Heim 1983), presupposition is stronger than entailment, explaining Russell’s intuition that examples like (27) are false: if there fails to be a familiar discourse referent in the common ground uniquely known to be a king of France, then (27) cannot be truthfully added to that common ground. Worse, if we know there is no king of France, then that addition would be contradictory. And if there were two or more kings of France, utterance of (27) would involve making a false presupposition. There is no evidence that we have direct intuitions of the diVerence between a presupposition and an entailment (otherwise, the theoretical argument would have been settled long ago!), so in this last case all the hearer knows is that the utterance has given rise to a falsehood. Something similar is going on with directions, as in (28): (28)

(Directions to a tourist) Once you’re in the square, turn right at the obelisk.

514

V. Anaphoric Pronouns and Dynamic Semantics

Directions are designed to be used in a context in which the user will likely have perceptual access to full information about the situation in question. When the tourist enters the square, if there were two obelisks, each giving rise to a discourse referent for an obelisk in the common ground she shares with the giver of directions, the directions would fail informational uniqueness, and would thereby fail to be optimally useful. Roberts examines other types of cases where uniqueness eVects arise and concludes that there are always conversational factors to which the eVects can be attributed. Hence, informational uniqueness is suYciently weak to be compatible with the failure of semantic uniqueness in the general case, while strong enough to explain the evident constraints on deWnites which go beyond familiarity (Kadmon 1990; Heim 1990; Neale 1990), including both the potential for uniqueness eVects and the diVerences between deWnites and indeWnites. In the latter connection, note that familiarity alone does not suYce to diVerentiate deWnites from indeWnites. For even if my interlocutor knows nothing about West Virginia, including that it has mountains, I cannot describe my weekend activities using (29), but must use a deWnite as in (30): (29) (30)

#Last weekend we climbed a biggest mountain in West Virginia. Last weekend we climbed the biggest mountain in West Virginia.

Superlatives and other NPs whose semantics entail that they are semantically unique can never be indeWnite, because use of the indeWnite article not only presupposes novelty (non-familiarity) in the common ground (Heim 1982), but also implicates that for all the speaker knows there are other discourse referents bearing the same descriptive content; see Roberts (2003) for discussion of how this implicature arises. Since this implicature would be contradictory in such cases, the deWnite article must be used.8 We also Wnd evidence for informational uniqueness in constraints on accommodation, as seen in the diVerence between (31) and (32): (31)

(32)

(Spoken to a business associate who knows very little about the speaker) So sorry to be late. I had to take the car to the garage because the gear shift was stuck. (Same context) So sorry to be late. #I had to take the car to the garage because the tire was Xat.

In (31), uttered with a minimal prior common ground involving only generic knowledge about cars, repair, and gear shifts, the deWnites the car, the garage, and the gear shift all have familiarity presuppositions which require accommodation. On the basis of the fact that people generally take care of their own cars, the addressee would reasonably assume in this minimal context that the car in question must be the speaker’s, accommodating a discourse referent for it, and moreover, given the default assumption that 8 The theory of conversational implicature assumed here diVers from that of Grice in assuming that these implicatures are not cancelable. See Welker (1994). Hence, the robust character of this implicature.

Roberts, Pronouns as Deﬁnites

515

people only have one car, would take that discourse referent to be informationally unique. We also cross-reference the garage in question to the speaker, assuming that, since people usually try to frequent a single repair shop, this must be the unique garage she regularly goes to. This cross-referencing involves giving the NP an implicit relational reading (see Clark 1975; Chierchia 1995). On the basis of our knowledge of gear shifts and the assumption of Relevance, we cross-reference the discourse referent for the gear shift with that for the speaker’s car, knowing that a given car has a unique gear shift. The accommodated discourse referents for the Wrst two deWnites, as well as that for the weakly familiar (entailed) third deWnite therefore all satisfy informational uniqueness, and the utterance is felicitous. But in (32), we know that a given car has four tires, so informational uniqueness fails, and the utterance sounds odd. It would be preferable to say because I had a Xat tire. Finally, note that unlike deWnite descriptions, pronouns almost never give rise to uniqueness eVects. There are many examples discussed in the literature where semantic uniqueness would yield the wrong interpretation for pronouns. For example, (33), from Rooth (1987), may be true even though many parents have more than one son still in high school: (33)

No parent with a son still in high school has ever lent him the car on a weeknight.

However, Evans (1977, 1980) argued that in examples like (34), pronouns do give rise to semantic uniqueness, claiming that it is this that diVerentiates them from examples with relative clauses, like (35): (34) (35)

John has some sheep. Harry vaccinates them. John has some sheep that Harry vaccinates.

First, note that in plurals, uniqueness amounts to maximality, that is to the claim that, say, the NP’s denotation is maximal in containing all the individuals of the relevant sort. In these examples, it does seem that there is a diVerence in meaning between (34) and (35), so that (34) appears to convey something like ‘Harry vaccinates all of John’s sheep’. However, note that (34) can be felicitously followed by material which entails the failure of semantic uniqueness, as in (36): (36)

John has some sheep. Harry vaccinates them. John has some others out to pasture in Cornwall, but Harry won’t go all the way over there.

What then is the diVerence between (34) and (35)? The use in (35) of an indeWnite with a more elaborate descriptive content implicates that John has other sheep besides those that Harry vaccinates. Otherwise, one could use the simpler some sheep, as in (34). (34) by itself has no such implication; so, if nothing else is said, one might assume that the sheep in question are all that John has. The fact that the entailed denial of maximality here does not seem like a correction or contradiction shows that semantic uniqueness would be too strong.

516

V. Anaphoric Pronouns and Dynamic Semantics

Salience Presuppositions of Pronouns Pronouns diVer from deWnite descriptions in other ways besides failing to give rise to uniqueness eVects. We see this in examples like the following: (37)

(38)

A woman entered from stage left. Another woman entered from stage right. p p #The woman/ The FIRSTwoman/ The SECOND woman was carrying a basket of Xowers. A woman entered from stage left. Another woman entered from stage right. p She was carrying a basket of Xowers, while #the woman/ the FIRST woman/#the SECOND woman led a goat.

In (37) and (38), the Wrst two sentences set a scene in which there are apparently two women, distinguished only by which side of the stage they have entered from. We see in (37) that we cannot felicitously use the deWnite description the woman, apparently because it refers non-uniquely in the scene in question—there are two salient women, two corresponding discourse referents. The more speciWc NPs are felicitous; Wrst and second may either be taken to allude to the order of a woman’s entrance on the stage or the order of our mention of her and her entrance, with the same results. (38) shows that the pronoun she, while even less contentful than the woman, may be felicitously used, but it can apparently only refer to the second woman, as shown by the possible references to the other woman in the adjunct clause. Note that we cannot argue that the Wrst woman is simply not salient by the time we interpret the pronoun; in the same relative linear position in discourse, she in (39) can take the Wrst NP as its antecedent: (39)

A woman entered from stage left. There was a basket of Xowers in the middle of the stage. She picked it up.

The diVerence in (38) seems to be that a pronoun takes as its antecedent the most salient entity in the context at the time of its utterance which is of a sort compatible with the pronoun’s features. But a deWnite description does not take relative salience of potential antecedents into account in this way. On the basis of these, plus a number of other types of examples, as well as empirical research on how we process pronouns versus deWnite descriptions, Roberts (2003) argues that the central diVerence between pronouns and deWnite descriptions is that pronouns presuppose that their discourse referent antecedents are maximally salient, whereas that is not necessarily the case with deWnite descriptions. The additional presupposition of salience yields (40), which is a special instance of the presuppositions for deWnites in (2):

Roberts, Pronouns as Deﬁnites (40)

517

Presuppositions of Pronouns Given a context C , use of a pronoun Proi presupposes that it has as antecedent a discourse referent xi which is: a. weakly familiar in C , b. salient in C , and c. unique in being the most salient discourse referent in C which is contextually entailed to satisfy the descriptive content suggested by the person, number and gender of Proi .

(40) is just like (2) through clause (a). Clause (b) is added to reXect the salience presupposition; for more on what constitutes salience in discourse, see Roberts (2003). The uniqueness presupposition in (2c) is specialized here to take into account the fact that the discourse referent for a pronoun need be informationally unique in satisfying its descriptive content only among the set of salient discourse referents, rather than, as is the case for deWnite descriptions, among all discourse referents. This yields an explanation for why pronouns do not generally display uniqueness eVects: The range of possible antecedents is so restricted that the uniqueness presupposition clearly only serves to pick out a uniquely intended antecedent. (41)–(45) illustrate the importance of the salience presupposition. In each, the discourse referent antecedent for the deWnite is weakly familiar because entailed to exist by preceding context. The Wrst sentences in (41)–(43) (related to examples due to Barbara Partee, cited in Heim 1982) are truth-conditionally equivalent, all entailing the existence of a missing marble. This is suYcient to licence the deWnite description in (41), though the familiarity is merely weak. But weak familiarity does not guarantee salience, and this is why the pronoun is infelicitous in (43). When the missing marble is explicitly mentioned, and hence salient, in the Wrst sentence of (42), use of the pronoun is felicitous. We see the same phenomenon in the contrast between (44) from Heim (1990a), where the pronoun is infelicitous because the bicycle is merely weakly familiar, and (45), where the deWnite description is felicitous in the same context: (41) (42) (43) (44) (45)

I dropped ten marbles and found only nine of them. The missing marble is probably under the sofa. I dropped ten marbles and found all of them, except for one. It is probably under the sofa. I dropped ten marbles and found only nine of them. #It is probably under the sofa. In Amsterdam, if a cyclist isn’t very careful, it ’ll be stolen. In Amsterdam, if a cyclist isn’t very careful, her bicycle will be stolen.

The relatively impoverished descriptive content of pronouns also makes it more diYcult to accommodate a discourse referent antecedent for a pronoun than for a deWnite description, as we see in the following contrast:

518

V. Anaphoric Pronouns and Dynamic Semantics (46) (47)

A: Can you help me get to the Central Station fast? B: (Pointing to a car) The keys are in the ignition. Help yourself. A: Can you help me get to the station, fast? B: (Pointing to a car) #They’re in the ignition. Help yourself.

The italicized pronoun in (47) fails to make the required accommodation retrievable, hence failing (20a). And, again, in such cases the intended antecedent is not suYciently salient to satisfy (40b), either. Hence, use of pronouns without an explicit antecedent most often requires either maximal situational salience, as we saw in (4), or deixis, as in the example in (6). Hence, pronouns diVer from deWnite descriptions in two important respects, both stemming from the fact that their descriptive content is so minimal: They do not generally give rise to uniqueness eVects, which we Wnd with some uses of deWnite descriptions, and they require maximal salience of their discourse referent antecedents, which is not required with deWnite descriptions. The theory outlined here makes the Wrst diVerence stem from the second. Any theory which collapses pronouns, in any of their uses, with deWnite descriptions, fails to capture these important distinctions.

3. Theories of Pronouns as Disguised DeWnite Descriptions A number of authors have claimed that pronouns have at least one interpretation in which they are treated as disguised, Russellian deWnite descriptions. DiVerent authors mean slightly diVerent things by this, only some of them truly reducing (some) pronouns to deWnite descriptions. As we just saw, there is now reason to doubt whether such theories will be adequate. Though pronouns are undoubtedly closely related semantically to deWnite descriptions, are themselves deWnite, and because they are anaphoric often borrow part of their understood descriptive content from other (sometimes deWnite) NPs, they are quite diVerent from deWnite descriptions in other respects. In this section, I will brieXy consider the speciWc content of several of the relevant theories, and then oVer additional arguments that, in order to be empirically adequate, the theories would have to be modiWed along the lines suggested in the previous section.

The Theories Among the authors who have suggested that pronouns behave like deWnite descriptions are Quine (1960), Parsons (1978), Cooper (1979), Evans (1980, 1982), Engdahl (1986), Kadmon (1987, 1990), Lappin (1989), Heim (1990a), Neale (1990), van der Does (1993), Lappin and Francez (1994), and Chierchia (1995). Here I will focus

Roberts, Pronouns as Deﬁnites

519

mainly on the representative work of Neale, Heim, and Chierchia, though much of what I have to say about their work applies to other recent theories.9 Evans (1980, 1982) gave the name ‘E-type pronouns’ to pronouns interpreted as deWnite descriptions, diVerentiating them from pronouns which are ‘merely coreferential’ with their referential antecedents, or are bound by quantiWcational operators. On the E-type interpretation, we retrieve the descriptive content of the Russellian logical form of the pronoun by copying the descriptive content of its noun-phrase (NP) antecedent, plus any predicates on the antecedent, into the logical form of a Russellian deWnite description. For example: (48)

Barney owns a donkey. He beats it.

Assuming a donkey is antecedent of it, the latter with an E-type interpretation, the two sentences in (1) have the logical forms: (480 )

9x[donkey(x) & owns(b, x)] 9x[donkey(x) & owns(b, x) & 8y[donkey(y) & owns(b, y) ! y ¼ x] & beats(b, x)]

The descriptive content of the antecedent plus the relation predicated of it in the Wrst line occurs twice (underlined) in the second line, copied over to serve as the descriptive content of the E-type pronoun it. For the truth of (48), (480 ) requires that Barney own a unique donkey, which he beats. The counterpart of uniqueness for plural NPs is maximality (Kadmon 1990), as illustrated in Evans’s (49): (49) John owns some sheep. Harry vaccinates them. (490 ) 9x[sheep(x) & owns(j, x)] 9x[sheep(x) & owns(j, x) & 8y[sheep(y)&owns(j, y) ! y 2 x] & vaccinates(h, x)] I assume that x in (490 ) can be a plural individual, the counterpart of a set of sheep in a lattice-based treatment of plurals (Link 1983; Roberts 1990). Then the second conjunct in the second line of (490 ) stipulates that any sheep owned by John is a member of this ‘set’ of sheep x. We then take vaccinates to be a lexically distributive predicate, so that the truth of ‘vaccinates(h,x)’ will require that Harry vaccinate all the individuals in x; hence, Harry vaccinates all the sheep owned by John. A similar proposal was made by Cooper (1979), who discussed Geach’s pronouns of laziness, by which Geach meant those which are replaceable by a mere repetition of their deWnite antecedents. Besides donkey pronouns like that in (48), a central motivation for Cooper was so-called paycheck sentences (Karttunen 1969), illustrated by (50)–(52): 9 See also Chierchia (1992), and the argument against that theory in Kanazawa (1994). And see Roberts (2003) for extended discussion of Kadmon (1987, 1990).

520

V. Anaphoric Pronouns and Dynamic Semantics (50) (51) (52)

The man who gave his paycheck to his wife was smarter than the onei who gave it to his mistress. (Karttunen 1969) Every woman who gave her paycheck to her broker was smarter than any womani who gave it to her lover. Every graduate student put his paycheck in the bank. Every professori lost it. (David Dowty, pers. comm.)

In each of these examples, it could be replaced by his/her paycheck, salva veritate. Cooper proposed that we interpret such pronouns as in (53), giving them the logical form of a generalized quantiWer: (53)

lP9x(8y(P( y) $ x ¼ y) & P(x)), where P is a property-denoting expression containing only free variables and parentheses.

One could take P in a particular example to be a relation variable R and its argument, a variable v, each then getting their interpretations via the assignment function in a standard (static) interpretation in Montague Grammar (Montague 1974b; Dowty et al. 1981): lP9x(8y(R(xi )( y) $ x ¼ y) & P(x)) For (48), R is assigned the relation between a donkey and its owner, and xi is assigned to Barney. For the underlined pronouns in (50)–(52), R is assigned the relation of being a paycheck owned by some individual, and xi is bound by the NP with the referential index i: lP9x(8y(paycheck(xi )( y) $ x ¼ y) & P(x)) So, for example, in (52) this will be the paycheck of some arbitrary professor, yielding the interpretation: (520 )

8xi (student(xi ) ! 9x(8y(paycheck(xi )(y) $ x ¼ y) & put-in-bank(xi , x))) 8xi (professor(xi ) ! 9x(8y(paycheck(xi )(y) $ x ¼ y) & lost(xi , x)))

Pronoun of laziness interpretations are not equivalent to Evans’s E-type interpretations, since in a theory like Cooper’s it is not required that part of the descriptive content of the disguised deWnite description include the relation predicated of the antecedent NP. Because of the predicate requirement on Evans’s theory, he cannot treat paycheck pronouns. For example, for (52), the descriptive content on Evans’s approach would have to be ‘xi ’s paycheck which xi put in the bank’, which would not yield the correct interpretation for it. Lappin (1989) and Neale (1990) (independently) propose an elaboration of Evans’s general account, allowing for ‘number neutral’ interpretations for E-type pronouns, as glossed informally for (1), repeated from above:

Roberts, Pronouns as Deﬁnites (1)

521

Every man that owns a donkey beats it. it ¼ ‘whatever donkey(s) the man owns’

This yields the interpretation captured by the logical form in (10 ): (10 )

8x[man(x) & 9y[donkey(x) & owns(x, y)] ! 9y[donkey=s(y) & owns(x, y) & 8z[donkey(z) & owns(x, z) ! z 2 y] & beats(x, y)], where donkey/s holds of y iV y consists of one or more donkeys, and beats is a distributive predicate, so that it is true that a plural individual is beaten iV all of its atomic elements (the members of the group) are beaten

This is truth-conditionally equivalent to the reading of Geach (1962), shown in (100 ): (100 )

8x8y[man(x) & donkey(y) & owns(x, y) ! beats(x, y)]

Hence a man beats all of his donkeys.10 Both Evans and Cooper assume that pronouns are ambiguous, with the E-type or pronoun-of-laziness interpretation only one possible interpretation. Neale (1990) is careful to spell out three interpretations which a pronoun can receive: as a referring expression, a bound variable, or a disguised deWnite (which he calls a D-type interpretation): (54)

Neale’s (1990: 166–8) interpretations of pronouns: (i) Pronouns are referring expressions when they are: a. non-anaphoric, i.e. fail to have an NP antecedent. These cases include demonstrative uses, as in (6) above (see his s. 3.3 for how he takes these to work); and examples where we refer to an entity which is discourse-new, but highly salient in the immediate context, as in (4) above. b. anaphoric on referring expressions. These include examples where the pronoun takes an NP like that man or Jones as antecedent. He is ‘neutral about how this happens’ (p. 168), but mentions the following possibilities:

10 Moreover, we obtain this result without encountering what is generally called the proportion problem (Kadmon 1987, 1990). Theories like those in Kamp (1981) and Heim (1982) give interpretations which are truth-conditionally equivalent to (10 ) and (100 ), by actually quantifying over pairs of a man and his donkey (unselective binding), as in (1000 ): 8 < x, y > [man(x)& donkey(y)& owns(x, y) ! beats(x, y)]. The problem is that in the general case the logical form in (1000 ) only yields the correct truth conditions if the quantiWer in question is universal. For example, consider a situation in which there are ten men, nine of whom own one donkey each, while the tenth owns ninety-one donkeys. Suppose that the rich man beats all of his ninety-one donkeys, while none of the others does. Now consider the truth of (i) in that situation: (i) Most men that own a donkey beat it. If we quantify over all the pairs of a man and one of his donkeys, requiring that most of these (say more than 50%) involve beating, we get the wrong truth-value. But since the logical form in (10 ) does not quantify over pairs, but only over men, with the universal force of it coming from the maximality of the disguised deWnite description, it does not encounter this problem.

522

V. Anaphoric Pronouns and Dynamic Semantics • (Geach’s) pronouns of laziness (going proxy for their antecedents), or • inheritors of their referents from their antecedents, or • getting their referent from the most salient entity in the context of utterance, which has been raised to salience by the utterance of the antecedent (ii) Pronouns are interpreted as bound variables when they are anaphoric on c-commanding quantiWers.11 (iii) Pronouns are interpreted as quantiWers when they are anaphoric on non-c-commanding quantiWers. In this case, they receive the Russellian, D-type interpretation.

Neale proposes the following account of D-type pronouns (1990: 182): (P5)

If x is a pronoun that is anaphoric on, but not c-commanded by, a quantiWer ‘[Dx: Fx]’ that occurs in an antecedent clause ‘[Dx: Fx](Gx)’, then x is interpreted as the most ‘‘impoverished’’deWnite description directly recoverable from the antecedent clause that denotes everything that is both F and G.

What (P5) means is that if the antecedent NP is itself a deWnite, and so in his theory denotes something which is unique (or, if plural, maximal), then the proxy interpretation of the pronoun need only contain that NP’s descriptive content, not that of the predicate as well. These are basically pronouns of laziness, in Geach’s sense. But if the antecedent NP is not itself deWnite, then, as in Evans’s account, the proxy deWnite contains the minimal predicate on the NP, as well as the NP’s own descriptive content. Hence, though he does not discuss them, Neale could account for the paycheck sentences in (50)–(52), on the assumption that the antecedents are deWnite: NPs with a deWnite possessive are generally considered to be themselves deWnite (C. Barker 1991). Consider (52): (52)

Every graduate studentk put his paycheck in the bank. Every professori lost it.

The borrowed descriptive content here (a pronoun of laziness interpretation, because the possessive his paycheck is deWnite) will be ‘xk ’s paycheck’. Unless k ¼ i, the referential index of every professor, the example will fail to have the correct interpretation. When they are not equivalent, the interpretation is nonsensical and arbitrary, because k is free in the second clause; when they are, then the desired reading results. Hence, there is no need for the theory to enforce the equivalence, nor is the resulting accidental binding a problem in natural language. 11 C-command is a syntactic relation, which requires that the c-commanding element and that which it ccommands occur in the same sentence. Moreover, in the syntactic tree for the sentence, the Wrst node dominating the c-commanding element must dominate the c-commanded element. There are many variations on this basic deWnition, but this will do for our purposes.

Roberts, Pronouns as Deﬁnites

523

The resulting theory oVers substantial improvement over Evans’s in several respects.12 Besides donkey sentences and paycheck sentences, Neale can predict the correct readings for crossing co-reference examples and ‘sage-plant’ sentences like (26) above. Whereas Evans had argued that the E-type pronouns were interpreted as ‘‘names whose references are Wxed by description’’, hence as rigid designators, Neale argues that they really ‘‘go proxy for deWnite descriptions’’, and hence he correctly predicts (1990: 186–9) that D-type pronouns can get de dicto readings under the scope of intensional operators, as well as intermediate scopes where multiple operators are involved. But he does not take a Wrm stand about what it means to ‘‘go proxy for’’ a deWnite description: My own inclination is to see (P5) as a very reliable descriptive generalization, and perhaps also as a processing heuristic, though, as we shall see, a rather impressive array of facts can be accounted for on the assumption that (P5) is a genuine linguistic rule that operates on syntactical representations in which scope assignments have been made. (Neale 1990: 184)

Heim (1990a) proposes a theory of personal pronoun interpretation which is quite close to Cooper’s, except that it is based on functions of variable arity, and hence does not need to posit pronoun ambiguity. She assumes a set of n-place functor variables f1n , f2n , etc., which: n • range over functions whose arguments are n-tuples in (A [ W [ T) , A the set of individuals in the model, W the set of worlds, T the set of times. For n ¼ 0, an nplace functor variable is just a plain individual variable. • yield values in A; hence, given the required argument(s), the pronoun always denotes an individual. • may be partial, that is, not deWned for some arguments.

A pronoun is represented at LF as a term of the form in (55): (55)

fi n (v1 , . . . , vn ) where f is a free function variable of arity n, n 0, and v1 , . . . , vn are variables of the appropriate type (individual, world, or time).

Among the types of examples this theory can treat are donkey sentences like (1) and paycheck sentences like those in (50)–(52). With a donkey in (1) interpreted as ‘exactly one donkey’, the pronoun is translated as f 1 (xk ), where the free variable f is interpreted via the function assigning values to variables as the function described in (56); there, the top line describes domain and co-domain, the bottom gives f(x) for arbitrary x. This is the function assigning to each man the unique donkey that he owns. It yields the number-neutral interpretation quite straightforwardly if we take the function to pick out the supremum of the set of donkeys instead of a singular donkey: Where there 12 Actually, the central feature of Neale (1990) is an extended, very convincing argument against viewing the distinction between referential and attributive readings of deWnites as following from ambiguity in the deWnite article, arguing instead for a view of referential readings closer to that of Kripke (1977).

524

V. Anaphoric Pronouns and Dynamic Semantics

is only one donkey owned by a given man, that donkey will itself be the supremum of the set; otherwise, the function will pick out the individual consisting of all the donkeys owned by the man in question. (56)

bf : {x: x is a man and owns exactly one donkey} ! A x j! the unique donkey that x owns

In a paycheck sentence, the pronoun is translated as fk1 k(xi ), interpreted as ‘paycheck of xi ’, yielding the logical form we saw in (52’). In (56), the function bf is not deWned over the entire domain A. The listener has no reason to assume that the existence of a unique association between individuals and the donkeys they own extends beyond a proper subset of the donkey-owning men. Since the function bf is only partial, it follows that the formula x beats f(x) does not have a well-deWned truthvalue for every assignment g that has g( f ) ¼ bf . If g(x) is outside the domain of g( f ), kx beats f (x)kg is undeWned. (Heim 1990a: 141)

Though superWcially the variable interpretation does not appear to equal a disguised deWnite description, in fact both existence and uniqueness here are presupposed, not (just) entailed. Presupposed is whatever it takes for the function bf to be well-deWned, at least over the domain that is relevant to determine a truth-value for the utterance as a whole; which is to say, since bf is a function, that there exists exactly one value for each element in the restricted domain. Chierchia (1995) develops a theory in which pronouns have two types of interpretations, dynamically bound versus E-type. In general, pronouns are translated as variables over n-place functions from entities to entities in the manner of Heim (1990a), with n ¼ 0 yielding the standard case where pronouns are ‘‘entity-level variables’’ (Chierchia 1995: 114). Dynamic binding is a variant of the binding of pronouns across discourse by dynamic operators Wrst developed as an alternative to Heim’s and Kamp’s theories by Groenendijk and Stokhof (1990). In these theories, existential operators (e.g. those introduced by indeWnite NPs) are dynamic in that they can take scope outside the sentences in which they are introduced, though their scopes are limited by any nonexistential operators with wider scope. The special variables which may be bound by such dynamic operators are called discourse markers, to distinguish them from ordinary variables bound by non-dynamic operators. In Chierchia’s theory, binding via these dynamic operators is only possible when the pronoun’s arity is 0. Chierchia’s E-type interpretation of pronouns arises when pronouns are, instead, treated as variables over n-place functions, essentially as in Heim’s theory just discussed except that the n arguments of the function are given syntactically rather than directly as part of the pronoun’s interpretation. The function’s characterization is retrieved pragmatically from the context, along with the binding of the values of the n variables which are its arguments. As in Heim’s theory, since the variable ranges over functions, the resulting interpretation of the pronoun presupposes semantic uniqueness.

Roberts, Pronouns as Deﬁnites

525

Chierchia’s (1995: ch. 4) general interpretation of deWnite descriptions involves the binding of regular variables, rather than discourse markers, and yields a Russellian interpretation via a function whose value is partly given by the descriptive content of the deWnite NP, partly retrieved from context (essentially as in Clark’s (1975) Bridging; see also Roberts (2003) for extended discussion). Chierchia shows there that the E-type interpretation of pronouns is a subcase of this functional interpretation of deWnites. Following Neale’s interpretations of his D-type pronouns, Chierchia claims that when pronouns receive an E-type interpretation, they are unmarked as to whether they are singular or plural. Hence, Chierchia’s E-type interpretation of donkey pronouns yields an interpretation which is truth-conditionally equivalent to Geach’s interpretation of donkey pronouns in (1’’), which Chierchia calls the universal interpretation; while the dynamically bound interpretation leads to the interpretation we Wnd in Pelletier and Schubert’s (1989) example (57), which Chierchia calls the existential interpretation: (57)

Every man who had a quarter in his pocket put it in the meter.

On the interpretation of interest, a given man need put only one quarter in the meter, even if he has several in his pocket; hence, maximality is undesirable. Chierchia (1995: 113) claims that his theory, though it yields two kinds of interpretations of pronouns, does so without requiring that pronouns themselves be ambiguous: the situation is just as with variables in logic, which can have free and bound uses without being ambiguous . . . Their content is provided in essentially two ways: via semantic binding . . . or via contextually available information . . . In a way, the present work as a whole is simply an attempt to spell out a bit further the traditional claim that pronouns can be bound or free, by arguing on the one hand that binding is not only static (under c-command), but also dynamic and by arguing on the other hand that interpreting free pronouns through the context has its own systematicity (embodied in the E-type strategy).

However, the claim that pronouns are unambiguous in this theory is rather misleading. There are two distinct types of variables in Chierchia’s formal theory—discourse markers and other variables. The discourse markers are all of type e, ranging over individuals in the model; and they are given values by a special type of assignment function, associated with dynamic operators, which maps them onto such individuals. Other variables may be of any type (among other things, allowing the n-place functional interpretation of pronouns), and they are interpreted by the usual Tarskian assignment functions, mapping non-discourse marker variables onto entities of the appropriate type in the model. If a pronoun is to be dynamically bound, it must be translated as a discourse marker, of type e, and receives its interpretation via an assignment v of values to discourse markers; there is no uniqueness eVect associated with this interpretation. But if the pronoun is to receive an E-type interpretation, it is translated as a regular variable, of the type of some n-place function, and receives its interpret-

526

V. Anaphoric Pronouns and Dynamic Semantics

ation via a diVerent sort of assignment function g, leading to a uniqueness eVect. Though both sorts of pronominal interpretation are technically variables, this is still semantic ambiguity. It is crucial that one type of pronoun is a discourse marker, the other a variable over functions. The fact that discourse markers of type e can also technically be characterized as 0-place functions should not obscure their essential diVerences from ordinary variables within the interpretive mechanism Chierchia provides. In all of the theories in question, there is an interpretation of pronouns in which they are like Russellian deWnite descriptions in carrying a uniqueness requirement, whether proVered or presupposed. In all of these theories except that of Heim (1990a), pronouns are treated as ambiguous, with the Russellian interpretation coexisting with a bound variable interpretation and/or a referential interpretation of some kind. I have already brieXy sketched the arguments from Roberts (2003) that the semantic uniqueness requirement is not desirable for deWnites in general, and especially for pronouns, which do not generally give rise to uniqueness eVects. And I have argued that pronouns carry an additional presupposition of salience, which none of these authors addresses. In the following section, I will present a number of other problems with these theories.

Problems with Theories of Pronouns as Disguised DeWnite Descriptions If pronouns are taken to have exactly the same semantic content as deWnite descriptions, a number of problems arise. Two could be ameliorated fairly straightforwardly. To any of the theories discussed in the preceding section, one could add the presupposition on pronouns that there is a corresponding discourse referent in the common ground which is maximally salient. Since discourse referents are by deWnition familiar, this takes care of both the requirement of weak familiarity and that of maximal salience. The resulting theory is essentially that of Kadmon (1990). Other problems, however, are not so easily addressed.13 Semantic uniqueness I have already brieXy sketched the argument that semantic uniqueness is too strong, for deWnites in general and pronouns in particular. As Heim puts it, uniqueness is the Achilles heel of an E-type analysis. The following example, from Heim (1990a), serves to reinforce the point: (58)

If a man is in Athens, he is not in Rhodes.

Since there is always more than one man in Athens, on an E-type account of the pronoun the descriptive content one can retrieve from the antecedent of the condi13 See S. Barker (1997) for additional arguments against Neale’s theory, some more valid than others, and none of which is addressed in what follows.

Roberts, Pronouns as Deﬁnites

527

tional will always lead to a false interpretation. Yet (58) is obviously true and unobjectionable. Heim’s response to this problem is to posit that conditionals are (roughly here, for brevity) generalizations over minimal situations—in this case, situations which contain at most a single man and the city of Athens. In each such situation there is, of course, at most one man; relativizing all arguments to that situation, then, he is interpreted as ‘the man who is in Athens in minimal situation s’ and hence, the problem of semantic uniqueness is avoided. However, minimal situations are notoriously diYcult to deWne adequately, as Heim herself admits; and the solution appears to be rather ad hoc to the problem. Moreover, this solution leads to the re-creation of the proportion problem in conditional versions of the donkey sentences, as in (59): (59)

In most cases, if a man owns a donkey he beats it.

Here, in the situation with ten men, one of them owning ninety-one donkeys and beating them all, if we range over minimal situations containing a man and one of his donkeys, we have one hundred pairs, ninety-one of them (all with the same man) involving beating. But most speakers agree that (59) should not be true when only one donkey owner in ten veriWes it, since it seems to be a generalization about what it is to own a donkey, and one bad owner (even if he is rich) does not suYce to make it. The point is not that there might be conditionals where this type of reading arises (like if a man meets another man, he greets him), but that one should not restrict the interpretation of conditionals in general to range only over minimal situations. See Kadmon (1990) for extended discussion of such conditionals. In a theory like the one outlined in the previous section, (58) presents no problems. There is at most one salient discourse referent (strongly) familiar in the context of interpretation of the consequent; that discourse referent is appropriately referred to using a third-person, singular, masculine pronoun (so that the pronoun’s descriptive content is satisWed); and anaphora goes through without a hitch. Suppose one added familiarity and salience presuppositions to an E-type account, with the semantic uniqueness presupposed instead of proVered; the resulting theory would be much like that of Kadmon (1987, 1990). Then one might appeal to these new presuppositions to implicitly restrict the domain in which uniqueness must hold, in much the same way as I have done for informational uniqueness above. However, this would still leave one with a Russellian account of deWnite descriptions, which I have argued is too strong for those as well. The scope issue All of the theories under consideration encounter a prima-facie problem: They predict that the descriptive content of a pronoun which is a disguised deWnite description is somehow borrowed from previous context, often being the descriptive content of a previous quantiWcational NP plus or minus any predicates on the latter. Except in Chierchia’s (1995) theory, there are no constraints on the relationship between the antecedent NP and the pronoun. In particular, this seems to predict that a pronoun

528

V. Anaphoric Pronouns and Dynamic Semantics

could take its content from an NP which is under the scope of an operator that does not have scope over the pronoun. The following, from Heim (1990a), illustrate the problem: (60) (61)

John owns no sheep1 and #Harry vaccinates them1. John doesn’t own a car1 , and #he drives it1 on Sunday.

She argues that the infelicity of the second clause in these sentences derives from a presupposition failure. The deWnite has an existence presupposition, which fails in these cases due to the fact that the Wrst conjunct denies the existence of the relevant entities. Hence, repair by accommodation is not possible, either, as it would lead to contradiction. Neale has much the same account of examples involving double negation, claiming that the pronoun ‘just makes no practical sense to use in these sentences’ because the resulting interpretation is contradictory (1990: 232). This may suYce for examples involving negation, but things are more complex than this. Consider, for example, the following example involving deontic modality: (62)

You should meet a nice man and get married. He has already seen you in his dreams.

If the second sentence in (62) is felicitous at all, he means ‘the nice man you’ll meet someday and marry, whoever he is’. This presupposes that the addressee will actually meet someone, a presupposition which must be accommodated here. Heim’s theory could handle such an example, taking the value of the variable over functions denoted by he to be the one-place function from women to the men they’ll meet and marry, applied to the addressee; this value is, presumably, pragmatically retrieved.14 But the application of Neale’s (P5) would predict something else: that he can mean ‘the nice man who you meet’, with no hint of futurity or marriage. There is no contradiction here, so no ready way to block such an interpretation. The attested interpretation, involving accommodation, can only be retrieved by considering the nature of the operator under which the licensing NP a nice man occurs, realizing that it is futureoriented, and understanding, moreover, that the conjunction with and get married implicates marrying this nice man. Examples involving modal and temporal subordination, and other cases of anaphora in intentional contexts (Roberts 1989, 1990, 1996a) show that generally the explicit descriptive content of an indeWnite in such contexts can only be borrowed when the irrealis mood is maintained. We saw this in (21)–(23), repeated here, where it can borrow the content ‘earthquake that hit San Francisco’ only under the scope of another irrealis operator in (21b); in (23b) the content must be a nominalized version of the possibility reported in (23a), that is, including the irrealis mood itself. 14 At the end of her paper, Heim oVers a diVerent version of the theory where, as in Evans (1977, 1980), the descriptive content of the pronoun is borrowed from the syntactically given descriptive content of its antecedent NP. With respect to this type of example, the second version of Heim’s theory would run into the same problems as Neale’s.

Roberts, Pronouns as Deﬁnites (21) (22) (23)

529

a. An earthquake might hit San Francisco. b. That would upset me. It would be frightening. a. An earthquake hit San Francisco in 1989. b. That upset me. It was frightening. a. An earthquake might hit San Francisco. b. That upsets me. It is frightening.

And consider (63): (63)

Most days, John brings a bagel for breakfast. a. Sometimes, Mary steals it from him. it: ‘the bagel John has brought for breakfast that day’ b. Mary stole it from him.

Here, it, with the interpretation ‘the bagel John brought for breakfast’, is acceptable in (a), where the mood of generalization introduced by the adverbial most days is continued with sometimes. This interpretation is not possible in (63b), where the mood of generalization is not continued. Yet all of the other theories considered require only that the function from John to his bagel on a particular day be salient, along with the day of utterance made salient by the present tense, and hence fail to predict the awkwardness of (63b). Of course, we can ameliorate (63b) by adding Today at the beginning, but then (on the theory proposed here) the pronoun is licensed by default entailment from the assumption that today was an ordinary day with respect to John’s breakfast. Heim (1990a) points out that an algorithmic approach like that in Neale’s (P5) to determining the descriptive content of these pronouns on the basis of the logical form of the antecedents leads to problems with examples like her (64) and (65): (64) (65)

Every student turned in a paper. They were all identical. Each time every student turned in a paper they were all identical.

The problem is that since a paper is under the scope of every student, this would lead to the prediction that they means ‘the papers that x turned in’, rather than the understood ‘the papers that the students turned in’. Even if we attempt to deal with (64) by claiming that they was somehow deictic, this will not help with (65), where no single incident or set of students or their papers was involved. The problem that all these examples reXect is that the theories in question depend on mere salience for retrieval of the descriptive content of a pronoun that is a disguised deWnite description. Note the following diVerence between the distribution of felicitous uses of the pronoun it and that of the anaphoric indeWnite one: (66) (67)

Either John has a new car, or else #Mary has it. Either John has a new car, or else Mary has one.

(67) argues that the descriptive content of a new car, though it is not accessible in the second disjunct, is suYciently salient to license anaphora. However, outside the Wrst

530

V. Anaphoric Pronouns and Dynamic Semantics

disjunct, there is no weakly familiar discourse referent for a new car, that is, no existential entailment, and hence it is infelicitous in the absence of any other plausible discourse referent antecedent.15 Salience of the descriptive content of a recently uttered NP is simply not suYcient to license the corresponding interpretation of a pronoun. Chierchia avoids several of these problems when the E-type pronoun has an overt NP antecedent by requiring that the pronoun’s logical form contain a variable which must be anaphorically co-indexed with the head of the antecedent, thereby determining ‘the range of the function’ (1995: 231), though he does not specify how this would be done. If there are operators with scope over the antecedent that do not have scope over the pronoun, this would yield infelicity. However, he does not discuss how he would handle examples of modal subordination and other cases of anaphora in intensional contexts, as in (9) and (63a), where there is a lack of accessibility to the licensing NP. And he does not discuss examples involving deixis or contextual salience, where there is no anaphoric antecedent NP. Moreover, he gives no motivation for the required anaphoric co-indexing with an accessible NP head, which hence appears ad hoc. In contrast, the scope constraint falls out of the theory encoded in (40) directly, as a consequence of the way that discourse referents are managed in discourse. Thus, the theories in question all fail to capture the fact that E-type readings seem to arise just when existential entailments or implicatures license weak familiarity (in the local context of interpretation) of the corresponding discourse referent. Using a dynamic interpretation to keep tabs on the familiar discourse referents at any given point in discourse permits us to capture this fact, giving improved empirical coverage across a wider range of types of examples. Of course, one might enrich an E-type account with the requirement that there be a local entailment of the sort required, for example, that the use of the NP from which the descriptive content of the pronoun is borrowed contextually entails the existence of the relevant discourse referent. But then it seems to me that the resulting theory is very close to what I proposed above, modulo semantic uniqueness, familiarity, and salience. Pronominal ambiguity, donkey sentence readings, and generic interpretations Any adequate theory of pronominal interpretation would have to involve an account of variants of Geach’s donkey sentences which have non-universal interpretations of the E-type pronoun. Examples like these were Wrst pointed out by Pelletier and Schubert (1989), and have since been discussed extensively in the literature; see C. Barker (1993), Kang (1994), Kanazawa (1994, 2001), Chierchia (1995), Yoon (1996), and Geurts (2002), among others. The existence of the two types of readings for donkey 15 I assume, following Chierchia (1995), that the logical form of a disjunction is roughly of the form [(:P) ! Q ], as a way of capturing both the inaccessibility of indeWnites inside the two disjuncts and the fact that this inaccessibility changes if the Wrst disjunct is negated, as in Partee’s Either there’s no bathroom in this house, or it’s in a funny place. In terms of the present theory, the negation of the Wrst disjunct entails the existence of a bathroom in the house in question, which is hence temporarily accessible from the second disjunct, though not beyond.

Roberts, Pronouns as Deﬁnites

531

pronouns is problematic for most theories of pronoun interpretation. For example, Neale (1990) predicts two readings, as shown, neither of which is the preferred reading: (57)

Every man who had a quarter in his pocket put it in the meter.

Neale’s readings: Singular: Every man who had a quarter in his pocket put the (unique) quarter he had in his pocket in the meter. Number neutral: Every man who had a quarter in his pocket put all the quarters in his pocket (whatever quarters he had) in the meter. Preferred reading: Every man who had a quarter in his pocket put one of the quarters in his pocket in the meter. Chierchia (1995) argues that the existential interpretation of pronouns in such examples is licensed by the fact that natural language quantiWers are conservative, in the sense of Barwise and Cooper (1981). In their relational terms, where a quantiWcational operator is a relation between two sets: Conservativity: if A,B E, then DE AB $ DE A(A \ B) That is, if a quantiWcational operator DE (the interpretation of a determiner) is conservative, taking as its two arguments A (the denotation of the common noun) and B (the interpretation of the predicate), with both of these taken to be sets (extensional interpretation) yields the same interpretation as taking as arguments A and the intersection of A and B. Equivalently, taking a quantiWcational determiner as a restricted generalized quantiWer, restricted by its common noun-phrase, this encodes the same property as Benthem’s (1987) lives-on: Lives-on: X 2 kDk(A) iff (X \ A) 2 kDk(A) Generalized quantiWers (in the extensional sense) are sets of predicates. If some predicate X is in the denotation of D with argument A, then the intersection of X with A is also in that denotation. The test for conservativity, which all English quantiWcational determiners pass, is: Test for conservativity: Det CN VP $ Det CN is a CN who VP Many men run $ Many men are men who run Every woman smiled $ Every woman is a woman who smiled Hence, when considering the properties that every man who has a quarter in his pocket has, we can be assured that one of them is that of being a man who has a quarter in his pocket. In terms of the theory proposed above, the quarter is weakly familiar under the scope of the quantiWcational determiner every because its existence is entailed by the

532

V. Anaphoric Pronouns and Dynamic Semantics

conservativity of that determiner. But this theory does not require semantic uniqueness, so the interpretation we get is closer to that suggested by the logical form in (57’): (57’)

8x[man(x) & 9y[quarter( y) & has_in_pocket (x, y)] ! 9y[quarter( y) & has_in_pocket (x, y)] & puts_in_meter (x, y)]

Of course, this is not the actual logical form of (57) on the theory proposed here, because the existence of the quarter is presupposed, not proVered. And the pronominal variable is not bound by a second existential quantiWcation. Rather, the discourse context at the time of interpretation of the VP will be such that (temporarily, while under the scope of the universal quantiWer) all the assignments of values to variables which capture the information the interlocutors share will assign to y a quarter in the pocket of x. Krifka (1998) oVers quite a diVerent account of the basic diVerence between universal and existential interpretations of quantiWcational donkey sentences, based on an observed diVerence between upward- and downward-entailing quantiWcational operators which he calls Rooth’s generalization:16 Basically, if the head quantiWcational determiner is monotone decreasing (e.g. no, few, at most n), we expect the existential interpretation, while if it is monotone increasing (e.g. all, any), we get the universal. But Kanazawa (2001) argues convincingly that this generalization is not suYciently nuanced to reXect all the facts about diVerent types of monotonicity in quantiWers; based both on judgments of her own informants and on work by Jackson (1994), it seems that monotone increasing quantiWcational determiners with existential interpretations (e.g. some, several, at least n, many) tend more often to lead to existential interpretations than every or any, the diVerence in generalized quantiWers theory terms being that the class with existential entailments is also persistent (or left-monotone increasing), while those without are anti-persistent (or left-monotone decreasing). Moreover, she argues that when Rooth’s generalization holds true, it seems to hold only for plural donkey pronouns, not for singular. We see instances of this in examples already cited above, where a pair of donkey sentences with singular donkey pronouns and the same quantiWcational head every display the two diVerent readings: (57) has a preferred existential reading, (1) a universal: (1) Every man that owns a donkey beats it. (57) Every man who had a quarter in his pocket put it in the meter. What is the diVerence between examples with preferred universal interpretations, like (1) and (68), on the one hand, and those with preferred existential interpretations, like (57) and the Wnal sentence in (69), on the other? (68)

If a farmer owns a donkey, he beats it.

16 After Rooth (1987), though Kanazawa (2001) points out that only part of this is properly attributed to Rooth. For deWnitions and discussion of monotonicity and persistence in generalized quantiWer theory, see Barwise and Cooper (1981), and the very helpful Gamut (1991).

Roberts, Pronouns as Deﬁnites (69)

533

The farmers in this area are friendly and generous, even though they’re quite poor. Suppose your car breaks down in the neighborhood, and you need to go for help. If a local farmer owns a donkey, he’ll loan it to you to ride back into town.

The diVerence, I contend, lies in what we know about the types of situations being described, and about the basis on which gnomic generalizations about such situations can be made. We know that one quarter is a standard parking meter fee, so we do not expect a person who has more than that to put them all in the meter. The generalization in (57) is not about men or quarters, but about parking meters and their limited requirements, hence about suYciency. Similarly, (69) is about the generosity of the farmer in helping the stranded motorist back to town; but for the purpose, the loan of at most one donkey would suYce. On the other hand, universal maltreatment of animals could have to do with the nature of the owners, the nature of the animals (how harshly one would have to behave to control them, for example), or both. C. Barker (1993) has a useful discussion of other factors that may bear on how we know which reading of a donkey sentence is preferred in a particular context. Whatever the basis of generalization in (1)/(68), it would suggest that if men and donkeys were of such a character to lead to beatings, non-beating would be the exception, not the rule. Hence, it seems that the universal cases could be viewed as pragmatically strengthened versions of the weaker existential. There is at least one other reading for unbound pronouns which seem to have a quantiWcational antecedent in prior discourse, besides the ones discussed above. Chao (1983) noted that generic interpretations of plural pronouns do not require bare plural antecedents, as illustrated in her (70): (70)

Many women from the village came to the fair. They like that sort of thing.

Here, we understand that it is women from the village in general who like fairs, and not just those who actually came to the particular fair in question; hence, the subject of the Wrst sentence is clearly only a licensing NP, not an anaphoric antecedent in the classic sense. Chao conjectured that such readings are available whenever we use a plural common noun, like women in (70). However, in other examples a singular NP, like Sam’s dog in (71), will serve to license these readings, as well: (71)

Sam’s dog got into my compost pile last night. They love rotten stuV. [‘dogs love rotten stuV ’]

Nor does the NP which appears to trigger the salience of the natural kind have to entail existence of stages of that kind, as we see in (72), due to Webber (1979), where few is generally taken to be a downward entailing operator without existential entailments: (72)

Few linguists smoke. They know it causes cancer. [‘linguists know that smoking causes cancer’]

534

V. Anaphoric Pronouns and Dynamic Semantics

(I note that the pronoun it in (72) illustrates yet another respect in which pronominal anaphora depends more on pragmatic factors like salience and plausibility than on explicit mention (strong familiarity). Here, it arguably refers to a nominalization of the preceding predicate, smoke.) Apparently, mention of an instance of a kind or quantiWcation over stages of a kind are generally suYcient to make the kind itself salient. The kind is already familiar in the common ground due to interlocutors’ general world knowledge that such kinds exist. And kinds are, by nature of what they are, semantically unique, and hence informationally unique to interlocutors who know what a kind is. In these examples, it is clear that it is a kind that is referred to by the pronoun from the predicate in question (both its content and its tense and aspect) and the context in which it occurs. In (71), there is no salient group of individuals, and the property of loving rotten stuV is plausibly predicated of dogkind, so we take it that way as an explanation of Sam’s dog’s behavior. This is all pragmatics. We Wnd this generic reading of pronouns in some donkey sentences, as well: (73)

Every farmer that owns donkeys breeds them.

(73) as generally understood does not require that any given farmer breed all of his donkeys—some are surely too young or too old or otherwise unsuitable—but only some of them; though generally he would breed more than one, in fact as many as would be proWtable to him. The simplest, most plausible way to interpret (73) is to take them to have a generic interpretation: ‘every farmer that owns donkeys breeds donkeys’. How do these generic interpretations arise from a technical point of view? Carlson (1977a) oVers an extended argument that bare plural NPs have a generic interpretation, denoting the relevant natural kind. When such NPs appear to have an existential interpretation, this is because they occur as arguments of a verb or complex predicate that he called stage-level. For example, in Firemen are available, the stage-level predicate be available takes the generic-denoting Wremen and says of it that there are stages of that kind—roughly, spatio-temporal instances, individual people— that are available. He notes that plural pronouns can be anaphoric to generically interpreted bare plural NPs, and that one of these pronouns can appear to have either a generic or a stage-level interpretation, depending on the predicate that takes it as argument, whether that predicate is stage-level or kind-level, and its tense and aspect. In (73), though breeds is a stage-level predicate, it appears here in the bare present tense used in non-stative verbs to indicate the habitual or generic interpretation, as opposed to pertaining to particular situations. Carlson also shows that kind interpretations of bare plurals are not equivalent to those of universally quantiWed NPs: we would expect exceptions, sometimes systematic ones, as with the systematic exception of the young and old or inWrm from breeding. Carlson’s account of the generic interpretation of bare plurals makes one more prediction. This is that when such plurals, denoting kinds, occur with a stage-level predicate and without the bare present tense driving a habitual interpretation, the

Roberts, Pronouns as Deﬁnites

535

stage-level predicate takes the generic as argument and ‘lowers’ it, entailing that the predicate holds of stages of the kind. This interpretation is equivalent to an interpretation of the plural as an existentially quantiWed NP, an indeWnite. If plural pronouns have such an interpretation, they should be susceptible of this lowering, and (74) argues that they are: (74)

A: Few teenagers are responsible. B: I know what you mean. They tore up my garden at Halloween!

They in (74) seems to mean ‘some teenagers’. We can derive this in Carlson’s theory by taking the pronoun to denote the familiar, salient, and informationally unique natural kind ‘teenagers’. This serves as argument to a stage-level predicate, ‘the property of being a kind x such that there exists a (possibly plural) stage of x which tore up my garden at Halloween’, combining to give the reading ‘there is a (plural) stage of teenager-kind which tore up my garden at Halloween’. Though apparently indeWnite, the interpretation of the pronoun itself is deWnite, and the indeWniteness arises from the meaning of the stage-level predicate. It seems that we can Wnd this type of reading in donkey pronouns, as well: (75)

If one of my neighbors likes dogs, he generally seems to breed them, and they end up running around in packs, terrorizing the neighborhood cats, and getting into the garbage.

In (75), the Wrst donkey pronoun, them, is directly anaphoric to the kind-denoting bare plural dogs, and hence has a generic interpretation itself. But the second pronoun, they, though interpreted in the same fashion, due to the lowering triggered by the conjoined stage-level predicates seems to denote a non-kind existential interpretation, indeWnite: ‘dogs that have been bred by the neighbor in question’. What the generic examples show us is that the theories which treat pronouns as deWnite descriptions have overlooked a whole class of examples where a pronoun is licensed by quantiWcational NP in prior discourse, but where the pronoun apparently has either a generic or an indeWnite, rather than a deWnite, interpretation. Note also that in most of these examples, there is no accessible NP antecedent; rather, the licensing NP makes the kind salient, it is already weakly familiar in the common ground, and this suYces to make the pronoun felicitous. The existence of donkey pronouns with existential and generic interpretations argues that neither E-type nor D-type pronouns alone, nor Heim’s functional variables, would suYce to cover the range of interpretations we Wnd in pronouns. Any theory which includes an E-type interpretation of pronouns, with the associated uniqueness/maximality, must treat pronouns as ambiguous, as in Neale’s and Chierchia’s theories. Chierchia (1995: ch. 2) oVers additional arguments that ambiguity is required if we take one reading of pronouns to be that of deWnite descriptions. But there is a marked lack of attested ambiguity in particular examples, including those involving donkey pronouns. Rather, it seems that in general either the bound

536

V. Anaphoric Pronouns and Dynamic Semantics

variable or the E-type reading or the generic reading is plausible, but not more than one. For example (57) has only the existential interpretation, not requiring of the men in question that they empty their pockets of quarters; while we seem to prefer the universal interpretation of (76), where we might expect for any given dog-owner that all his dogs will be vaccinated: (57) (76)

Every man who had a quarter in his pocket put it in the meter. Every Lancaster resident who has a dog had it vaccinated for rabies last year.

This lack of ambiguity in particular examples suggests that what are really at issue in deriving the reading for a pronoun in a given utterance are contextual, pragmatic factors, triggered by the implications of the content of the sentence, as discussed above. For with lexical ambiguity generally, we would expect that there should be single examples which display the ambiguity truth-conditionally. And where lexical ambiguity is in question, I see no principled way to constrain the ambiguity so that certain readings only arise in certain contexts. This suggests that it is not that pronouns themselves are ambiguous. Rather, there are several diVerent ways of using a combination of syntactic, semantic (scope) and pragmatic factors to Wx the intended discourse referent antecedent for a pronoun. In turn, this suggests that Neale’s (P5) and its counterparts for retrieving the descriptive content of pronouns in other theories are at best ‘processing heuristics’, as Neale puts it. Number neutral pronouns A number of recent authors have claimed that pronouns can have number neutral readings, where their semantic number need not correlate with their overt, syntactically marked number. These are generally theories which give pronouns the semantic content of a disguised deWnite description, and number neutrality is appealed to to explain readings like the universal reading of donkey sentences. In addition to the theory in Neale (1990) discussed above, this view has been promoted by Davies (1981), Lappin (1989), Lappin and Francez (1994), Kang (1994), Yoon (1996), Krifka (1998), and Larson and Segal (1995). A recent article by Kanazawa gives both a good overview of what she calls the Number Neutrality Thesis and a convincing and thorough argument that this thesis is ‘‘not well-motivated, makes wrong predictions, and does not do the job it is intended to do in some cases’’ (2001: 383). Kanazawa points out, contra the claim that syntactic number is merely a marker of agreement with an antecedent, that syntactic agreement is not necessary, or even always preferred. (77b) is at least as good as (77a), and (78b) is preferable to (78a) for most speakers: (77)

a. Every farmer who owns at least one donkey beats it. b. Every farmer who owns at least one donkey beats them.

Roberts, Pronouns as Deﬁnites (78)

537

a. #Every farmer who owns more than one donkey beats it. b. Every farmer who owns more than one donkey beats them.

Moreover, as discussed and illustrated in the previous section, what Krifka (1998) calls ‘Rooth’s generalization’ regarding the distribution of universal versus existential readings of donkey sentences holds, when it holds at all, only of plural donkey pronouns, not of singular ones. Neale (1990) had pointed out that treating donkey pronouns as numberless, interpreted as denoting the maximal sum of the individuals satisfying the relevant description, permitted him to account for the ‘sage-plant’ sentences of Heim, like (26) above. However, in order to obtain the attested interpretation, it was necessary to stipulate that the predicate over the pronoun be distributive. If not, his (79) would mean (790 ), which is contradictory. But with distributivity, we get the attested meaning in (7900 ): Every man who bought a beer bought Wve others along with it. For every man x who bought a beer, and for the maximal amount of beer y that x bought, x bought Wve other beers (non-identical with y) along with y. (7900 ) For every man x who bought a beer, and for each beer y that x bought, x bought Wve other beers (non-identical with y) along with y.

(79) (790 )

Not only is the stipulation that the predication be distributive ad hoc, but Lappin and Francez (1994) point out that it fails to predict the correct interpretation for other examples where the donkey pronoun is plural. (80) is understood to entail (800 ), but Neale would predict that them denotes the sum of all beers bought by a man and is interpreted distributively, yielding the incorrect interpretation in (8000 ): (80) (800 ) (8000 )

Every man who bought two beers bought four others along with them. Every man who bought at least two beers bought at least six beers. Every man who bought at least two beers bought at least Wve beers.

Moreover, Kanazawa shows that singular donkey pronouns lack the collective and cumulative interpretations and the narrow scope readings that one would expect if they have a semantically plural interpretation. Hence, the unacceptability of her (81b), compared to (81a)—despite the fact that in (81a) there may be only one donkey!— and the lack of cumulative reading in her (82a), as opposed to (82b): (81)

(82)

a.

Every farmer who owns one or more donkeys rounds them up at night. b. #Every farmer who owns more than one donkey rounds it up at night. a. Every thief who stole a painting from the museum earned at least one million dollars by selling it. b. Every thief who stole any paintings from the museum earned at least one million dollars by selling them.

538

V. Anaphoric Pronouns and Dynamic Semantics

I know of two more types of cases that bear on this question, where minimal pairs can be adduced to show that pronominal number is semantically signiWcant. First, number in pronouns is semantically important in giving rise to the generic interpretation discussed in the previous section. For example, the most available interpretation of the plural pronoun in (83), the generic interpretation, is not available for the singular pronoun in the otherwise identical (84): (83) (84)

Every farmer that owns donkeys breeds them. Every farmer that owns a donkey breeds it.

Instead, the preferred interpretation of (84) is the gnomic universal interpretation found in (1), where every donkey must be bred. To see the diVerence, consider that (83) is not necessarily about the relationship of any particular farmer with his current donkey possessions, but is about his tendency to breed donkeys when appropriate. But (84) says that donkey-owning farmers run something more like the donkey counterpart of a puppy-mill. Second, in arguing that pronouns are number neutral, Neale claims that universally quantiWed NPs can be antecedents to both singular and plural pronouns, indiscriminately. For example, he claims that (85) and (86) are semantically equivalent, and oVers (87) and (88) as additional examples where singular pronouns may apparently refer back to universally quantiWed NPs: (85) (86) (87) (88)

Every new recruit is armed. He is ready for combat at a moment’s notice. All new recruits are armed. They are ready for combat at a moment’s notice. Every Swiss male over the age of 21 owns a gun. He is required to do so by law. Each candidate will be debriefed by Mrs Hendrix. He will be given some advice on how to tackle the press.

He does not seem to recognize that (85), (87), and (88) are examples of what Roberts (1989) dubbed telescoping, wherein a universal generalization is followed by an apparent universal instantiation. As discussed there and in Roberts (1996a) and Poesio and Zucchi (1992), occurrence of these types of examples is tightly constrained. One of the main constraints is that all information predicated of a singular pronoun in such examples must be entirely generic, not speciWc to a particular instance or even pertaining to a particular event. This plays out in the diVerence between (89a) and (89b), on the one hand, where the plural pronoun is acceptable whether the subject of the Wrst sentence is plural or singular, and (89c), on the other, where the same Wrst sentence as in (89b) does not license use of the singular pronoun, contra the predictions of the number neutral theory: (89)

a.

New recruits are armed. Last night, the president questioned whether they should be trusted so soon, but the chief of staV convinced him it was necessary.

Roberts, Pronouns as Deﬁnites b.

c.

539

Every new recruit is armed. Last night, the president questioned whether they should be trusted so soon, but the chief of staV convinced him it was necessary. Every new recruit is armed. #Last night, the president questioned whether he should be trusted so soon, but the chief of staV convinced him it was necessary.

In the last example, the occasion of the discussion is so particular that it seems as if the president must have been talking about a particular recruit. And if we make the predicate on the singular pronoun sound like something that would not plausibly be true of (most) all instantiations of the relevant kind, this also gives rise to infelicity: (90)

Each candidate will be debriefed by Mrs Hendrix. #He had a practice press conference this morning that reminded him of the trauma of his Wrst college speech class.

In all of these examples, I would argue that the universally quantiWed NP is not a true antecedent, but only a licensing NP. Its use in the Wrst sentence implicates the existence (so long as the quantiWcation is not vacuous) of a salient group of individuals, and it is this weakly familiar group which can serve as antecedent to a plural pronoun. But there is no informationally (or semantically) unique member of that group that is made suYciently salient or distinguished in such a way that informational uniqueness would be satisWed for a discourse referent understood to be correlated with an atomic (nonplural) individual; hence, the unacceptability of singular pronouns in the general case. Whatever the proper analysis of telescoping, it is clearly a special type of case, where the universal instantiation is reminiscent of the continuation of irrealis mood in modal subordination. Plural pronouns often have quantiWcational licensing NPs which are non-universal in force. So, along with (91), we have (92): (91) (92)

Every man went for a walk in the park. They greeted each other as they passed. No man went for a walk in the park. They’re (all) staying home tonight.

Application of Neale’s (P5) to the interpretation of the pronoun in (92) would predict that it meant ‘the men who are walking in the park’, since this is the most impoverished deWnite description directly recoverable from [no x: man(x)] (is_walking_in_the_ park(x)] that denotes everything that both is a man and is walking in the park. Of course, this would lead to a contradiction, so he would predict that we would not say (92)—it would not be reasonable. But, of course, (92) has a sensible interpretation where it means ‘all the men are staying home’, obtainable on the assumption suggested above, that utterance of the Wrst sentence implicates the existence of a relevant group of men, which is hence both weakly familiar and informationally unique.

540

V. Anaphoric Pronouns and Dynamic Semantics

Given the problems cited by Kanazawa and the other examples just discussed, I conclude that the number neutral hypothesis regarding E-type pronouns is not viable.17 Of course, if Neale and other proponents of the analysis of pronouns as disguised deWnite descriptions were to adopt the hypothesis I proposed in the previous section, and treat existential interpretations of donkey sentences as their basic meaning, with the universal interpretations derived via pragmatic strengthening, then the lack of number neutral readings is not a problem. However, if we can obtain all the types of reading cited here—including the generic interpretations, sage-plant sentences, and non-generic plural pronouns with non-existential quantiWcational antecedents—with the theory outlined above, why posit the E-type interpretation? The status of rules for pronoun interpretation in E-type theories A Wnal problem with theories of pronouns as disguised deWnite descriptions is the question of the status of the rules that would derive the descriptive content of the disguised deWnite. We have already seen several examples which argue that we cannot account for all of the pronominal interpretations of interest with an algorithmic rule like Neale’s (P5), based solely on the descriptive content and syntactic context of the purported antecedent NP. Moreover, algorithmic rules will not work in cases of what Neale calls pronominal contradiction, like the following: (93) (94)

A: A man jumped oV the cliV. B: He didn’t jump, he was pushed. (Strawson) A: Every time I was there, a man jumped oV the cliV. B: I bet that in most cases he didn’t jump but was pushed.

Neale argues that we might give a referential interpretation to he in (93), taking it to refer to the individual raised to salience by the utterance of the sentence containing its antecedent, here ‘the man who jumped oV the cliV’. And he also mentions Davies’s (1981) suggestion that pronouns in such examples have an ‘ironical’ character, captured with scare quotes: ‘the man who ‘‘jumped’’ oV the cliV didn’t jump’. But (94), where the phenomenon occurs under quantiWcation, argues that referential analysis to a particular individual does not oVer a general solution to the problem. And Davies’s solution seems forced, and entirely implausible under quantiWcation, as in (94). But a more pragmatic account would permit us to understand the second part of (94) as meaning ‘the man who you said jumped’ or ‘the man who was seen to fall oV the cliV ’, or the like, any of them readily retrievable, and hence accommodable, in the given context. 17 See ch. 3 of Roberts (1990: s. 3.2.3), for a preliminary discussion of matters pertaining to number in NPs. There, I argue that, while singular NPs are semantically singular, plural NPs denote elements in the supremum of the denotation of their head, so that they include atomic individuals. Whether we understand a syntactically plural NP to be semantically plural or not is then largely a matter of pragmatics. This is a complex question and deserves to be explored in much more detail. But it is almost certain that pragmatics is as important as semantics in understanding what is conveyed by nominal number.

Roberts, Pronouns as Deﬁnites

541

In conclusion, the theories in question are basically theories where we accommodate additional descriptive content for a pronoun, either algorithmically as a function of syntactic structure or through pragmatic retrieval. The algorithmic approaches fail to be suYciently Xexible to handle the full range of cases. And pragmatic retrieval, wherein it is not so much that pronouns are hidden deWnite descriptions as that they seem to act like them in certain contexts, leaves lots of important questions unanswered. To answer these questions, you need a pragmatic theory, one which elucidates notions like Relevance, salience, and accessibility, and gives an adequate and general accounting of the ways we update information in context. It is my contention that when you answer these questions, you’ll come out with a theory much like the one proposed in section 2 above.

4. Conclusion: Pronouns are Unambiguous DeWnites The theory oVered in section 2 can handle all of the examples discussed in connection with the theories in section 3. For example, the paycheck sentences in (50)–(52), repeated here, involve contextual entailments: (50) (51) (52)

The man who gave his paycheck to his wife was smarter than the onei who gave it to his mistress. Every woman who gave her paycheck to her broker was smarter than any womani who gave it to her lover. Every graduate student put his paycheck in the bank. Every professori lost it.

We assume, at least by default, that all men receive a regular paycheck. In (50), reference to his paycheck in the Wrst relative clause makes these paychecks maximally salient, so that the pronoun in the second relative clause can pick up the entailed, salient paycheck for the less wise man. (51) sounds strikingly feminist partly because of the assumption that every woman not only has a paycheck, but a broker as well. In most academic departments in the US today, graduate students are also employees in some sense, as are professors, so default assumptions license the anaphora in (52), as well. All of the problems for the E-type theories cited in section 3 can be handled straightforwardly by the pragmatically based theory in section 2. The theory eschews semantic uniqueness in favor of informational uniqueness: deWnites make use of facts about the interlocutors’ shared common ground of information in order to diVerentiate quickly and accurately particulars with whom interlocutors are familiar. Pronouns can do so even more eYciently because their potential range of interpretation is limited to those discourse referents which are maximally salient. Because pronouns pick out discourse referents, and the means of keeping track of discourse referents in dynamic theories

542

V. Anaphoric Pronouns and Dynamic Semantics

encodes facts about scope, accessibility constraints fall out of this type of theory without the need for stipulation. There is no need for ambiguity if we treat all pronouns as simple deWnites with impoverished descriptive content. Pronouns and other deWnites can take as antecedents discourse referents for all kinds of entities—Wctional entities or abstract entities like natural kinds can be as familiar and salient as concrete individuals. This permits a uniWcation of the account of pronouns which have true, co-referential NP antecedents with that of pronouns which are deictic, pronouns that refer to a highly salient individual not previously mentioned, pronouns in intensional (modal, temporal) subordination, generic pronouns, and pronouns whose non-c-commanding licensing NPs are quantiWcational, along with others whose discourse referent antecedents are merely weakly familiar. And in doing so, we can respect the semantics and the pragmatics of pronominal number. If we take these all to be separate cases, we fail to see that the same general principles govern the retrieval of the intended interpretation in all these cases, as in the retrieval of other kinds of context-sensitivity in discourse. These are pragmatic principles and cannot be reduced to a purely structure-based algorithm, for they depend on context, background, and the perceived intent of the speaker, as well. A few years ago, the type of theory of pronoun interpretation proposed here would have been taken to be too weak to be of interest, precisely because of what I now take to be its strength: it is crucially based on assumptions about the nature and operation of pragmatic factors in interpretation. Although there is not space here to present this material in any detail, note that I assume that all the pragmatic principles involved in this account are independently motivated within a general theory of interpretation in discourse (Roberts 1996b, forthcoming). Moreover, the virtue of an adequately constrained, explicit formal theory of information structure in discourse is that it permits one to make predictions about when a particular reading will or will not be possible and felicitous in a particular context. Views of what context is like other than that of Roberts (1996b) are clearly possible; for example, see Asher and Lascarides (1998a, b) for a rather diVerent approach. But this sort of theory of the role of discourse context in interpretation is the general ground on which our understanding of deWnites, anaphora resolution, and other types of context-sensitivity must be worked out. Syntax, semantics, and algorithmic rules alone cannot suYce, because of the inherently pragmatic features of these phenomena. Of course, throughout this chapter though I have been talking about pronouns simpliciter, all the evidence I provided was about pronouns in English. It should be clear that, before one can make any really sweeping generalizations about pronominal anaphora in natural language, one would have to carefully examine a wide range of types of examples in a very wide range of languages, something I have not attempted to do here. So the results are really much more limited than the discussion above might suggest. In particular, there are languages, including the Slavic languages, Japanese, and many others, which have no deWnite article. And there are many languages, including Spanish and Japanese, where null arguments are often (or even almost

Roberts, Pronouns as Deﬁnites

543

always) preferred to overt pronouns; in such languages we might expect signiWcant additional constraints on the use of pronouns, and perhaps additional implications in their meaning. However, all the languages with which I am familiar seem to have demonstrative pronouns, and those were arguably the historical source of the deWnite article and of pronouns in English. I think we know enough about what pronouns do that we can conjecture that some combination of factors discussed in section 2, ultimately constituting a cluster of properties deWning deWniteness, will turn out to underlie the use and meaning of pronouns, as well as null arguments, in all human languages.

16 Dynamic DeWnite Descriptions, Implicit Arguments, and Familiarity Alice G. B. ter Meulen

1. Summary Asserted, presupposed, implicit, and inferred existential noun-phrases diVer importantly in dynamic binding potential. Implicit existential arguments of predicates cannot constitute antecedents for pronouns, yet they bind the presuppositions of deWnite descriptions in subsequent sentences and support inferred existentially quantiWed statements. A proper semantics must allow for free variables and an inference rule of existential generalization, a standard deductive rule in predicate logic. Lambda DRT principally separates the composition of a lambda-term for VPs from the declaration of its reference markers for NPs, projected from nuclear argument structure to constitute accessible antecedents. This division of labor of composition and binding restores the balance between representation and inference to model human reasoning in cognitive science, where natural language semantics may be usefully connecting to empirical cognitive or psycholinguistic modeling.

2. The Issue In dynamic semantics the content of a sentence, commonly assumed to be true, constitutes an instruction to update the context, which has already been agreed upon, the so-called common ground (Stalnaker 1999). Of course, this updating process must take into account the often complex background in which the new information is given, material circumstances such as its time of utterance and its prosodic properties, and a variety of constraints based in our shared knowledge of the language and understanding of the world. In preparing to present information in a sentence, important though largely unconscious choices must be made regarding the way the information is to be connected to the information that has already been shared, as well as the way in

ter Meulen, Dynamic DeWnite Descriptions 545 which speakers envisage to continue. It is of particular importance to analyze in detail diVerences created by asserting information or presupposing it, or leaving it implicit or entailed, in order to understand how the common ground is structured and restructured during a conversation. Information that is left implicit, tacitly understood, or presupposed does not serve to constitute the context in the same way as asserting information does. Presuppositions appear to have a more dual function, as they may aVect the context by introducing a new referent, but often get bound to already familiar referents. This chapter is intended to contribute to our understanding of the diVerent ways contexts are constituted in the process of interpreting discourse and reasoning with the information it provides. Ordinarily indeWnite noun-phrases (NP) introduce a novel, unfamiliar referent into the context, whereas deWnite noun-phrases require their referent to be available already in the context.1 Both deWnite and indeWnite NPs are interpreted as referring to discourse referents, which remain available for binding pronouns across sentential boundaries, whereas singular quantiWcational NPs in general cannot bind across sentence boundaries. But in ordinary English an argument of a predicate may be suppressed, yet we understand it to refer to an unfamiliar referent. Such suppressed arguments are called ‘implicit arguments’, which serve to optimize eYcient communication in speciWc ways. Implicit arguments do contribute to the context, but their contribution, even when they are interpreted existentially, should not be equated with that of indeWnite NPs. Implicit arguments are not asserted, but inferred from the sentence which contains the predicate in which argument structure they serve. A simple example serves to illustrate how implicit arguments may bind deWnite descriptions but not pronouns, though both are interpreted as referring to an already familiar referent. Although anyone readily infers from (1a) that one of the ten marbles is not in the bag, the pronoun it in (1b) cannot refer to that one marble outside the bag, whose existence is entailed in (1a).2 (1)

a. Only nine of the ten marbles are in the bag. b. It is under the sofa.

The sentence (2a), truth-conditionally equivalent to (1a), does assert of a marble that it is not in the bag. This asserted indeWnite NP one of the ten marbles provides an antecedent for the pronoun it in the subsequent sentence (2b), if it is to be interpreted coreferentially. (2)

a. One of the ten marbles is not in the bag. b. It is under the sofa.

1 This distinction between deWnite and indeWnite NPs is addressed in a large amount of linguistic literature. See Heim (1982), Reuland and ter Meulen (1987) and Kamp and Reyle (1993) for good starting points. 2 The marbles example is attributed to Barbara Hall Partee in Heim (pers. comm.). It was Wrst used to argue that truth-conditionally equivalent sentences may diVer in their anaphoric potential. Here its use is extended to argue that inferred indeWnites are not accessed in resolving pronouns, and pronouns require overtly asserted antecedents.

546

V. Anaphoric Pronouns and Dynamic Semantics

Originally these observations in (1) and (2) served to show that logically equivalent sentences, which must be true in exactly the same situations, may diVer in providing antecedents for binding pronouns across sentential boundaries, their ‘dynamic binding potential’. In addition it is clear from (1) and (2) that pronouns generally require overtly asserted antecedents, although there may be a limited set of exceptional cases where bridging inferences, based on indeWnite NPs in tacit common-sense assumptions, provide such antecedents.3 Although deWnite descriptions and pronouns are both deWnite NPs, which require that their referent is already given, (3) shows that deWnite descriptions, unlike pronouns, may easily be interpreted as co-referential with the indeWnite inferred from asserting (1a). (3)

The missing marble is under the sofa.

The presupposition of the deWnite description the missing marble is the existential claim that a marble is missing. Such a presupposition carries a weak familiarity requirement that may allow it to be resolved by referents of matching NPs entailed by the prior discourse, if there is no asserted antecedent (cf. Van der Sandt 1992). Hence in this context, the presupposed marble is not a new one, even though the presupposition is existential and hence indeWnite. We understand the presupposition to refer anaphorically to one of the ten marbles, and presuppose of it that it is missing from the bag. This presupposition obviously matches the entailment of (1a) that one of the ten marbles is not in the bag. The presupposition of (3) and the entailment of (1a) are merged in a resolution of the underlying variables. As a last resort, presupposed indeWnites may be accommodated globally, if neither an inferred antecedent nor an asserted one is available, overruling their weak familiarity requirement. Global accommodation would introduce a novel referent into the context by interpreting the existential presupposition as having widest possible scope over the entire discourse. It is clear that such global accommodation would be rather unnatural for our marble case, as we would no longer be able to determine what that new marble was missing from. However, not every context allows equally easily for the introduction of novel referents by simply asserting an indeWnite NP. In some cases the novelty of the referent must be explicitly asserted in a disjoint reference condition, which ensures that the new referent is distinguished from any of the referents already in the context. Consider, for instance, the perhaps slightly unnatural discourse in (4a), where a simple indeWnite NP is used twice. Intuitively, (4a) should allow for the two occurrences of a book to co-refer, for there is no semantic reason to think that the book referred to by the second occurrence must be diVerent from the book referred to by the Wrst occurrence, as the novelty requirement would enforce. Taking pragmatic Gricean 3 A clear case of an inferred antecedent for pronoun resolution is the following bridging inference: (i) Jane is pregnant. (ii) It is a girl. (iii) Jane carries a baby. The common-sense knowledge in (iii), inferred from (i), that ordinarily a pregnant woman carries only one baby apparently suYces to constrain the situation to a unique potential referent of the pronoun it in (ii), where presuppositions support such binding.

ter Meulen, Dynamic DeWnite Descriptions 547 conversational maxims into consideration, whenever there are two books intended, the speaker is required to say so, as in (4b). If John is reading the same book for a second time, again the speaker should say so, using (4c) with a deWnite NP to ensure coreference to the given book, supported in addition by the presupposition triggering adverb again. (4)

a. John has read a book. He is reading a book. b. John has read a book. He is now reading another book. c. John has read a book. He is reading the book again.

The problem here is to explain why (4a) is logically entailed by both (4b) and (4c), although the novelty requirement of indeWnites would strictly speaking make (4a) incompatible with (4c). It seems that inferring an indeWnite weakens its novelty condition, and prefers an anaphoric interpretation, rather like the anaphoric character of presupposed indeWnites, as argued above in the marble case. The novelty requirement of indeWnites is clearly contextually constrained, and in certain contexts the novelty of a referent needs to be asserted or explicitly stated in a disjoint reference condition. Insight is needed into underlying causes of these diVerences in the dynamic binding potential of asserted, inferred, or presupposed indeWnites.

3. Binding with Implicit Arguments In ordinary English ellipsis may allow us to use a verbal predicate without overtly expressing all of its arguments, even though they are understood to contribute to the interpretation of the sentence. Verbal predicates always require a Wxed number of arguments syntactically, but, unlike the ‘arity’ of predicates in any syntactically disambiguated and fully explicit logical language, natural languages do not require all of these arguments to be overtly expressed by full noun-phrases. Obviously, in agentless passive constructions, the argument for the causer of the action described by the verb is suppressed, but still entailed, as in (5). (5)

a. John washed a car. b. A car was washed.

In (5b) the passive construction of the transitive verb wash may optionally delete its original subject John in the corresponding active (5a). However, (5a) and (5b) each entail that someone washed a car, and (5a) entails (5b), but not vice versa. If the agent argument is not overtly asserted, it is not available as antecedent for binding pronouns, neither sentence internally, as in (6a), nor across sentential boundaries, as in (6b). (6)

a. Every car that was washed, earned him $20. b. A car was washed. He earned $20.

548

V. Anaphoric Pronouns and Dynamic Semantics

From (6) is it evident that implicit NPs, just like entailed indeWnites, do not constitute antecedents for binding pronouns. As we have already learnt from the marble case in (1)–(3), implicit indeWnites are still available to bind deWnite descriptions, which trigger existential presuppositions, as in (7). (7)

a. Every car that was washed, earned the washer/kids $20. b. A car was washed. The washer/kids earned $20.

The implicit agent in (7a) may be used to bind the presupposition of the washer or the kids, supporting the inference that the kids washed the cars. Since implicit arguments have no way of carrying syntactic agreement features, they bind both singular and plural deWnite descriptions. Similarly, in (7b) the implicit agent of the washing is understood to be co-referential with the deWnite description in the second sentence, regardless of number. Besides passives, agentless resultatives in complements of light verbs behave on a par with entailed indeWnites and implicit arguments, as we see in (8a). (8)

a. John had his car washed. {The guy= He} did a good job. b. John had someone wash his car. {The guy/He} did a good job.

From the Wrst sentence in both (8a) and (8b), a disjoint reference condition is inferred, requiring that someone other than John washed John’s car. This must be based on the general syntactic principle that arguments of the same predicate cannot be interpreted as co-referential, unless they are reXexively marked. The relevant predicate here is the light verb have, which creates an exceptional case marking complement, assigning the subject of the washing accusative case, if it is overtly expressed.4 It is important to realize that neither this disjoint reference inference nor the binding of the presupposition constitute what is commonly called a ‘bridging inference’, which must appeal to common-sense or normal possible worlds.5 A paradigmatic case of a bridging inference in (9) requires the default conditional (9c) to correlate the house and the kitchen, in order to support the intuitively valid inference that it is the kitchen in John’s house that needs remodeling. (9)

a. John bought a house. b. The kitchen needs to be remodeled. c. Every house normally has a kitchen.

Bridging is a way of linking two descriptive predicates, where one is often deWnite, but its antecedent is not explicitly asserted. The disjoint reference condition derived from (8a, b) does not appeal to any notion of possible worlds at all, and hence avoids having 4 Exceptional case marking is studied extensively in generative syntax. Radford (1997) constitutes a good Wrst reference. 5 Ko¨nig and Mauner (2000) argue that the accommodation of presuppositions by binding it to an implicit argument requires bridging. Although bridging may be needed for their particular examples with French impersonal subjects, it is clear from (8) that bridging is by no means required to get binding with implicit arguments as antecedents.

ter Meulen, Dynamic DeWnite Descriptions 549 to spell out the notoriously diYcult notion of what would constitute a ‘normal’ possible world. The disjoint reference condition is based in the lexical syntactic argument structure of the verb wash, which requires a causer argument, the one who controls and performs the action, and a second, theme argument for what is undergoing the washing. The argument structure of predicates is wired into the lexicon of the language, as a basic relation between its thematic roles. Other uses of the same verb, its alternations, may be derived from this lexical entry by speciWc lexical operations. For the present purposes it suYces to conclude that implicit arguments in agentless resultatives behave semantically entirely analogously to entailed indeWnites and anaphoric presuppositions. Summarizing what we have concluded from this review of the linguistic data, it is clear that the binding of pronouns and deWnite descriptions should be diVerentiated to explain why only the latter bind to implicit arguments or entailed indeWnites. This requires a fundamental separation of the compositional interpretation of verb-phrases (VP), where the full argument structure of the verb must be present, from the dynamic binding potential of its arguments. Only asserted NPs may be related to declared reference markers, if they are to serve as antecedents for pronouns or deWnite descriptions. As implicit arguments and entailed indeWnites cannot constitute antecedents for pronouns, they should not declare a referent in the context. The existential presupposition carried by all deWnite descriptions may be attributed the semantic force to declare a reference marker with an associated weak novelty condition, seeking local binding. That reference marker, once declared, may serve to state an identity condition with the implicit argument, eVecting a binding between the deWnite description and its implicit argument antecedent. To make this sketch of an account more precise calls for a reWned account of the diVerences in anaphoric potential of asserted, inferred, implicit, and presupposed indeWnites, and the associated locality constraints on the accommodation of existential presuppositions.

4. Implicit Arguments in Lambda DRT In Kamp and Reyle (1993) a semantic theory is presented, Discourse Representation Theory (DRT), in which indeWnite NPs always introduce new discourse referents, also called ‘reference markers’, into the domain.6 DeWnite NPs, which include pronouns as well as deWnite descriptions, require identiWcation with the accessible reference marker of the intended antecedent, which must already be present in the Discourse Representation Structure (DRS). An essential step in the process of interpretation of NPs is the declaration of their discourse referent at the level at which the NPs are interpreted, 6 See Kamp and Reyle (1993) for the standard introduction to DRT. A basic understanding of the construction of DRSs and the dynamic semantics of DRT is presupposed in the remainder of this chapter.

550

V. Anaphoric Pronouns and Dynamic Semantics

except for proper names, which are always declared at the top level. QuantiWcational NPs create a layered DRS, where the determiner (e.g. every, most, all) is interpreted as a relation between the restrictor and the subordinate nuclear scope. At least the common noun of the quantiWcational NP, possibly with its restrictive relative clause, is mapped into the restrictor and the VP of the sentence into the nuclear scope. Although the wellformedness conditions on Discourse Representation Structures allow for a reference marker to be used in its conditions, consisting of predicates with their arguments, without having to be declared, no construction rule in the representational system of Kamp and Reyle (1993) actually makes use of this possibility.7 Declaring a reference marker renders it equally accessible for pronoun resolution and presupposition accommodation at that level or at any subordinate one, where the nuclear scope of a quantiWer is considered subordinate to its restrictor. Reference markers that remain undeclared are existentially closed by embedding the DRS into the model to determine its truth conditions. Undeclared reference markers, like the free variables of predicate logic, are interpreted accordingly as wide scope existential quantiWers over the entire DRS. There is no inference rule of existential generalization in DRT, which would support the inference of an indeWnite NP in a conclusion, based on an undeclared reference marker used in a condition in the DRS. The tools of DRT as such do not oVer any account of why implicit arguments or entailed indeWnite NPs may bind the existential presuppositions of deWnite descriptions and yet fail to bind ordinary pronouns. What is needed is a way for implicit arguments to play their role in the argument structure within the VP interpretation, without having to be declared at any level. If implicit arguments act as such undeclared reference markers, they should not be forced to receive a wide scope interpretation, for their scope tends to be narrow and dependent on any quantiWcational NPs in the sentence, as in (10). (10)

Every farmer who had his car washed paid the guy twenty dollars.

In (10) the implicit argument for the washer of the car may, but need not, refer to diVerent individuals depending on the choice of farmer, and yet each washer so picked must be able to serve as referent for the deWnite description the guy. Clearly in this narrow scope interpretation of the implicit argument for the washer, the existential presupposition of the deWnite description the guy is also captured within the restrictor of the quantiWer every farmer, where it is locally accommodated to co-refer with the implicit argument for the one washing the car. Existential closure of the undeclared reference markers would force only a wide scope reading of them, which would eliminate the preferred interpretation of (10), assigning the implicit argument narrow scope. The existential closure in DRT of all undeclared reference markers by the verifying embedding should hence somehow be prevented from applying to those undeclared reference markers that represent implicit arguments. 7 Ko¨nig and Mauner (2000) are the Wrst to propose that implicit arguments should be analyzed as undeclared reference markers in DRT. They do not seem to be aware of the consequence that they would be forced into wide scope by existential closure.

ter Meulen, Dynamic DeWnite Descriptions 551 To do justice to the fact that we infer indeWnites from implicit arguments by reasoning resembling the natural deduction rule of existential generalization, DRT needs to allow for indeWnites to be interpreted anaphorically, seeking local binding like presupposed indeWnites. The novelty condition associated with indeWnites must hence be weakened to prefer local binding for indeWnites that are inferred or presupposed. The DRT account of entailment simply adds the conclusion of an inference, as if it were asserted, to the DRS constructed for its premises, and seeks to extend every verifying embedding of the DRS for its premises to one which in addition veriWes the conclusion. This strategy of modeling inference cannot be maintained if asserted indeWnites do aVect the context, whereas inferred or presupposed indeWnites prefer to be interpreted as locally bound within the given context. If a theory of natural deduction could be welded onto DRT, there should be an easy way to deWne a rule of existential generalization, as well as other structural rules of inference. Such a solution would lighten the labor of DRS construction rules, and be likely to lead to a more eYcient form of processing, taking into account whether information was asserted, inferred, or presupposed. Finite as we human beings are, we often avoid asserting what is already entailed or may be inferred otherwise. We produce information eYciently by trying, with varying degrees of success, to maximize entailments and presuppositions as a general strategy of linguistic economy. This eYciency of sharing information in natural language should also be modeled in dynamic semantic systems, redistributing the division of labor between representation and inference. EYciency in human information processing should not be equivocated with the algorithmic eYciency of uniWcation in the computational sense of this notion. In the recent developments of dynamic semantic systems based on the principles of DRT, Blackburn and Bos (1999) and Kohlhase et al. (2000) designed a variant of DRT, called Lambda DRT, which constitutes a compositional discourse semantics with a solid deductive theory. The key idea of Lambda DRT is to separate the composition of the VP interpretation by lambda abstraction over its arguments, well known from Montague’s famous paper PTQ (Montague 1974b), from the declaration of its NPs to render them accessible to pronoun or presupposition resolution. For the present purposes the main gain of Lambda DRT is its allowing for lambda abstraction over the implicit argument, which avoids forcing its wide scope interpretation by the verifying embedding. The presupposed indeWnite of a deWnite description declares a new reference marker, which seeks to bind that description to the implicit argument. Pronouns carry no such presupposition and hence do not declare a reference marker at all. On a par with all asserted NPs, both pronouns and deWnite descriptions are still interpreted as generalized quantiWers as in PTQ, that is, denoting sets of Wrst-order properties of individuals.8 Implicit arguments are, however, of the basic type e of individuals. They are not interpreted as generalized quantiWers, as they are syntactically 8 An elementary introduction to the theory of generalized quantiWers in natural language is given in ch. 14 of Partee et al. (1993).

552

V. Anaphoric Pronouns and Dynamic Semantics

not NPs, but remain implicit arguments within the VP. These are the essential ingredients to specify the semantics of implicit arguments and the associated issues as discussed above. Although the remainder of this chapter is of a somewhat more technical nature in comparison with the discussion thus far, the main ideas of the proposed analysis are presented informally in an attempt to render them accessible to more readers. To understand the technical motivation of Lambda DRT as a development from classical DRT, the interested reader is referred to Kohlhase et al. (2000), and references cited there. Anyone at ease with PTQ style formal semantic interpretation may Wnd the full logical details of the interpretation of implicit arguments speciWed in the appendix.

5. Declaration Versus Composition The language of Lambda DRT is an enriched logical language, which includes DRSs as well-formed formulae. A DRS is constructed from a set X of declared reference markers and a truth-conditional formula f, written as D{X}f. This enrichment facilitates not only a strictly compositional semantics, as opposed to the non-compositional classical form of DRT, but it also creates room for a proper theory of natural deduction to govern inferences, instead of putting all the weight into the DRS construction rules, which is still customary in classical DRT. In Lambda DRT the interpretation of a sentence is composed by an associative and commutative operation called ‘merge’ deWned over DRSs. Sentences in a discourse are composed by an associative and noncommutative (hence order-sensitive) operation called ‘join’ over their respective DRSs. Implicit arguments are interpreted as referring to individuals, but pronouns, interpreted as generalized quantiWers, denote sets of properties of individuals. Neither implicit arguments nor pronouns declare a reference marker, but both trigger a familiarity condition, which seeks to identify them with a suitable local antecedent. DeWnite descriptions, however, do declare the reference marker of their presupposed indeWnite, and merge it to its descriptive content. Accordingly, they are translated in Lambda DRT as given in (11), where ^ is the PTQ-style cap operator, raising the type of the expression to its intensional type, interpreted by functions from worlds to the corresponding extensional type. (11)

Implicit arguments Pronouns Def. descriptions

PROi (s)hei theji

^

xi lP P(^ xi ) lPlQ ((D{xi }xj ¼ xi ) P (^ xi ) Q (^ xi ))

Verbs or verbal predicates declare a reference marker for the event they describe, which is the Wrst argument in its argument structure. Their other arguments are not declared, but they are bound by the lambda abstraction, as is customary in PTQ. For extensional

ter Meulen, Dynamic DeWnite Descriptions 553 predicates, like wash, the arguments should refer to individuals, so the ‘cap/cup’cancellation of PTQ is applied to lower their initially intensional type to the required extensional one. The lexical entry for a simple extensional verb like wash is accordingly speciWed as in (12), where #¼ cup operator (i.e. ( # (^ x)) ¼ x). (12)

lv lu D{ea } WASH (ea , # u, # v)

wash

In (12) the lexical predicate is entered as a relation between an event argument ea , indexed for later binding, and two extensional individual denoting variables. The event argument is declared in D{ea }, and the undeclared NP arguments are not declared, but bound by lambda abstraction. Proper names are also interpreted as generalized quantiWers, which declare their reference marker too, as in (13). (13)

lP (D{xj } xj ¼ j P (^ xj ))

John

Ordinary common nouns again do not declare a reference marker, but provide a lambda-bound individual variable, as for the noun car in (14). (14)

car

(l z D { } CAR( # z))

A possessive pronoun as determiner creates a full NP with a common noun, so it must again be interpreted as a generalized quantiWer. Since this NP should be available as antecedent to bind subsequent pronouns, the resulting interpretation should declare a reference marker, seeking a given reference marker to identify the possessor, as in (15). (15)

his lP lQ (D{x i } poss: (x j , x i ) P (^ x i ) Q (^ x i ))

Accordingly, the compositional interpretation of the NP his car is obtained by lambda conversion and results in (16), where the individual variable xi is declared, and xj triggers a familiarity condition. (16)

his car

lQ (D{xi } poss: (x j , x i ) & CAR(xi ) Q (^ x i ))

Composing this NP as direct object with the lexical predicate wash, we obtain the VP interpretation with the argument u for its washer still lambda-bound, as in (17). The subscript on the variable serves to distinguish diVerent variables of type e, and the superscript carries the anaphoric information. Accordingly, in (17) the possessive pronoun his is indexed by the subscript i to distinguish it syntactically from x j , and the superscript j to mark it as semantically anaphoric to the agent argument of wash Xj . (17)

j

wash hisi car

lu D{ea , x i } poss: (x j , x i ) & CAR(x i ) & WASH(ea , # u, x i )

The next step is to merge the implicit argument y K with # u in the subject position of the VP, where syntactically the implicit argument is the phonologically null PRO in (18).

554

V. Anaphoric Pronouns and Dynamic Semantics (18)

j

PROK wash hisi car (ea , y K , x i )

D{ea , x i } poss:(x j , x i ) & CAR(x i ) & WASH

The conversion of the active clause in (18) to its passive counterpart deleting the agent aVects only its syntax, but not its semantic representation, resulting in (19). (19)

[IP his car[VP washed tj , ]] ) D{ea , x i } poss:(x j , x i ) & CAR(x i ) & WASH (ea , y K , x i )

The next step in the derivation is composing past tense marked the light verb have with (19), which results in (20), skipping here a few basic steps, fully speciWed in the appendix. (20)

hadb [IP his car [VP washed tj , ]] lz D{e a , e b , x i } poss: (x j , x i ) & CAR(x i ) & WASH(e a , y K , x i ) & HAVE (e b , z, e a ) & e a ¼ e b & e b precedes NOW :

The proper name John is merged into (20) to result in (21) as translation of the sentence John had his car washed. (21)

D{e a , e b x i , x j } x j ¼ j & poss:(x j , x i ) & CAR (x i ) & WASH (e a , y K , x i ) & HAVE (e b , j , e a ) & e a ¼ e b & e b precedes NOW :

If the second sentence, continuing this discourse, is He paid the guy cash, we interpret the pronoun he as co-referential with John, and the deWnite description the guy as coreferring to the implicit argument for the one washing John’s car. (22)

D{ec x n xm } y K ¼ x n & GUY(x n ) & PAY(ec , j , x n , xm ) & CASH(xm )

The deWnite description the guy carries a weak familiarity assumption of its presupposed indeWnite a guy. This is rendered as a dynamic binding to any available implicit argument from prior discourse, that remains undeclared in the domain, in casu yk . This localized accommodation of the presupposition triggered by the deWnite description prevents wide scope interpretation of such indeWnites and eVects the binding by any quantiWer in the restrictor, had the subject, for instance, been every student instead of John. A deeper, conceptual reason or cause may be desired to account for why presupposed indeWnites seek such local binding and implicit arguments remain available for exactly such binding, other than this technical simulation of how this may be accomplished compositionally. I can only remark that this may be an indication of the horror vacui in natural languages, well known from physics, appealing to overarching coherence considerations and our need to make sense. Joining the two clauses results in the union of the domain of declared reference markers, ensuring that the implicit argument yK itself is still not declared, but identiWed with and declared via this identiWcation with the argument for the deWnite description, as in (23).

ter Meulen, Dynamic DeWnite Descriptions 555 (23)

D{e a , e b x i , x j , ec , x n , xm }x j ¼ j & poss:(x j , x i ) & CAR(x i ) & WASH (e a , y K , x i ) & HAVE (e b , j , e a ) & e a ¼ e b & e b precedes NOW & y K ¼ x n & GUY(x n ) & PAY(ec , j , x n , xm ) & CASH(xm )

The Wnal semantic representation declares the reference markers for the three events, for the proper name John, and the deWnite NPs the car, the guy, and the cash, but not for the one who was washing the car, nor for the pronoun. This is the desired result, for the implicit argument cannot bind the pronoun, but it can be bound indirectly via the declared reference marker for the deWnite description and its associated weakly familiar existential presupposition, accommodated to co-refer with the implicit argument.

6. Conclusions The Lambda DRT analysis presented in this chapter made essential use of the separation of the compositional interpretation of the VP projection from the independent process of declaring reference markers to determine their binding potential. The novelty and familiarity conditions associated with respectively indeWnite and deWnite NPs are in this approach best understood as meta-constraints on reference markers, instead of being part of the DRS construction rules as in classical DRT. They determine that a new reference marker should be introduced only for asserted indeWnites, and one which is to be identiWed either locally, for presupposed or inferred indeWnites, or at any accessible level, for all remaining deWnite NPs. In this way you can have your cake and eat it too—compositional semantics with a structured semantic representation, determining accessibility, and binding. The structural insights from DRTare hence integrated in a fully compositional semantics with a dynamic, denotational semantics. This approach has been shown to account properly for the limited binding potential of implicit arguments, and allows for an important reorganization of the division of labor between semantic representation and inference. Instead of coding anything that may be inferred into the semantic representation, this Lambda DRT strategy aVords us to be light on representation, encode only what is asserted, but heavy on inference, let structural inference rules support all the entailments. It remains to be seen whether other theories of anaphora, such as E-type theories or game-theoretic semantics, may be extended to account for the data on binding by implicit arguments, but such considerations await another occasion (see Ter Meulen (1999) for a sketch of a game-theoretic analysis). The program to model human information processing as a very eYcient and highly structured form of information processing seeks to place more of a burden on a separate module for inference rules in order to alleviate the encoding of information at the representational level. This must also enhance its empirical interest to cognitive scientists and improve the testability of its predictions.

556

V. Anaphoric Pronouns and Dynamic Semantics

7. Appendix A fully detailed dynamic semantics of Lambda DRT is presented in Kohlhase et al. (2000), on which this appendix is based. The technical details of the derivations in the chapter are given here for readers who are familiar with formal, denotational semantics for natural language. For the details of the interaction of lambda conversion with the process of declaring reference markers the reader is referred to Kohlhase et al. (2000). The set of types is deWned as follows: TYPE: e—individuals d—individual concepts o—truthvalues t—DRS If a and b are types, then (a, b) is a type too. DRS construction allows for expressions to be DRSs themselves in the formal language, which are composed from sentences by the merging operation and combined with each other to represent sequences of sentences in discourse. The deWnition is as follows: (i)

X is an inWnite subset of distinguished variables, f is of type o, D{X}f is of type t, a DRS (ii) merge D1 D2 (associative, commutative) to compose sentences (iii) join D1; D2 (associative, not commutative) to compose discourse The main idea of the chapter was to associate declaring reference markers only with deWnite descriptions, but not with pronouns, though both are interpreted, as in PTQ, by generalized quantiWers. Implicit arguments Pronouns Def. descriptions

PRO i (s)he i the ji

^

xi lP P(^ xi ) lPlQ ((D{x i }x j ¼ x i ) P(^ x i ) Q (^ x i ))

The derivation of John had his car washed (#¼ cup operator): wash lv lu D{e a }WASH(e a , # u, # v) j hisi car lP lQ (D{x i }poss:(x j , x i ) P(^ x i ) Q (^ x i ))(lzD{ }CAR( # z)) ) lQ (D{x i }poss:(x j , x i ) & CAR(x i ) Q (^ x i )) j wash hisi car lu D{e a , x i }poss:(x j , x i ) & CAR(x i ) & WASH(e a , # u, x i ) j PROK wash hisi car D{e a , x i }poss:(x j , x i ) & CAR(x i ) & WASH(e a , y K , x i ) [IP his car [VP washed tj , ]] ) D{e a , x i }poss:(x j , x i ) & CAR(x i ) & WASH (e a , y K , x i ) leluD {e a e b }HAVE(e b , # u, # e) & e a ¼ e b & e b precedes NOW : hada hada [IP his car [VP washed tj , ]]

ter Meulen, Dynamic DeWnite Descriptions 557

) John

lR lzD{e a , e b x i }poss:(x j , x i ) & CAR(x i ) & WASH(e a , y K , x i ) R(^ z, ^ e b ) (lelu D{e a , e b } HAVE (e b , # u, # e)) & e a ¼ e b & e b precedes NOW : lzD {e a , e b x i }poss: (x j , x i )&CAR(x i )&WASH(e a , y K , x i ) & HAVE (e a , z, e b ) & e a ¼ e b & e b precedes NOW : lP(D{x j }x j ¼ j P(^ x j ))

John had his car washed D{e a , e b , x i , x j } x j ¼ j & poss:(x j , x i ) & CAR(x i ) & WASH(e a , y K , x i ) & HAVE (e a , z, e b ) & e a ¼ e b & e b precedes NOW : The second sentence He paid the guy cash is interpreted as follows: pay cash the guy pay the guy cash he paid the guy cash

) lvlwluD {ec } PAY (ec , # u, , # w, # v) ) lyD{ }CASH( # y) ) lPlQ ((D{xz }xs ¼ xz GUY (^ xz ) Q (^ xz )) ) luD{ec xz xm }xs ¼ xz GUY(xz ) PAY(ec , # u, xz , xm ) & CASH(xm ) ) D{ec xz xm }xs ¼ xz & GUY(xz ) & PAY(ec , j , xz , xm ) & CASH(xm )

Joining this second sentence to the Wrst one produces the Wnal result: D{e a , e b x i , x j ec xz xm }x j ¼ j & poss:(x j , x i ) & CAR(x i ) & WASH (e a , y K , x i ) & HAVE (e b , z, e a ) & e a ¼ e b & e b precedes NOW & y K ¼ xz & GUY(xz ) & PAY(ec , j , xz , xm ) & CASH(xm ) The free y K is now forced to refer to the same individual, identiWed with the guy who was paid cash by John. This reference marker y K for the implicit argument was never declared and hence never made available for binding pronouns. It was identiWed in accommodating the familiarity condition of the deWnite description via its weak indeWnite presupposition xs ¼ xz , by unifying y K ¼ xz . This concludes the presentation of the analysis of binding with implicit arguments.

17 IndeWnites and Scope Choice Ruth Kempson and Wilfried Meyer-Viol

1. Introduction Ever since the early 1980s, there has been recognition that the intrinsic contribution of a natural language expression systematically under-determines the way it is interpreted in any given context. In the initial part of this period, there were Werce debates between Fodor and situation theorists, principally Barwise (see Fodor 1981, 1983, 1987; Barwise and Perry 1983; Barwise 1988) as to whether this phenomenon should be modeled relative to the assumption that languages are directly interpreted (much as the familiar formal languages, though with an enriched semantic vocabulary), or rather only indirectly with an intermediary process mapping such strings into some form of representation (for Fodor some ‘language of thought’) which is transparent with respect to such interpretation and nontrivially distinct from the natural-language string itself. In the manner of academic controversies, this debate has been largely set aside, with the protagonists developing independent lines of research leaving the issue unresolved. In the subsequent decade, formal theories of context-dependency have in the main implicitly sided with the Barwise anti-representationalist stance, by taking the phenomenon of context-dependence to impose the challenge of deWning concepts of context within some semantic algebra (Groenendijk and Stokhof 1991; Milward 1995; Rooth 1992). By this route, the concept of deWning a notion of denotational content for natural language expressions, albeit more abstract, is preserved. Contrarily, informal work from a pragmatic perspective has retained the assumption that natural languages are interpreted only indirectly, via the construction of a propositional form representing content, a form which is only partially determined by language-internal speciWcations, hence presuming that denotational content is deWned over propositional structures which result from interaction between language-internal speciWcations and general cognitive processes (Sperber and Wilson 1986/1995; Carston 1988, 2002). Intermediate between these two positions is Discourse Representation Theory, DRT (Kamp and Reyle 1993). DRT deWnes a construction algorithm mapping syntactic structures onto Discourse Representation Structures. These structures are themselves model-theoretically deWned as partial models, for which embeddability

Kempson & Meyer-Viol, Scope Choice

559

conditions are deWned into some overall model. In so far as the articulation of a model does not involve commitment to an additional level of structural representation, these intermediate structures might be seen as part of a program in which natural languages are directly model-theoretically interpreted objects. However, in so far as structural properties of the discourse representation structures may be manipulated in determining the interpretation to be assigned to a sentence, they constitute an essential level of representation intermediate between the characterization of the natural language string and the model-theoretic evaluation of its assigned content. Despite the lack of continuing debate about the so-called representationalist position within the semantic literature, the attempts to address under-speciWcation and the nature of the gap between speciWcations of linguistic content for individual expressions and their interpretation relative to context have led—following the impetus of Discourse Representation Theory—to the articulation of increasingly rich concepts of structure within the semantic algebra. Two recent examples are within situation theory and categorial grammar, both originally bastions of the view that natural languages are assigned denotational content directly without intervening structural conWgurations. In situation theory, proof-theoretic models of natural language content are being articulated and then used, amongst other things, to characterize the phenomenon of natural-language context-dependency (Milward 1995; Cooper 1998). And in typetheoretic formalisms of categorial grammar, proof-theoretic concepts of context are deWned (as arbitrary sets of premises) so that anaphora resolution is deWned as a process of term-substitution, with proof-theoretic constructs such as labels of proof-terms able to be manipulated as input to such a substitution process (Ranta 1994; Fernando 2000). The result is a coming together of what historically have been divergent representationalist and nonrepresentationalist perspectives about the nature of natural-language interpretation. In the light of this coincidence of originally opposing viewpoints, it is worth providing a reminder of the empirical reasons for adopting representationalist assumptions about natural language interpretation. If we suppose to the contrary that natural language expressions are deWned as having a denotational content directly assigned, the phenomenon of the context-dependent variation in interpretation immediately poses the puzzle of widespread cross-categorial ambiguity. For example in the muchstudied case of anaphora, despite repeated eVorts to analyse away such ambiguity, some distinctions, for example, between discourse and bound-variable anaphora, remain irreducible and hence have to be deWned as projected from discrete lexical items. This solution leaves completely unexplained the fact that all anaphoric expressions in all languages display such ambiguity: languages in general do not have distinct anaphoric forms for such apparently discrete uses.1 This ambiguity is, moreover, observable across pronouns, deWnite and demonstrative noun-phrases, tense and temporal adver1 Zribi-Hertz and Mbolatianavalona (1999) argue that Malagasy provides a language in which an itemized form of pronoun is reserved for bound-variable construals, but it is notable that this language is highly exceptional in displaying such a property.

560

V. Anaphoric Pronouns and Dynamic Semantics

bial construal. The conclusion that all anaphoric and temporal expressions are intrinsically ambiguous though consistent with the facts of denotational ambiguity is disappointing, for it is an implicit assertion that there is no uniform explanation of what it means to be an anaphoric expression—only a number of discrete answers. The problem is that to establish a unitary basis from which all interpretations can be said to stem demands an analysis of intrinsic content for the anaphoric expression which is weaker than any individual use of it (Kempson et al. 2001 provide detailed exempliWcation across pronouns, deWnite noun-phrases, demonstrative noun-phrases involving that and this as determiners, and tense: see also Kamp and van Eijck 1997). This threat of ineliminable diversity of the natural-language vehicle (i.e. word) at least with respect to bound-variable and indexical construals of pronouns (with resumptive uses as an unaddressed addition) is removed if such words are said to project some partial speciWcation, a mere place-holding device relative to which a substituend has to be provided, on the basis of which some structural conWguration is built up and subsequently evaluated (see Kamp and van Eijck 1997; Kempson et al. 2001 for detailed arguments to this eVect). Ellipsis phenomena constitute a second area where the representationalist debate is re-emerging: (1) (2) (3)

Joe cut himself but Susie managed not to. John saw his mother damage his paintings and Bill did too. You’re sitting on my chair. No, I’m not.

At one extreme, there have been those who advocate analyzing elliptical strings as having syntactic structure with some unrealized elements (Fiengo and May 1994; and many within the minimalist paradigm). At the other, there have been those who advocate analyzing elliptical strings via an interpretation procedure articulated exclusively within the semantic algebra (Hardt 1993). Intermediate between these is the view that interpretation of such strings involves the projection of structure as the Wrst step in a construction process, of which only the output is model-theoretically evaluated. The signiWcance of ellipsis is that it provides evidence buttressing that of the anaphora debate: the mapping from linguistic input onto a conWguration for evaluation of elliptical strings is essentially structural and yet this output conWguration is not inhabited by words of the language under analysis (Kempson 1988, 1996; Kempson et al. 1999). The ellipsis reconstruction required in (1)–(3) for example involves a conWguration over which binding principles have to hold, yet the result of building up the interpretation for the fragmentary form from its preceding clause must not preserve the form of words of that antecedent string.2 In this chapter, we add to this growing representationalist picture the phenomenon of quantiWer scope ambiguity. As in other areas of interpretation, analyses of quantiWer 2 On the assumption that ellipsis is a process of copy and delete, a process of ‘vehicle reconstruction’ is posited to provide for such copying without morphological identity. See Kempson et al. (1999) who draw out the relevance of these data to the representationalism debate.

Kempson & Meyer-Viol, Scope Choice

561

scope ambiguities have divided into three families according to whether: (i) the phenomenon is seen as exclusively syntactic, with a structural operation mapping the structure corresponding to the string onto a structure at which scope is fully determined by a process of covert movement (‘covert’ because not reXected in the surface sequence of words—this is the movement analysis popular since May 1985); (ii) the phenomenon is modeled as exclusively semantic, with a process of evaluation projecting the varying interpretations as ambiguities of content from a single syntactic structure as input (Cooper 1983); (iii) the phenomenon is modeled in the semantic algebra but involving semantic under-speciWcation, hence with additional intermediate partial denotations from which mapping operations determine the scope-diVerentiated interpretations (Kempson and Cormack 1981; Reyle 1993; Erk et al. 2002; Kallmeyer and Joshi 2002). We shall argue for a fourth position, that the phenomenon of quantiWer scope should be analyzed in terms of a structure-building process of term construction with place-holders which require substitution—in other words, directly analogous to the process of anaphora resolution. So the analysis is one which involves under-speciWcation of content and a process of completing that speciWcation that is both structural and subject to pragmatic constraints. In this chapter we set out: • the puzzle provided by quantiWer scope; • the informal presentation of the view that quantiWed expressions inherently project incomplete speciWcations of content, with indeWnites requiring pragmatically driven choices to complete such speciWcations, directly analogous to pronoun construal; • a brief presentation of the formal account which gives substance to this view with a sketch of the framework within which this is embedded; • an outline of the consequences that follow if this view is correct.

2. The Problem of Scope Variation In all discussions of quantiWer scope variation in natural language, one assumption remains constant. A reconstruction of scope construal between one expression and another has to be deWned at a propositional level to reXect the fact that quantifying expressions are propositional operators, functions from (open) proposition to (possibly open) proposition. This very familiar assumption is illustrated by the fact that sentences containing more than one quantiWer can be interpreted in a number of ways within a clause according as the one is taken to have scope over the other:3 3 English speakers report a preference for an interpretation of (5) reXecting linear order, with (5) accordingly describing that one nurse interviewed the full set of patients, but also allow as possible an interpretation in which not all patients are interviewed by the same nurse.

562

V. Anaphoric Pronouns and Dynamic Semantics (4) (5)

Every nurse interviewed a patient. A nurse interviewed every patient.

This phenomenon of scope ambiguity is standardly modeled along the pattern of predicate logic; and it has been assumed to be uncontroversial that the formal reconstruction of scope must be deWned as a general process. Furthermore, in the face of the apparent function of wh expressions as propositional operators taking scope over the clause which they immediately precede: (6)

Who do you think will get the job?

It is also assumed that, just as wh expressions move overtly to a position from which their supposed function as a variable-binding operator is directly in evidence, so quantiWed NPs move, albeit covertly, to a comparable position in the structure projected from the sentence sequence. According to this syntactic modeling, (7) is assigned a structure (70 ) at a level of logical form: (7) (70 )

I think a friend of mine will get the job [ [A friend of mine]i [I think ei will get the job] ]

Pursuing this parallel between wh and quantiWed expressions within such a movement account, it is assumed that the ability of a quantiWed expression to take broader scope than is indicated by its position in the string is at least as free as the process of overt ‘Move a’ in establishing long-distance dependency eVects.4 From all perspectives within which quantiWer scoping is taken to be a general process, there are a number of puzzles. First, not all pairs of quantiWers allow reversed scope interpretations: (8) (9) (10)

A student was meeting every visiting dignitary at the station. Most anarchists tolerate few politicians. Almost no student has read every book by Chomsky.

Though (8) allows an interpretation ‘For every visiting dignitary at the station, there was a student meeting them’, neither (9) nor (10) allows such a reversed scope interpretation. Secondly, of those quantiWers which allow reversed scope interpretations, only indeWnites freely allow interpretations in which they are interpreted as taking scope wider than the clause in which they are contained. Thus (11) can have the interpret4 This assumption of parallelism between construal of long-distance dependency eVects and quantiWer scoping not determined by position, though not uniformly held, has been extremely inXuential. In May 1985, and all subsequent analyses which involve quantifer raising, it was explicitly assumed that the processes of establishing structures in which quantiWer scope was explicitly represented involved the same rule of Move a. This parallel is sustained in paradigms without movement. In Morrill (1994), relative to categorial assumptions, long-distance dependency and quantiWer binding are taken, equally, to involve the construction of an additional assumption, duly discharged at the higher point in the tree at which combination with the wh or quantiWed expression is to take place.

Kempson & Meyer-Viol, Scope Choice

563

ation in which most particularly good scripts is interpreted as taking wide scope with respect to two examiners, but it cannot be interpreted as meaning that for some large proportion of scripts it is the case that for each such script everyone agreed that two examiners had marked it: (11)

Everyone agreed that two examiners marked most particularly good scripts.

To the contrary, (12) has just such an interpretation, allowing the interpretation in which for each of two particularly good scripts everyone agrees that most examiners marked that script: (12)

Everyone agreed that most examiners marked two particularly good scripts.

It is clear that this wide-scoping eVect of indeWnite expressions cannot be modeled as the very same process as (the analogue of ) ‘Move a’, since it is not sensitive to the constraints to which long-distance dependency is subject, such as construal across a relative-clause boundary: (13) (14)

Which professor do few students succeed at the problemsets which are assigned by? Few students succeed at the problemsets which are assigned by one professor.

Thus (14) allows an interpretation in which ‘there is one professor for whom few students can succeed at his problemsets’, despite the fact that construal of the wh expression as in this position is not possible. Indeed, it appears that the wide scope taking potential of indeWnite quantiWers is not sensitive to any of the familiar syntactic restrictions (falling under the general label of subjacency).5 (15)–(18) which contain structures out of which a wh expression in such a position cannot be moved, all allow interpretations in which the indeWnite can be interpreted independently of that structure: (15) (16) (17) (18)

Two of my colleagues wrote an article which a friend of mine complained about. John is planning to have several parties in which we all celebrate the work of two Dutch colleagues. John went to see a Wlm and in it many friends of mine were portrayed as idiots. John was visiting his mother while two painters I know were stripping his house.

5 It might be argued that subjacency is not an LF phenomenon, so the question of whether the wide-scope potential of indeWnites is sensitive to such restrictions is irrelevant; but the current stance that co-called movement at LF is not subject to subjacency arose as a response to the debate over issues such as freedom of quantiWer scope as displayed for example by indeWnites. See Reinhart 1997.

564

V. Anaphoric Pronouns and Dynamic Semantics

It has been argued (Ruys 1992) that, despite apparent violations of island-sensitivity for scope eVects with quantiWed noun-phrases, there is evidence of sensitivity to structural restrictions displayed by quantiWer scoping, the cited evidence being that quantiWcation over set membership associated with a plural noun-phrase is sensitive to such island restrictions, even though the existential quantiWcation representing the set denoted by that NP is not. The evidence in this connection is: (19)

If three of our employees take maternity leave, we shall be very shortstaVed.

where it is observed that the only natural interpretation is one that for some set of employees, if each member of that set take maternity leave, then we shall be very shortstaVed. The conclusion by Ruys (1992) is that the representation of quantiWcation over members of a set denoted by a plural NP (called its ‘Distributivity Operator’) has to be restricted by familiar structural restrictions, and hence it is a syntactic phenomenon. However, there are interpretations of indeWnites inside the antecedent clause of a conditional sentence in which not only the existence implication of the set denoted by the plural indeWnite but also universal quantiWcation over its members are independent of the conditional (see also the data of Abusch 1994): (20) (21)

(22) (23)

(24)

If two of my sons ever oVer to help me, I buy them a drink in celebration. On the one occasion on which they both did, I burst into tears of surprise. If two of my sons ever oVer to help me, I buy them a drink in celebration; but fortunately the third is as considerate as the others are not, and never fails to help. If three of my friends get a Wrst, I will be surprised; and if they all do, I will be amazed. Books which three professors had omitted from their reading lists were out of stock when the School eventually ordered them, so the reading lists were very incomplete this year. Some students who professors had singled out for criticism in class were later accused of cheating.

The relevant interpretations follow the pattern of (200 ) as the possible interpretation of (20): (200 )

9Son2 , 8x, x 2 Son2 , (x oVer to help me) ! (I buy x a drink)

If, adopting the movement metaphor as in the proposed analysis of Ruys 1992, the distributivity operator is restricted to disallow its movement out of the antecedent of the conditional, (20) should imply that there is only one occasion on which I bought my sons a drink in celebration of their helping me, and it was on that occasion that I burst into tears. To the contrary, however, one natural interpretation of (200 ) is an assertion that each of the unhelpful sons gets a drink should he elect to help me. Indeed, as the follow-up setting makes plain, it is only this interpretation that

Kempson & Meyer-Viol, Scope Choice

565

is intended. The interpretation of (22) as involving an assertion about three of my friends should be one in which the assertion of the second conditional if they all do would be nothing more than a repetition of something already asserted, which it is not. And as (23)–(24) show, plural indeWnites seem to freely license such interpretations. One attempt to address the puzzling potential for interpretation of indeWnites independent of their environment is to invoke lexical ambiguity, analyzing the indeWnite as corresponding to two lexical items, positing an additional ‘speciWc indeWnite’ that is deWned as quasi-referential and not a scope-bearing term at all. Such an indeWnite is said to denote a particular, albeit unnamed individual which the speaker has in mind (Fodor and Sag 1982; and many thereafter; see also Szabolcsi 1997).6 This form of indeWnite is in addition to a lexical speciWcation of the indeWnite as existentially quantiWed. Such an analysis has the advantage of leaving undisturbed a generalized quantiWer analysis of a subclass of indeWnites with no need of a general process of quantifer-storage. However, as observed by a number of people over the years (Farkas 1981, 1997; Cormack and Kempson 1990; Abusch 1994; Winter 1997; Reinhart 1997), such an ambiguity account fails to capture cases in which the indeWnite can be construed as taking intermediate scope, with scope over the clause or noun-phrase within which the indeWnite is contained but nevertheless interpreted as within the scope of some preceding quantiWer: (25) (26)

Every professor decided three students should evaluate a recent MIT thesis. Each student has to come up with three arguments that show that some condition proposed by Chomsky is wrong.

Each of these examples allows an interpretation in which the indeWnite in the embedded clause can be understood as taking broader scope than the nearer of the two quantiWed expressions preceding it but narrower scope than the subject quantiWer. So for these cases, in any event, indeWnite expressions must apparently be analysed as expressions that function as quantiWers taking scope over some arbitrarily larger domain. A further, apparently independent, puzzle is presented by the fact that this phenomenon is subject to cross-linguistic variation, and many informants in other languages robustly deny the variation which most English informants provide for (5) and (8):7 (27)

mei ge ren mai le yi ben shu every CL man bought ASP one CL book Every man bought one book [ambiguous].

6 Equivalently, one might analyse this phenomenon as one in which the indeWnite on such an interpretation takes wider scope than any other quantiWed expression in the string. 7 There is variation between English speakers also over the degree of sensitivity to linear order with which they evaluate interpretations.

566

V. Anaphoric Pronouns and Dynamic Semantics (28)

yi ge ren mai le mei ben shu one CL man bought ASP every CL book One man bought every book [unambiguous].

In particular in Chinese, though (27) is reported to be ambiguous, as indeed an ambiguity analysis of indeWnites would allow, (28) is said to have only a single interpretation, with no interpretation of the indeWnite as taking narrower scope than the following universal mei. For such speakers, then, inverted scope interpretations are possible with indeWnites following some other quantiWer, but not in the reverse order, a phenomenon which the ambiguity account of indeWnites does not provide a means of addressing. One property of indeWnites has so far gone largely unnoticed. There is a parallelism between indeWnite scope eVects and pronoun construal, which is displayed in all languages, despite cross-linguistic variation as to which quantiWer-scope interpretations the indeWnite makes available. Whatever role linearity considerations play in a language in determining how a pronoun is understood will also determine relative scope of indeWnites in that language. Chinese, for example, is a language in which, for many speakers, pronominal construal is strictly determined by linear order: (29) xiang-xin mary chong-bai ta de mei-yi-ge ren dou cuo-le manj all mistaken believe Mary admires himi REL every Everyonej who believes Mary admires himi is mistaken [unambiguous]. (30)

hushi

ta

muqin

de

mei

ge

ren

dou mei

kao

jige.

ignored

hei

mother

REL

every

CLASSIFIER

manj

all

test

pass

not

Everyonej who ignored hisi mother failed the test.

In (29)–(30) the pronoun ta cannot be understood as dependent on the following quantiWed expressions, even though with reversed order of head and restrictive relative (as Chinese allows), there is an interpretation in which the pronominal is interpreted as bound by the quantiWer: (31) mei ge hushi ta juqin de ren every classiWer ignored he mother REL man Everyonei who ignored hisi=j mother failed the test. (32) mei-yi-ge xiang-xin Mary chong-bai ta de everyone believe Mary admires him REL Everyonei who believes Mary admires himi=j is mistaken.

dou meikao all not ren dou man all

jige test pass

cuo-le mistaken

These facts go hand in hand with (27)–(28). In (28), choice of dependency of the indeWnite cannot be made relative to some quantiWed expression succeeding it, but if the quantiWed expression precedes as in (27), then the indeWnite may be interpreted as being dependent on that quantiWer.8 8 Not quite all our informants reported the asymmetry between (28) and (27), but, commensurate with the assertion of parallelism between indeWnites and pronominal construal, if they reported that (28) allowed a

Kempson & Meyer-Viol, Scope Choice

567

The parallelism between anaphora and indeWnites is sustained even in languages where the correlation between word order and scope is not deterministic. For example, German is a language in which linearity eVects in quantiWer construal have been observed, but these eVects are not tight enough to warrant encoding in the grammatical system. The phenomenon arises in German subordinate clauses, where all nounphrases precede the verb.9 In subordinate structures, an indeWnite expression preceding some quantifying adverbial is generally not interpreted as within the scope of the following adverbial, whereas indeWnites that follow the adverbial are freely interpretable as taking either narrow or wide scope with respect to that adverbial:10 (33) weil ein Assistentsarzt zweimal einen Patienten von because a doctor twice a patient from gefa¨hrlichen Krankheit geheilt hat. dangerous disease cured because one doctor twice cured a patient of a dangerous disease (34) weil zweimal ein Assistentsarzt einen Patienten von because twice a doctor a patient from gefa¨hrlichen Krankheit geheilt hat. dangerous disease cured because twice a doctor cured a patient of a dangerous disease (35) weil von einer gefa¨hrlichen Krankheit zweimal ein because from one dangerous disease twice a einen Patienten geheilt hat. a patient cured because from a dangerous disease one doctor twice cured a patient

einer a

einer a

Assistentsarzt doctor

So in (33) ein Assistentsarzt is interpreted as taking wide scope with respect to zweimal, whereas both einen Patienten and einer gefa¨hrlichen Krankheit freely allow interpretations with wide or narrow scope with respect to zweimal. In (34) all three indeWnites can be interpreted in both ways, and in (35) it is einer gefa¨hrlichen Krankheit that is restricted to take wide scope with respect to the adverbial. However, the restriction does not appear to be structurally deWnable, because if two indeWnites precede the adverbial in such constructions, the second may take narrow scope with respect to the adverbial which immediately follows, as in (36), which allows an interpretation in which one doctor twice cured a patient of some dangerous disease (diVerent patient and diVerent disease on each occasion): reversed scope interpretation, then they also invariably reported that (29)–(30) allowed interpretations in which the pronoun is construed as a bound variable. All such speakers report the possible inXuence of English on their judgements. This restriction on the interpretation of indeWnites as ambiguous only if they follow some other quantiWed term is very widespread, though it is not often expressed in the terms suggested here (see Hoji 1986 for a report of similar data in Japanese). 9 In main clauses, where the V-2 eVect ensures that at least one expression is isolated as preceding the verb, the eVects of linearity are less clear-cut. 10 These data are, we believe, due to Hans Kamp.

568

V. Anaphoric Pronouns and Dynamic Semantics (36) weil ein because a gefa¨hrlichen dangerous

Assistentsarzt einen Patienten doctor a patient Krankheit geheilt hat. disease cured

zweimal von einer twice from a

In the same circumstances, anaphora resolution also allows an interpretation of the pronominal ihren in (37) as bound by the quantiWed expression that follows: (37)

weil

ein

because

one doctor

Assistentsarzt

ihren Krankheitsverlauf

zwei Patienten

zeigte

their

two

showed

disease-progress

patients

because one doctor showed two patients their hospital record

The claimed parallelism between anaphora and indeWnites is not to deny the existence of idiosyncratic restrictions. As is well-known, pronoun resolution is subject to a locality restriction, which indeWnite construal is not. Conversely, there are restrictions on indeWnite construal that do not apply to anaphoric expressions. An indeWnite can only stand in a scope relation to other expressions in the string: no such restriction applies to anaphoric expressions (which freely allow indexical construal). There is, furthermore, lexical variation between types of anaphoric expression, and so, equally, there may be varying lexical constraints on indeWnites, both within and across languages. Nevertheless, the parallelism between anaphora and indeWnites is striking. On the assumption that pronoun construal is correctly analysed in terms of a choice mechanism determining what term provides the antecedent for Wxing its interpretation, one might seek to model the indeWnites in terms of a similar choice mechanism to determine what other term the indeWnite should be dependent on.

Scope Variation as an Anaphoric Choice To give substance to this parallelism, we propose that all quantiWers are deWned to provide only a partial speciWcation of their content, that constraints on scope dependency are deWned separately from the general process of projecting logical forms for sentences, and that these together determine the assignment of interpretation to a string. The mapping from such lexical speciWcation onto interpretation is part of a process of incrementally establishing a logical form on a left–right basis following the dynamics of parsing, in which lexical speciWcations characteristically under-determine their content, which is established through processes of structural and pragmatic enrichment that yield logical forms. Anaphoric expressions provide one central example, projecting metavariables, Ui , Uj . . . , which act as place-holding devices that have to be substituted by some concrete term of type e during the construction process. The scope constraints associated with quantifying expressions are another. These are made available from lexical items as they are incrementally processed, and collected together at the end of the interpretation process to determine the interpretation of the whole. This enables the parallelism between indeWnites and pronouns to be straightforwardly expressed. IndeWnites are deWned as projecting a scope restriction that the term which

Kempson & Meyer-Viol, Scope Choice

569

an indeWnite introduces must be interpreted as taking narrow scope with respect to some other, yet unspeciWed, term. This will be represented by an expression of the form Ui < x where Ui is a metavariable whose value has to be chosen at some later point. The value of Ui , as in anaphora resolution, is established in context (unlike pronouns, such a value must be another term occurring in the interpretation of the string). This gives us a basis for characterizing the arbitrarily wide scope properties of indeWnites, if we make the assumption that the index of evaluation is a term in the logical form of the sentence, represented by Si : c. Here Si denotes the index of evaluation for c. Now, Si can be a term that replaces a metavariable. The relatively unrestricted range of scope eVects for indeWnites follows then immediately. If the value selected for the Wrst argument of the scope relation is some variable associated with a determiner noun sequence, we get narrow scope interpretations; if it is chosen to be Si , then the indeWnite will be construed as taking wide scope relative to the other quantiWed expressions in the sentence. In (38), for example, there happens to be only one possible choice, that the indeWnite is interpreted depending on Si , the term denoting the index of evaluation: (38)

A student fainted.

But in cases such as (39), (12), and (14)–(18), with more than one clause and/or more than one quantiWed expression, there are several possibilities: (39)

Two professors insisted that each student review a recent paper by Chomsky.

The advantage of this view is that it directly classiWes pronoun antecedent construal and choice of scope construal for indeWnites together, as involving a pragmatic choice.

Scope reversal as a delayed scope choice Support for this anaphora-based approach to modeling indeWnite NPs comes from phenomena that seem to conXict directly with the assumption that linear order of elements in the string underlies restrictions on quantiWer construal. Not only can indeWnites take wide scope over expressions that precede them in the string, they can, as we have seen, take narrow scope with respect to some following quantiWed expression, for example, taking narrow scope relative to every patient in (40): (40) (41)

A nurse visited every patient. Two examiners marked most scripts.

If the choice process is, as claimed, a free pragmatic one, then it might seem surprising that this phenomenon is subject to language variation. In particular, Chinese, as indicated by (28), does not allow any such ambiguity. Furthermore, we cannot simply

570

V. Anaphoric Pronouns and Dynamic Semantics

conclude that English allows Xexibility as to whether or not the scope of all indeWnites is Wxed relative to linear order, on the strength of the construal of indeWnites in subject position. There are other positions in which an indeWnite cannot be interpreted relative to some following quantiWed expression. For example, an indeWnite preceding a quantiWed expression in a double object construction cannot be interpreted as taking narrow scope with respect to the following expression, though an indeWnite following some such quantiWed expression certainly may be interpreted with respect to that preceding expression: (42) (43)

John showed a student every problem case. John showed every student a problem case.

The observation so far, then, is the puzzle that, though English appears to follow the general cross-language pattern that the scope of indeWnites is Wxed relative to what precedes it, this pattern is suspended in certain structures, one of which is when the indeWnite occurs in subject position. Recall however that subject position is a position in which expletive pronouns characteristically occur: (44) (45)

It is unlikely that Mary will be at the party. It is unfortunate that Sue is a linguist.

Expletive pronouns are speciWcally itemized pronouns which ensure that their interpretation is determined from some subsequent expression (see Cann et al. 2001). It is not a freely available cataphora process, and is only available for speciWc expressions in speciWc contexts. Suppose now we analyze indeWnites in English as projecting a metavariable as the Wrst argument of their scope relation, with this argument identiWable from some subsequently introduced term. This will license an interpretation in which the indeWnite is construed as having narrow scope with respect to some following expression. Interestingly, such forward-looking construal parallels expletive pronouns. Expletive pronouns in English have an associated locality restriction: the value to be provided for the expletive must be identiWed by some expression occurring within the same clausal domain. This value may be identiWed across a sequence of nonWnite verbal forms; but it may not be identiWed with an expression within some following clause (the ‘rightroof ’ constraint familiar in putative cases of rightward movement—see Bu¨ring and Hartmann 1997): (46) (47)

It is likely to be important that I succeed. It is likely because I am certain that I will fail. 6¼ That I will fail is likely because I am certain.

In just the same way, indeWnites can be construed as dependent on some following quantiWed expression in a post-verbal position in the same clause, and across a sequence of nonWnite verb forms; but they cannot be construed as dependent on a quantiWer in some following adverbial clause:

Kempson & Meyer-Viol, Scope Choice (48) (49) (50)

(51)

571

One student seems to have assisted every professor. Two examiners are required to be seen to mark every script. A student left because every problem was too hard. 6¼ For every problem, there was a student that left because the problem was too hard. A student failed the course because I had mishandled every subject. 6¼ For every subject there was a student failed the course because I had mishandled it.

Putting these expletive cases within the general picture, what we observe is that the restrictions on when identiWcation of a pronominal antecedent may or may not be determined against the linear order also apply to scope dependency. There is the case of double object constructions, for example. In these, just as a pronoun occurring second in a double object construction may but need not be construed as identiWed with a variable introduced by a quantiWed expression in the Wrst of such a pair and be construed as bound by it, so, equally, an indeWnite in such a position may but need not be construed as dependent on the quantiWed expression which is the Wrst of such a pair. Contrarily, just as pronominal dependency cannot be established for the Wrst constituent in a double object construction from the second constitutent in the pair, so also an indeWnite occurring Wrst in a double object construction cannot be construed as being within the scope of a quantiWed expression occurring second in such a pair: (52) (53) (54) (55)

I showed every student his room. I showed every student a data-puzzle I was considering. I showed his mother every student’s room. I showed a student every data-puzzle I was having.

Thus double object constructions appear to disallow either anaphora or scope construal in which the Wrst term of the pair can be interpreted relative to the second. Further parallelisms between anaphora and scope construal for indeWnites are displayed in preposed adjuncts, which allow inverse anaphor-antecedent and inverted scope-dependency construals: (56) (57)

With his hat on his head, every boy left the room. With a frown, every boy left the room.

In the light of these facts, we conclude that in all cases where some non-indeWnite quantiWer appears to take wide scope over a preceding indeWnite, the analysis is not due to any storage or lifting process associated with the non-indeWnite expression: it is, rather, the indeWnite which is interpreted as having scope narrower than the following quantiWer. This has the bonus of explaining the otherwise puzzling diVerence between (8) and (10) repeated here: (8) (10)

A student was meeting every visiting dignitary at the station. Almost no student has read every book by Chomsky.

572

V. Anaphoric Pronouns and Dynamic Semantics

The scope variation available in (8) and but not (10) is not due to some inexplicable property of every displayed in one context only. It is due to the fact that (8) contains an indeWnite in subject position. Thus the apparently problematic counter-example to the proposed anaphoric-like choice of dependency for an indeWnite expression posed by indeWnites in subject position, turns out to fall within the general pattern of anaphora resolution after all.11 Finally, indeWnites divide into subclasses according as the choice of scope they require is or is not subject to locality restrictions. For example, the indeWnite in English when modiWed by the adjective certain imposes a constraint that the phrase it introduces may not be construed as dependent on some expression within the same clausal domain, a principle B style of eVect. So, in (58) a certain historical problem cannot be construed as ranging over as many historical problems as students:12 (58)

Two professors insisted that every student in their class studied a certain historical problem.

It can however be construed distributively with respect to two professors. If we deWne this as a locality restriction precluding scope dependency of the indeWnite on any preceding co-argument of the predicate for which it provides an argument, then we can characterize the exclusion of an interpretation of (58) in which the indeWnite is dependent on its subject while nevertheless allowing interpretations of (59) and (60) which contain occurrences of a certain in a simple clause: (59) (60)

A certain student has failed the exam. I criticized a certain student.

In these latter cases, the indeWnite is independent of any other co-argument if there is one, but takes narrow scope with respect to the index of evaluation. Choice of scope for indeWnites thus seems to follow the patterns displayed by anaphora. To sum up, choice of antecedent for a pronoun and choice of source of dependency for an indeWnite can be made across a structure of arbitrary complexity; antecedent choice and source of dependency for indeWnites cannot be established from the Wrst NP of a double object construction onto the second; prepositional adjuncts provide systematic exception to linearity eVects both for anaphora and for dependency of an indeWnite; if the language allows expletive subject pronouns, then dependency of an expression in subject position on some post-verbal complement will be available; 11 This perspective on indeWnites opens up new possibilities for analysis of cases in which some quantiWer following an indeWnite has been analyzed as involving QR (or quantiWer storage) to characterize its ability to take scope over the preceding indeWnite (see Farkas and Giannakidou 1996): (i) Many people taking the exam thought that each question was hard. 12 Certain is one of the adjectives associated with so-called ‘speciWc’ readings, commonly said to involve a referential use of the indeWnite, to which such examples as these stand as direct counter-examples. Interestingly, speciWc is no more successful in inducing such supposedly referential readings. Indeed speciWc may be associated with interpretations in which the indeWnite in question remains distributive with the narrowest possible scope, so it does not appear, even, to be associated with the type of nonlocality restriction that certain is: (i) Two professors insisted that every student study a speciWc linguistic problem.

Kempson & Meyer-Viol, Scope Choice

573

and, Wnally, both anaphoric and indeWnite expressions may have locality restrictions determining their interpretation. These locality restrictions are not identical. Unlike a pronoun, an indeWnite does not, except with the addition of certain, have an antilocality restriction. And the dependence of indeWnites on what follows is somewhat freer than the specially deWned expletive pronouns. But the general dynamics of interaction between partially speciWed terms and pragmatic choices in the Wxing of interpretation in completed logical forms is shared by both anaphora and indeWnites, both making available backward- and, more restrictedly, forward-looking dependencies. This parallelism between anaphora and quantiWer dependencies is extremely puzzling from the perspective of most accounts of quantiWcation.13 In movement accounts of the quantiWed expression, the correlation between indeWnite and anaphora is at best partially expressible by invoking ambiguity as between referential and quantiWcational uses of the indeWnite, so that parallel eVects can be established between referential uses of the indeWnite and referential uses of pronouns. However, we have already seen that this move is not supported by the evidence.14 In any case, such a move misses the point of the parallelism: it is not at the level of denotational content that the parallelism is to be observed, it is rather in the types of process whereby such content is established. In Xexible categorial grammar (e.g. Hepple 1990; Morrill 1994), it might at Wrst glance seem that parallelism between anaphoric binding and quantiWer construal is captured directly, since all discontinuous dependency is characterized by assumption of some variable and subsequent discharge of that assumption, for long-distance dependency, quantiWers, and pronoun-binding as a bound variable alike. Indeed it is surprising on this view that these phenomena ever diverge. However this uniWcation is close to spurious, for anaphoric dependency is deWned solely to cover the boundvariable construal of pronouns (see Hepple 1990), with no attempt to unify all types of anaphoric dependency; and there is no reason whatever to correlate availability of wide scope interpretations for indeWnites and anaphoric binding in contra-distinction to any other form of binding. Indeed, on all accounts in which indeWnites are deWned to be generalized quantiWers, the correspondence between anaphora and indeWnite scope eVects is at best unexpected. If however we characterize indeWnites as projecting some form of name with an associated constraint on what is otherwise a free scope choice, then we have a basis from which we can explain these phenomena. The lexical speciWcation of an indeWnite will include the projection of an under-speciWed term in order to allow the Wxing of scope relation to be carried out as part of the interpretation process. And it is the analysis at the level of making choices between available terms that correctly reXects the parallelism between quantiWer scope and anaphora. 13 Farkas (1997) also notes the parallelism between anaphora construal and interpretation of indeWnites, but her analysis, deWning indeWnites as variables with functions determining the range of available interpretations, does not bring out the extent of the parallelism between the two, with both requiring an analysis in which their lexically assigned content underdetermines interpretation. 14 Cormack and Kempson (1990) argue that even in cases where the ‘referential’ interpretation of the indeWnite seems clear enough, it is in fact no more than a pragmatic implicature, and not part of the established content projected by the expression itself.

574

V. Anaphoric Pronouns and Dynamic Semantics

3. Dynamic Syntax The suggested account of quantiWed expressions in terms of partial speciWcations and a left–right process of construction may seem impossible to square with familiar syntactic or semantic accounts of the phenomena. What is needed is some general dynamic perspective; and this is what we now sketch. In the dynamic syntax framework, syntactic properties of sentences are expressed in terms of a goal-directed process whereby information is built up on a left–right wordby-word basis. The methodology adopted is to use tree structure to represent not syntactic structure but logical forms as assigned to clauses/sentences; and then to deWne a concept of tree growth that reXects accumulation of the interpretation, using the transitions across partial trees to constitute the basis for syntactic generalizations. The goal is to establish a logical form corresponding to the proposition expressed by the sentence/clause where the logical form is represented as a tree structure whose rootnode is decorated with a propositional formula, and each internal node decorated with some subterm of that formula. The interpretation process starts with the introduction of the rootnode of a tree with a ‘requirement’ (represented by ‘?’) that it be decorated by some propositional type, and proceeds by unfolding and decorating the tree node by node, as information from the words progressively enables a tree to be constructed. For an example of a completed tree, consider the interpretation of John upset Mary (Figure 17.1). This decorated tree presents a propositional formula, each node decorated by a subterm (not the word of the natural language itself ) and semantic Ty(t) PAST : Upset (Mary) (John)

Ty(e t) Upset (Mary)

Ty(e) John

Ty(e Formula-type

Fig. 17.1

(e Upset

t))

Ty(e) Mary

Classification

Ty(t)

=

formulae representing assertions, of 'type t ' (t stands for 'true')

Ty(e)

=

formulae representing individuals of 'type' (e is for 'entity')

=

formulae representing properties of 'type e t'

=

formulae representing relations between pairs of individuals of 'type e (e t) '

Ty(e

t)

ty(e

(e

t))

Kempson & Meyer-Viol, Scope Choice

575

type (whether proposition, predicate, etc.), which indicates how this term will combine with other terms. Following Sperber and Wilson (1986/1995), the starting point is simply the goal to establish some propositional formula as interpretation, represented as ?Ty(t): ? indicates a goal/requirement of the process, ?X on a node being a ‘requirement’ for some formula X . We take the process involved in two examples: (61) (62)

He upset Mary. Mary, he upset.

The structure provided at each step is a pointed partial tree, the ‘pointer’, , identifying which particular node is under development. In processing (61), an early step involves a choice. Given the word he, the hearer has to establish a representation of who is being talked about. All that the pronoun itself provides is a place-holding device. Suppose we provide (63) as the context: (63)

John was unnecessarily oVensive at the party yesterday.

And suppose also this pronoun is construed as subject, then what is established in the processing of the Wrst word is a four-step process (Figure 17.2). That is, in these Wrst steps, we have started out from our goal, expanded it to introduce a pair of nodes with the pointer indicating the subject is to be developed Wrst, decorated that subject node with a variable provided by the lexical speciWcation of the pronoun, and replaced it with the term John (as available in the particular context), leaving the pointer at a node requiring some predicate. In the step which occurs next, the subgoal ?Ty(e ! t) is a trigger for the set of actions provided by the irregular verb upset which contains a temporal speciWcation as part of its lexical speciWcation.15 The result of updating the tree is to provide the term ‘Upset’, a speciWcation that the action to be depicted is in the ‘past’; the only bit of information still needed is a speciWcation of who got upset— hence the remaining subgoal ?Ty(e) (Figure 17.3). The word Mary then gets processed in much the same manner as the word John, providing a value to the object for ‘Upset’ (Figure 17.4). Through regular semantic rules (b-reduction), we build the predicate Upset Mary by combining Upset with Mary, and we combine that with John, then Wnally we establish a complete logical form representing the assertion that this particular John upset this particular Mary ? Ty (t)

? Ty (t)

? Ty (t)

? Ty (t), ? Ty (e),

? Ty (e

t),

U,

? Ty (e

t),

John

? Ty (e

Fig. 17.2 15 The word upset, exceptionally, does not have a morphological reXex of past tense.

t),

576

V. Anaphoric Pronouns and Dynamic Semantics PAST :? Ty (t)

John

? Ty (e

Upset

Fig. 17.3

t),

? Ty (e),

PAST :? Ty (t)

John

? Ty (e

t),

Upset

Fig. 17.4

Mary

PAST : Upset (Mary) (John),

John

Upset (Mary)

Upset

Fig. 17.5

Mary

(Figure 17.5). Notice how the structure gradually builds up: nothing is taken away except requirements, and these go as they are progressively fulWlled. The tree structure displays how the parts are put together; and the result is a structured representation of the interpretation assigned to that sentence in the given context. This same property of progressive tree growth is displayed in the cases of longdistance dependency. (62)

Mary, he upset.

Again the word he poses the problem of not providing information as to who is being talked about, and a choice has to be made in the particular context. But there is more to interpretation than just this. In this case, the word Mary is construed as providing a term for the resulting logical form, but how it is to contribute within that structure is not at Wrst identiWed. This under-speciWcation we reXect directly by representing the term as decorating a node which is described as within the overall structure but not yet Wxed (Figure 17.6). With this term, so to speak put on hold, the processing of the remainder of the string is as before. So, again substituting the variable provided by the pronoun with the term John and so identifying the person picked out as John, the partial structure established

Kempson & Meyer-Viol, Scope Choice

577

is (Figure 17.7). And after processing the word upset, we have a structure into which the predicate ‘Upset’ has been inserted, and an object node introduced (Figure 17.8). But, now, when the structure is almost complete, the string Wnishes. Indeed, nothing more is needed if the formula Mary in the overall structure is taken to provide the object of the predicate Upset. And reXecting this, we unify the unWxed node with the object node to establish its location in the tree Figure 17.9). Notice the signiWcance of this analysis. The process starts with some weak speciWcation of structure and progressively enriches it. The phenomenon of long-distance dependency, one of the central properties of natural language syntax, is analyzed as just one aspect of the growth of information—information as it comes in may be underspeciWed with respect to some outcome, in this case, under-speciWed with respect to its role in the Wnal structure, an under-speciWcation which is resolved during the construction process.

? Ty (t)

Fig. 17.6

Mary,

? Ty (t)

Fig. 17.7

John

Mary

? Ty (e

t),

? Ty (t)

John

Mary

Fig. 17.8

Upset

? Ty (e),

? Ty (t),

John

Fig. 17.9

? Ty (e

Upset

t)

Mary

578

V. Anaphoric Pronouns and Dynamic Semantics

QuantiWcation With the gradual build-up of trees, the projection of under-speciWed terms for natural language quantiWcation is just one type of under-speciWcation. IndeWnites are deWned as projecting an incomplete (epsilon) term. The task then is to deWne the precise form of underspeciWcation required, the process which updates this speciWcation, and the rules determining scope construal once appropriate choices are made. The particular problems posed by the left–right constraints of a parsing perspective are how to deWne constraints on scope assignment, how these constraints can be collected incrementally, and to formulate the algorithm for determining the Wnal scope. The Wrst problem is that we have to incorporate the information supplied by quantiWed noun-phrases into the tree before their scope relations to other noun-phrases and to modalities can be determined. This is a consequence of the fact that we take natural language strings to supply their information in a sequential order. We therefore want to be able to collect any restrictions on scope between quantiWed expressions as they become available in the course of a parse through the string. To reXect this, our analysis of quantiWcation falls into two parts. First, we analyse quantiWed NPs as structured objects of type e. Secondly, we assign scope relations to these structured objects.16 QuantiWed NPs, like proper names, project values of the formula predicate of type e. These NP representations will be formed using variable-binding term-operators. For instance, the phrase some man will be represented by (, x, Man(x)) This quantiWed NP can be completely described by Wxing four parameters: 1. The Binder (e.g. ) indicates the mode of quantiWcation (i.e. existential). 2. The Variable (e.g. x) indicates the variable being bound by the binder. 3. The Restrictor (e.g. Man (x)) indicates the binding domain of the variable. By their restrictors, objects are introduced in a discourse; and predications on these objects are then constructed from items more standardly seen as falling within the nuclear scope of the determiner. These three features exhaustively specify quantiWed NP denotations in isolation. However, to characterize the interpretation of NPs as occurring within a sentence, a scope statement is required. 4. A Scope statement is a proposition of the form x
Kempson & Meyer-Viol, Scope Choice

579

where x and y are arbitrary variables of type e stating, in this case, that the quantiWer binding x has scope over the quantiWer binding y. Each of these these four features needs to be Wxed, by whatever encoded or pragmatic means, so that the semantic, that is, truth-functional, interpretation of a clause projection will be fully determined. With a given formula value c of type t, c ¼ f((n1 , x1 , c1 ) . . . (nn , xn , cn )) where the (ni , xi , ci ) displayed are the projections of all quantiWed NPs occurring in a clause, each with a unique variable xi , we can associate a strict partial order: B ¼ hB,
580

V. Anaphoric Pronouns and Dynamic Semantics

Having parsed the string which is (64), the decorations of the root node of the resulting tree description may form the set {Fo(Ate((, y, Biscuit( y))) (t, x, Dog(x))), Ty (t), . . . }[ {(S < x, x < y)} That is, the scope relations, however projected, appear as atomic propositions involving the bound variables on the root node. These propositions have been collected in the course of the parse. Because every term will have been assigned a distinct variable, we can identify a term with its variable. The scope propositions can now be used to algorithmically update the formula value, Ate(t, x, Dog(x)) (e, y, Biscuit(y)), into the more familiar shape: {Fo(S: 8x(Dog(x) ! 9y(Biscuit( y) ^ Ate(x, y)))), Ty(t), . . . }: In this update the quantiWer binding x indeed has scope over the quantiWer binding y. So, once a parse through a natural language string has set up such a pair, a complete and well-formed representation has been constructed (see Kempson et al. 2001: ch. 7).

The Dynamics of Building up Interpretations In this formal sketch of quantiWcation, what we have focused on so far is the resulting interpretation and its representation using scope statements. There are, in addition, actions that build up such interpretations. A full account of the formal details of these actions is not necessary here. It will suYce to give an indication of how individual determiners provide distinct forms of input. Lexical speciWcations of words give instructions to update a tree, with each lexical action involving a triple of a condition S and two actions: IF THEN ELSE

all formulae in S hold at the pointed node, do action a1 , do action a2 .

where ‘ELSE’ is commonly an instruction to abort. This is because the action of interpreting a lexical item in a string is not something which is optional. The action is obligatory, and its eVects unavoidable. ReXecting the dynamics of the construction process, these speciWcations characteristically provide considerably more than just a term Wxing the denotational content. They may provide an array of decorations either on the node currently under development, or on some other node. In the case of quantiWed expressions, we assume that, for every determiner–nominal pair, there is an action, adding to the root a scope statement speciWc to the type of determiner. In the case of indeWnite noun-phrases, we assume that the scope action involves a scope statement of the form: Ui < x

Kempson & Meyer-Viol, Scope Choice

581

where x is the variable projected by the indeWnite, Ui is a metavariable whose value remains to be chosen. As in the case of pronouns, such a metavariable has to be replaced by some appropriate term. To express the additional restriction imposed by Chinese, where the replacement for that place-holder has to be a term subsequently introduced in the construction process, we need to add a restriction on the variable x that it take wider scope than any subsequently introduced variable. Additional restrictions on the set of scope statements may also be added by predicates. Since scope statements are added to the rootnode of the tree during the course of the tree construction, it is not only determiners that may have the eVect of adding scope restrictions: other words may do so too. So we can, for example, deWne the predicates that induce double object constructions as imposing a restriction on scope statements concerning the Wrst such object. Once all the scope statements are collected, an algorithm dictates the Wnal interpretation. This representation can take one of two logically equivalent forms, either a predicate logic formula, or a formula of the epsilon calculus in which the term restrictors fully specify the scope relations between the various terms.

4. Summary Rather than pursuing the formal details of such lexical speciWcations and the general algorithm for completing the logical form using the collected scope statements, we prefer to reXect in closing on the general properties of this analysis. Notice that the mode of analysis, with its direct modeling of the dynamics of the incremental process of understanding, is essentially structural. Partial terms are introduced and updated. Determiners introduce binders, nominals introduce variables. Constraints on scope are projected either by determiners or other lexical items. The Wnal scope is derived from the scope statements accumulated during the process of building up a logical form. It is only once the scope statement is determined that the evaluation of the tree can be completed and the full restrictor of the term operators Wxed. As an integral part of this construction process, interpreting the indeWnite article involves substitution, replacing a metavariable with a concrete term. This substitution process is unlike the model-theoretic characterization of a variable as Wxed with respect to some assignment of values to variables. Rather it is a process of choosing what term is to be constructed as input to such model-theoretic evaluation. It is this process of choice which renders indeWnites parallel to anaphoric expressions. In both anaphora and indeWnites, the input determines an intermediate representation, and these representations are completed according to accompanying locality or other constraints. It is essential to the analysis that the speciWcation provided by the determiner is not some function from properties to sets of properties but rather an instruction on the construction process. Direct model-theoretic characterizations of quantiWed expressions in natural language may provide an accurate characterization of the diversity of denota-

582

V. Anaphoric Pronouns and Dynamic Semantics

tional content expressible by quantiWed expressions; they do not address the issue of how such denotational content is established from items which only partially determine their content. In particular, model-theoretic characterizations do not provide a natural vehicle for expressing what is common to some base speciWcation from which the truth-theoretic diversity for an individual expression can be determined.17 In contrast, a perspective in which the process of establishing interpretation is central, and, furthermore, is deWned as a process of structural growth, not merely provides a unitary basis from which to express the diversity of anaphora resolution in an integrated way but also enables the parallelism between anaphora and indeWnite construal to be directly brought out. And so it is that discussion of quantiWer scope turns out to be a defense of representationalism. The parallelism with anaphora cannot be brought out any other way. In advocating a level of representation as the interpretation of the string, it may seem that there is an unwanted multiplicity of levels of representation with, presumably, syntactic and semantic representations in tandem. However, within the dynamic syntax framework this is not so: the only types of representation invoked are those which are partial speciWcations of subterms of some logical form. The concept of syntactic structure independent of semantic representation is abandoned, replaced by the concept of tree growth yielding a logical form as outcome. Wellformedness for a sentence-string is accordingly deWned as the availability of a set of transitions from the starting point (the postulation of a single (root) node with a requirement to achieve Ty(t)) to the required outcome (a tree representation of a logical formula of type t) such that no words remain to be processed, all unWxed tree-relations and metavariables are replaced by Wxed values, and no requirements remain outstanding. It is notable in this connection that the prime evidence set out in the 1960s and 1970s for a level of syntactic representation independent of semantics was that of longdistance dependency. With this phenomenon reconstructed in dynamic syntax through the concept of a growth process leading to a logical form representing interpretation, the major justiWcation for a concept of syntactic structure independent of semantics is undermined, and the vocabulary for describing the progressive and left– right build-up of logical form for a sentence is all that is needed for a formal account of natural language syntax. The detailed justiWcation of this claim is provided in Kempson et al. 2001: here we consider what this variant of representationalism amounts to. According to the dynamic syntax perspective, the grammar of a language induces monotonic development of trees representing unreduced lambda terms. These are the only type of representation invoked; and the syntax of a language is given by deWnitions of the set of representations derivable from a string plus an articulation of the licensed transitions that can be used in establishing the mapping from sequence of words onto such representations. So while the characterization of indeWnites, as one example of a 17 It is notable that DRT, though deWned in model-theoretic terms, with discourse representation structures deWned as partial models, is now invariably taken as an intermediate form of representation with structural properties that may play a role in the interpretation algorithm (see Kamp and van Eijck 1997).

Kempson & Meyer-Viol, Scope Choice

583

solution to a puzzle about natural-language semantics, is explicitly structural, it is nevertheless a conservative solution, for the level of representation corresponding to the interpretation of a string is the only level of structure required. All the rest is the dynamics of getting there from the sequence of words. To put this result back into the vocabulary of representationalism, the syntax of natural languages is characterized solely in terms of the structure of the internal representation system (the Language of Thought in Fodor’s terms) plus the dynamics of progressively establishing such structures from the linear Xow of words.

This page intentionally left blank

PA RT V I

Names and Descriptions

According to John Stuart Mill (1843), proper names do not connote; they do not ‘imply attributes’. Instead, they merely denote; they ‘signify’ the entities they are said to name without attributing any properties to those entities. Thus, in so far as names can be said to have ‘meanings’, the meaning of a name is simply its denotation. In a more contemporary idiom, the ‘semantic value’ of a proper name is nothing more nor less than its referent. Thus, the semantic value of the expression ‘Vienna’ as it occurs in an utterance of (1) is the city itself, Vienna: (1)

Vienna is my favorite city.

This approach to proper names is undeniably intuitive. Pre-theoretically, proper names seem diVerent from other linguistic expressions; they do not seem to have meanings in the way other expressions do. Ask a linguistically competent non-philosopher what ‘favorite’ or ‘city’ means, and you will likely get a response of some sort. ‘Favorite’, the non-philosopher might tell you, means something like the most preferred; ‘city’ means something like a population center larger or more important than a town or a village. Ask the same person what ‘Vienna’ means, and you might well receive a puzzled look. Alternatively, you might be told: ‘It doesn’t mean anything! ‘‘Vienna’’ is just the name of a city in a Austria.’ These intuitive reactions are captured by the Millian view that proper names, in contrast to other linguistic expressions, do not have meanings in the ordinary sense. Both Frege (1892/1970) and Russell (1917, 1919) took issue with the Millian view, claiming in eVect that ordinary proper names do indeed connote. In particular, both philosophers argued that proper names behave semantically like deWnite descriptions. Let us begin with Frege’s view. For Frege, the meaning of a name is no diVerent in kind from that of a deWnite description (or any other natural language expression). For any given proper name, its meaning is its ‘sense’: that which accounts for its cognitive signiWcance, determines its reference, and constitutes its reference in embedded contexts. Consider the following sentences: (2) (3)

Hesperus is Hesperus. Hesperus is Phosphorus.

586

VI. Names and Descriptions (4) (5)

Fred believes that Hesperus is Hesperus. Fred believes that Hesperus is Phosphorus.

Suppose, as is in fact the case, that ‘Hesperus’ and ‘Phosphorus’ are co-referring: both refer to the planet Venus. This means that both (2) and (3) are true. However, whereas (2) is trivial, (3) is informative. How can this be? Frege’s answer: Although co-referring, the names ‘Hesperus’ and ‘Phosphorus’ have diVerent senses. These might be (something like): the brightest heavenly body appearing in the evening sky and the brightest heavenly body appearing in the morning sky, respectively. These diVerent senses account for the diVerence in cognitive signiWcance between the names themselves and between the sentences in which they occur. The names’ senses also provide a ‘criterion’ for reference: a conceptual condition that the reference must somehow ‘satisfy’. In the case of both ‘Hesperus’ and ‘Phosphorus’, the criterion is satisWed by the planet Venus. Now consider (4) and suppose it to be true—which it surely must be if Fred has any beliefs at all about Hesperus. Suppose, with Frege, that the reference of a sentence is its truth-value and that the reference of a sentence is compositionally determined by the references of its parts. If the truth-value of a sentence (its reference) is compositionally determined by the references of its parts, then it would seem (5) must be true if (4) is. But it might not be: Fred might believe that the names ‘Hesperus’ and ‘Phosphorus’ name diVerent heavenly bodies, in which case (4) might be true while (5) is false. How is this to be explained? Frege’s answer was that the divergence in truth-values is the result of the fact that the reference of an expression, when that expression occurs in an embedded context, is its ordinary sense. Because the (ordinary) senses of ‘Hesperus’ and ‘Phosphorus’ are diVerent, so are the references of those expressions as they occur in the embedded contexts provided by (4) and (5). This divergence in reference makes possible a divergence in truth-value between these two sentences. For Russell, viewing names as descriptions meant that names ‘abbreviate’ the descriptions speakers ‘have in mind’ when using them. Thus names, being (in eVect) descriptions, are subject to the theory of descriptions. Suppose that the description associated with ‘Aristotle’ in a particular assertive utterance of the sentence ‘Aristotle was brilliant’ is ‘the pupil of Plato and teacher of Alexander the Great’.1 Then, the proposition expressed is a complex existential one to the eVect that, (6)

There is exactly one individual who is a pupil of Plato and teacher of Alexander the Great and that individual is brilliant.

The arguments in favor of a descriptive (as opposed to referential) approach to proper names parallel the arguments in favor of a descriptive (as opposed to referential) approach to deWnite descriptions. As we have just seen, a descriptive approach to proper names can account for puzzles involving identity statements between coreferring names and propositional attitude attributions. Such an account can also 1 As speculated by Frege (1892/1970).

VI. Names and Descriptions 587 accommodate sentences containing non-referring (or ‘empty’) names, including true negative existentials. It can thus account for the informativeness of (3), the possible divergence in truth-value between (4) and (5), and the meaningfulness of sentences like, (7)

Homer lived in ancient Greece.

For even if the name ‘Homer’does not denote, it still abbreviates some deWnite description. Suppose it abbreviates the description ‘the author of the Iliad and the Odyssey’. The sentence uttered is thus meaningful as it expresses a proposition to the eVect that there is exactly one author of the Iliad and the Odyssey and whoever is an author of the Iliad and the Odyssey lived in ancient Greece. Kripke (1980) argued against the Frege–Russell view that proper names are semantically akin to deWnite descriptions. Although the arguments he oVered were varied, one stands out as especially compelling. Following Salmon (1994),2 we can refer to this argument as the ‘semantic argument’. The argument is directed against the view that names refer by way of associated descriptions, and has no direct bearing on the view that the semantic content of a name is descriptive in nature. Kripke points out that a speaker can succeed in referring to a name’s nominatum even if the description he associates with that name fails to denote the nominatum. Indeed, reference can be successful even if the associated description denotes some individual other than the name’s nominatum. Thus, one refers to Cicero by way of ‘Cicero’ even if the description one associates with the name—‘a Roman orator’—fails to ‘pick out’ a single individual. One refers to Einstein by way of ‘Einstein’ even if the description one associates with the name—‘the inventor of the atomic bomb’—denotes Oppenheimer rather than Einstein. The reference of a proper name must accordingly be secured by means other than associated descriptive content. In place of the Frege–Russell view of nominal reference, Kripke proposed the view that proper names refer via a causal chain of communication linking the user of the name to the episode of the name’s introduction. Generally, a name is introduced (and its reference Wxed) ostensively at a ‘dubbing’ of sorts. However, as Kripke realized, a name can be introduced by description, as when the name is stipulated to refer to whatever is the so-and-so. (See immediately below for details.) Kripke did not supplement his causal theory of reference with a theory of semantic content to rival the Frege–Russell view. He did, however, argue that Frege and Russell were wrong to construe the semantic content of a proper name as conceptual or descriptive in nature. And although Kripke never explicitly advocated a Millian view of semantic content, such a view would seem to be compatible with a causal theory of reference. Nathan Salmon (1986) is perhaps the staunchest supporter of the Millian view, having provided a detailed defense of the view against Fregean worries of the sort mentioned 2 Salmon (1994) provides a clear and detailed account of the central arguments against the Frege/Russell view.

588

VI. Names and Descriptions

above. A number of contemporary philosophers of language have followed Salmon’s lead.3 Despite its plausibility and current popularity, the Millian account of semantic value is not without its problems. One problem concerns so-called descriptive names: proper names that have their reference Wxed by description, rather than ostension. Consider the case of Neptune. In the nineteenth century, prior to the telescopic identiWcation of Neptune, the astronomer Leverrier speculated that there was a planet causing perturbations in the orbit of Uranus. Leverrier introduced the name ‘Neptune’ into the language, stipulating that it was to refer to the planet (if any) causing perturbations in the orbit of Uranus. Call such expressions ‘descriptive names’. Other examples of descriptive names are: ‘Vulcan’, ‘Deep Throat’, and ‘Jack the Ripper’. Such expressions are of theoretical interest because their semantics is perplexing. They look and sound like ordinary proper names, names whose reference is Wxed by ostension. That is why we call them ‘names’. One might thus expect descriptive names to function like ordinary proper names, which arguably function as logically proper names. Nevertheless descriptive names denote like deWnite descriptions: they ‘refer’, or so it appears, via the relation of satisfaction. Thus, one might expect such expressions to function like descriptions, which arguably function as quantiWers. What, then, is the semantic content (or ‘propositional contribution’) of such expressions? In particular, is such content Millian or descriptive? Gareth Evans (1979) suggested that such expressions are semantically equivalent to rigidiWed deWnite descriptions, where the description in question is the description that Wxes the reference of the name. To say that such descriptions are rigid is merely to say that they denote the same individual in every possible world in which that individual exists. Thus, consider the descriptive name ‘Julius’, introduced by Evans to refer the individual who (actually) invented the zipper. Then, ‘Julius’ (as introduced by Evans) refers to the individual who actually invented the zipper in every possible world in which he exists—even in worlds in which he did not invent the zipper. Although a number of philosophers have found Evans’s analysis of descriptive names persuasive,4 others have been convinced by the arguments of Donnellan.5 Donnellan (1979) advocates a Millian view of descriptive names and claims that the intuitions behind accounts like Evans’s are the result of confusing sentences with the propositions they express. Thus, consider an utterance of (8): (8)

Neptune causes perturbations in Uranus’s orbit.

as uttered by Leverrier upon stipulating that ‘Neptune’ is to refer to the (actual) planetary cause of Uranus’s orbital perturbations. The utterance would surely sound 3 See, for instance, Soames (1989a). 4 See Davies (1981), Forbes (1989), and Recanati (1993). Note that Evans did not hold that ordinary proper names are descriptive names. According to Evans (1973), ordinary proper names refer to the causal source of the information associated with them. 5 See Salmon (1987, 1989) and Soames (1998).

VI. Names and Descriptions 589 trivial—something easily explained on the hypothesis that ‘Neptune’ is semantically equivalent to the description that Wxes its reference: ‘the planet causing perturbations in Uranus’s orbit’. But Donnellan argues that this intuition is the result of conXating the sentence—which Leverrier would have known a priori to express a truth—with the (singular) truth expressed. The latter, but not the former, is something knowable only a posteriori. Donnellan thus purports to have undermined Kripke’s alleged examples of contingent a priori truths. What is known a priori to be true is not the contingent proposition expressed by the sentence, but the fact that the proposition expressed by the sentence (whatever it is) is true. Robin Jeshion (2001a) has contended that Donnellan’s argument is Xawed and that the (singular) proposition expressed by Leverrier’s utterance of (8) was indeed knowable by Leverrier a priori. Like Donnellan, she is a Millian about descriptive names; but unlike Donnellan, she believes that singular propositions of the sort expressed by (8) are sometimes knowable by those who lack acquaintance with the named entity. This was the case with Leverrier when he introduced the name ‘Neptune’ into the language, stipulating that it was to refer to the (yet to be identiWed) planetary cause of Uranus’s orbital perturbations. Leverrier was able to grasp the singular proposition expressed by (8) despite a lack of acquaintance with the planet Neptune. In her contribution to this volume, Jeshion argues that Evans-inspired descriptive accounts of descriptive names are mistaken. In particular, she argues that such views are not compatible with any plausible explanation of why descriptive names are introduced into the language in the Wrst place. In defending this view, Jeshion considers and rejects half a dozen explanations of the introduction of descriptive names that might be oVered on behalf of a descriptive view of such expressions. These explanations appeal to (inter alia) modal discourse, syntactic ambiguity, and abbreviation. It is a mistake (according to Jeshion) to suppose that descriptive names are introduced to allow for modal thought and discourse. We could engage in such thought and discourse with the assistance of rigidiWed descriptions, descriptions of the form ‘the actual F ’. It is a mistake to suppose that such expressions are introduced to avoid ambiguity due to wide and narrow scope readings of deWnite descriptions. Such ambiguity arises in using ordinary deWnite descriptions, and yet we do not introduce descriptive names to avoid ambiguity in such cases. And it is a mistake to suppose that such expressions are introduced as convenient ‘shorthand’ devices for the descriptions that Wx their reference. If this were the case, why would stipulators not introduce the shortest name possible—why not ‘Nep’ instead of ‘Neptune’? Jeshion contends further that a Millian analysis of descriptive names coheres with the most plausible explanation for the introduction of descriptive names: to enable speakers to think of the nominatum in a psychologically neutral way. An interesting (if surprising) consequence of her view is that it allows for the possibility of de re thoughts about objects/individuals with whom one does not (and in some cases cannot) have an acquaintance-based relation.

590

VI. Names and Descriptions

Marga Reimer is more sanguine about the possibility of descriptive names functioning as rigidiWed descriptions, and proposes a hybrid account of descriptively introduced names. According to this account, such names start oV as semantically equivalent to the rigidiWed descriptions that Wx their reference, though they may subsequently acquire a Millian content. Reimer argues that they acquire such a content if and when acquaintance-based applications of the name become established. Such applications require either that the speaker be acquainted with the nominatum (qua ‘the F ’), or that she be a member of a Kripkean causal chain originating in such acquaintance-based uses. Consider again the case of Neptune. At the time of its introduction into the language, the name functioned semantically as the rigidiWed description that Wxed its reference. Sometime after the telescopic identiWcation of Neptune, acquaintance-based applications of the name became established, at which time the name functioned semantically as a name. Reimer claims that the proposed account is the only account that does justice to all of the intuitive data: the data that support Millian accounts of descriptive names, as well as the data that support descriptive accounts of such expressions. In developing and defending the proposed view of descriptive names, Reimer draws attention to some interesting facts concerning both natural kind terms and empty names, facts that can be accommodated by extending some of the ideas underpinning her hybrid account of descriptive names. In particular, she notes that empty names are plausibly treated as descriptive in content—as are natural kind terms introduced into the language prior to the (perceptually-based) identiWcation of the kind in question. Only when the kind in question is perceptually identiWed can acquaintance-based applications of the kind term become established, thereby allowing for the acquisition of Millian (rather than descriptive) content.

18 Descriptive Descriptive Names Robin Jeshion

1. Ostensive Names and Descriptive Names On June 5, 1599, in Sevilla, a baby boy let out his Wrst wail. The next day, he was baptized. His parents, or a priest, proclaim that his name is Diego Vela´zquez. Fifty-six years later, we can imagine Vela´zquez before his newly Wnished fascinating masterpiece declaring that its name is Las Meninas. As we all know, Leverrier isolated discrepancies between the calculated orbits and the measured orbits of the other planets and hypothesized that there was another planet in our solar system whose gravitational force was responsible for the discrepancy. Prior to any telescopic identiWcation of such a planet, Leverrier introduced the name ‘Neptune’ to refer to the planet that was responsible for the perturbations in the orbits of the other planets. In London in 1888, detectives believed that one person was responsible for a series of brutal murders of prostitutes from the Whitechapel district, and introduced the name ‘Jack the Ripper’ to refer to their serial killer. ‘Vela´zquez’ and ‘Las Meninas’ are examples of what I shall call ostensive names. They are names whose reference was Wxed via ostension. ‘Neptune’ and ‘Jack the Ripper’ are examples of what I shall call descriptive names, names whose reference was Wxed exclusively with the use of a deWnite description. I call such names descriptive and ostensive only because of the way in which their reference was Wxed. As I use the terms, nothing about the semantic content or how we think with names contributes to identifying them as ostensive or descriptive.

I presented various drafts of this chapter at the 2001 Barcelona Conference on the Theory of Reference, the Bellingham Summer Philosophy Conference, the University of Arizona, the University of Maryland, Brown University, University of California, Berkeley, University of Vermont, Yale University, and Syracuse University. My thanks to the audiences for challenging questions. Aaron Boyden, Emma Borg, Dave Chalmers, Paul Coppock, Michael Della Rocca, Adam Elga, Mike Harnish, JeV Horty, Jerry Levinson, Michael Morreau, Mark Moyer, Jim Pryor, Franc¸ois Recanati, Houston Smit, Ted Sider, David Sosa, and Jason Stanley asked questions I tried to tackle in later drafts. I am especially grateful to Michael Nelson and Marga Reimer for extended discussions about descriptive names and pointed criticisms of my ideas. Warm thanks to Marga and Anne for inviting me to contribute to this volume.

592

VI. Names and Descriptions

Although it is intensely debated how exactly ostensive names secure their reference, pretty nearly all parties agree that the reference-Wxer stands in some acquaintance relation to the individual to be named and introduces the name as the name for that individual. The priest stands in a direct perceptual acquaintance relation to the baby boy and declares that it—this very individual, picked out demonstratively—shall be named Vela´zquez. With descriptive names, matters are diVerent. Ostension and acquaintance play no role in Wxing the names’ reference. Instead, the reference-Wxer introduces a deWnite description, used attributively, as the means of picking out the individual that the name names. Whoever (or whatever or wherever) satisWes the description ‘the F ’ is the descriptive name’s referent. The detectives lay down that ‘Jack the Ripper’ names whoever it is that murdered these prostitutes.1 It is important to notice that this analysis of descriptive names does not entail that the reference-Wxer is not acquainted with the name’s referent. Maybe Jack the Ripper is the reference-Wxer’s father or neighbor, someone with whom he is well acquainted. It is just that no acquaintance relations with the named object play a role in the reference-Wxing. Throughout this chapter, I will be concerned only with the semantic and cognitive role of descriptive names prior to the identiWcation or alleged identiWcation (if any) of the named object. Of course, it is precisely this use of descriptive names that is of central importance.2 Descriptive names are interesting because, at the pre-theoretical level, they have both a descriptive component and a nondescriptive component. The descriptive component is associated with the avenue by which the name secures its reference. Instead of the use of ostension, the name’s reference is Wxed with a deWnite description. ‘N ’ refers to object O because O satisWes the deWnite description ‘the F ’. This parallels the way deWnite descriptions (used attributively) secure their reference—satisfactionally.3 The nondescriptive component is associated with the fact that a name itself is being introduced into the language. Names characteristically refer to their bearers in a way that is fundamentally diVerent from the way that deWnite descriptions refer to their bearers, though of course the precise nature of the diVerence is hotly debated. Furthermore, 1 Within this chapter, I characterize descriptive names as names whose reference is Wxed exclusively with an attributive use of a deWnite description. There are cases of reference-Wxing in which someone uses a deWnite description, used referentially, to assist in explicitly identifying the individual the reference-Wxer wishes to name. Consider Kripke’s example in which the reference of ‘one meter’ is being Wxed. There are actually two scenarios here—one in which the reference-Wxer uses only the deWnite description, attributively used, as a means of picking out the name’s referent. The other is a case in which the reference-Wxer is standing before the stick, aims to name the stick’s length, and uses the deWnite description referentially to assist in communicating which object ‘one meter’ is to name. Only in the Wrst scenario do we have a descriptive name. The second scenario might not be properly classiWed as either descriptive or ostensive since it is at least possible that one might not be able to demonstratively identify the length of the stick. This shows that descriptive names and ostensive names do not exhaust the class of names. And there are other ways for a name to come to refer to its referent. Cf. n 22. 2 Interesting questions remain concerning any possible changes in their semantic and cognitive status resulting from the establishment of known acquaintance relations. I hope to address these questions elsewhere. 3 The term ‘satisfactionally’ is due to Bach (1987/1994).

Jeshion, Descriptive Descriptive Names

593

there appears to be a fundamental diVerence between the way in which we think with deWnite descriptions (used attributively) and the way in which we think with names.

2. Exotic Birds Descriptive names are usually seen as exotic birds or semantic mutants—phenomena that we should examine in the interest of a complete psycho-semantic theory, but which will have no bearing on and, moreover, can shed no light on a general analysis of names and the interplay between semantics and cognition. Analysis of them is typically subordinated to and constrained by the analyses of the semantics and psycho-semantics of ostensive names. By this I mean that, almost across the board, philosophers of language and mind develop theories of ostensive names that include certain varieties of strict conditions on naming and thinking with names, and then attempt to develop analyses of descriptive names that preserve these conditions. In particular, it is widely (though certainly not universally) held that when we think with ostensive names, we think ‘directly about’ the referent of the name. Our thought about the object is not mediated by a description, and as such is not mediated by concepts. Our thought is ‘direct’ in the sense that the object itself is part of the content of our thought. To use the jargon, by thinking with an ostensive name, we think de re, not de dicto, thoughts about the named object. And, by and large, philosophers of language and mind agree that to think de re thoughts we must stand in some variety of acquaintance relation to the object of our thought.4 The acquaintance condition on de re thought is the holy of holies, and psycho-semantic analyses of descriptive names work around it. Within the literature there are roughly three positions about descriptive names that, in one way or another, are designed to honor the acquaintance condition. One is that it is not possible to introduce descriptive names into the language.5 Another is that descriptive names are semantically and cognitively distinct from ostensive names. This is the view that I shall be addressing in this chapter.6 And a third is that, although descriptive names have the same semantic features of ostensive names (namely that they refer directly to their objects and utterances of sentences containing them express singular propositions), we cannot grasp those singular propositions by mentally tokening the descriptive name-containing sentence.7 4 The acquaintance condition has roots in Russell. Contemporaries on board for an acquaintance condition on de re thought include: Kaplan (1969), Burge (1977), Donnellan (1979), Evans (1982, 1985), Lewis (1983a), Soames (1995), Boer and Lycan (1986), Bach (1987/1994), Salmon (1987), Recanati (1993), Reimer (this volume). A noteworthy exception is Kaplan (1989a). 5 Cf. Kim (1977). My discussion in this chapter about the prevalence of descriptive names in natural language should cast doubt on this position. 6 This type of view was, I believe, originally proposed by Evans (1979). Other advocates include Davies (1981), Forbes (1989), Recanati (1993), Sutton (2001), Reimer (this volume), Stanley (2002). 7 Donnellan (1979), Salmon (1989, 1990), Soames (1998), Bach (1987/1994) all advocate a version of this position. In Jeshion (2001a), I suggest some concerns about the stability of the position.

594

VI. Names and Descriptions

This basic attitude towards subordinating and constraining our analyses of descriptive names to our theories of ostensive names is, I think, partly rooted in two widely shared views about descriptive names. One is that descriptive names are extremely rare in natural language. The other is that we are always free to introduce a descriptive name just by stipulating ‘Let ‘‘N’’ refer to the F ’. The idea of descriptive names’ rarity coupled with an artiWcial means of generating more has encouraged the tendency to perceive them as exotic birds and to accordingly treat them exotically.8 I think that this basic attitude is misguided. In my view, descriptive names should not automatically be taken as having a ‘lesser’ status in theorizing than ostensive names. I hope to promote this more digniWed status by showing that descriptive names are not really terribly rare in natural language; they are perhaps just as common as ostensive names. Furthermore, we are not ‘free’ to generate them artiWcially. Descriptive naming, like ostensive naming, is a bona-Wde speech act and is subject to felicity conditions. By recognizing the basic attitude as rooted in two very simple yet false views, we can begin to appreciate how descriptive names put intense pressure on the widely embraced acquaintance condition and push us toward altering our understanding of de re thought. I will not attempt to directly argue for these points. What I want to do instead is to discuss and, ultimately, criticize a group of important theses about descriptive names, all of which have roots in the writings of Gareth Evans. I will then use this criticism as a springboard for sketching an alternative, much less popular view about descriptive names, whereby we are capable of having de re thought about objects with which we lack acquaintance. In the course of my critical discussion, I will challenge the rarity and free introduction of descriptive names theses, and, I hope, set us on the right track for acknowledging acquaintanceless de re thought.

3. Descriptive Descriptive Names Evans was the Wrst to forcefully develop and promote the following set of theses about the psycho-semantics of descriptive names:9 Suppose the N is a name whose reference was Wxed with an attributive use of the deWnite description ‘the F ’.10 8 Evans (1985) explicitly (and casually) advanced both theses in the following inXuential passage: ‘Very few names which naturally occur in ordinary language can be regarded as descriptive names. Nevertheless, no matter how rare examples may be, it would appear always to be open to create descriptive names by stipulation. For example, we might stipulate: Let us use ‘‘Julius’’ to refer to whoever invented the zip, and, governed by such a stipulation, ‘‘Julius’’ would appear to have the properties of a descriptive name.’ 9 Evans (1985) reveals a commitment to all of them. On the Semantic-DDN thesis, Evans says that descriptive names are ‘Fregean’; on Understanding-DDN, he claims that to understand a descriptive name one need only know that it refers to the F ; he is explicit in advocating Thinking-DDN. 10 From here on in, I’ll usually drop the qualiWer ‘attributive use’, but it is always intended.

Jeshion, Descriptive Descriptive Names

595

Semantics-DDN: N is semantically equivalent to the description ‘the F ’ (or its rigidiWcation) in the sense that N is synonymous with ‘the F ’ (or its rigidiWcation). Understanding-DDN: Understanding N requires recognizing a priori that N is the F (or N is the actual F ).11 Thinking-DDN: In saying or mentally tokening ‘N is G’ and ‘The F is G’ (or ‘The actual F is G’), one must think the same thought; the only thought available to the thinker is descriptive/de dicto. Evans’s inXuence has been widespread. The theses enjoy support from a diverse and distinguished group, including neo-Fregeans, two-dimensionalists, and Millians. Some or all of them have been advanced by Martin Davies, Graeme Forbes, Franc¸ois Recanati, Marga Reimer, and Jason Stanley.12 But these theses are sometimes not clearly distinguished for they do Wt naturally together as an overall position about the psycho-semantics of descriptive names. Although I will usually specify the individual thesis under discussion, I shall sometimes casually refer to the collective as the descriptive descriptive name (DDN) view. (‘Descriptive descriptive’ because, here, descriptive names have semantic descriptive content.) It is important to notice that the theses are all formulated so as to allow some divergence between the reference-Wxing description and the description that gives the semantic content, understanding condition, or content of thoughts involving the name. So, if a stipulator Wxes the reference of a name N with the unrigidiWed deWnite description ‘the F ’, the semantic content of ‘N’ may be that of ‘the actual F ’. Since my basic criticism in large measure cuts across variations in descriptive descriptive name views due to modal matters, the disjunctive formulations are adequate. Two related lines of thinking stand out as motivations for the DDN view. Both emphasize the diVerence in the way descriptive names and ostensive names have their reference Wxed, and suggest that it is therefore natural, and to be expected, that their semantic contents/psycho-semantics diVer as well. The Wrst motivation is semantical, and pinpoints the presence of the deWnite description in the reference-Wxing: because the stipulator Wxes the name’s reference with a description of the object, the semantics of the name should reXect the description’s privileged role. In particular, the name’s content should be the same as that of the reference-Wxing deWnite description (or its rigidiWcation).13 This semantical consideration in turn encourages the theses about 11 For those who shun appeals to the a priori, an alternative formulation is this: Understanding -DDN Understanding N requires presupposing, and recognizing on the basis of meaning alone, that N is the F (or N is the actual F ). 12 Advocates of this view diVer widely in their views about the semantics of ostensive names. Consequently, the way in which their (common) analysis of descriptive names diVers from their respective analyses of ostensive names will diVer. 13 This line is perhaps implicit in Evans (1979: 180): ‘A descriptive name is a name whose reference is Wxed by description. This formulation covers two points. First, a descriptive name is a referring expression; . . . Second, there is a semantical connection between the name and a description; the sense of the name is such that an

596

VI. Names and Descriptions

understanding and thinking with descriptive names. So, understanding the name requires recognizing a priori that N is the F . And so any thought that N is G will be identical to any thought that the F (or the actual F ) is G. The second motivation is epistemic, concerning the absence of acquaintance in the reference-Wxing. Because the name’s reference was Wxed without reliance on the stipulator’s acquaintance with the object named, the only information he (thinks he) possesses about the object is the fact that it is the (actual) F . Lacking further information, any belief or any thought that the stipulator has by mentally tokening ‘N is G’ would be exactly the same as he has by mentally tokening ‘The (actual) F is G’. In particular, the stipulator could not have a de re thought or attitude about the named object. This epistemic motivation is also rooted in Evans’s writing, in the following famous passage: It seems clear that the two sentences [‘Julius is F’ and ‘The inventor of the zip is F’] are epistemically equivalent. . . . I cannot imagine how the belief that Julius is F might be characterized which is not simultaneously a characterization of the belief that the inventor of the zip is F, i.e., that one and only one man invented the zip and he is F. Belief states are individuated by the evidence which gives rise to them, the expectations, behavior, and further beliefs which may be based upon them, and in all these respects the belief states associated with the two sentences are indistinguishable. We do not get ourselves into new belief states by ‘the stroke of a pen’ (in Grice’s phrase)—simply by introducing a name into the language. (Evans 1985: 202)

For those inclined to keep their semantics sensitive to the epistemic considerations, this rationale for Thinking-DDN can support Semantics-DDN. That is, if de re thought about N ’s referent cannot be secured by mentally tokening sentences containing N , it is, for many, natural to take N as having the descriptive content of ‘the F ’ (or ‘the actual F ’).14 And then it is natural to assume that understanding N requires recognizing a priori that N is the (actual) F .

4. Basic Criticism My main concern about the descriptive descriptive name view is this: it fails to account for why the descriptive name was introduced in the Wrst place. Semantic-DDN states that the descriptive name will be semantically equivalent to a certain deWnite description. There is, consequently, no semantic advantage gained by introducing the name. The same goes for Understanding-DDN. One will be competent with the descriptive object is determined to be the referent of the name if and only if it satisWes a certain description. . . . In this way, a descriptive name has a descriptive content.’ It is not clear whether Evans regarded Semantics-DDN, as suggested by the special variety of reference-Wxing, as following from his characterization of descriptive names, or as an assumption that does not need argument. Reimer (this volume) is much more on target. She regards the special variety of reference-Wxing as a rationale for embracing Semantics-DDN. 14 Not everyone would agree, however: e.g. Bach (1987/1994) would maintain otherwise; Donnellan (1979) and Salmon (1987) might also disagree.

Jeshion, Descriptive Descriptive Names

597

name just in case one is competent with a certain deWnite description. And, according to Thinking-DDN, all thoughts and attitudes with the descriptive name are identical to thoughts and attitudes with the corresponding deWnite description. So there is no cognitive advantage secured by introducing the name. The DDN view gives descriptive names the same semantical and cognitive status as a deWnite description. Why, then, introduce the descriptive name? To save breath? We need an analysis of descriptive names that can explain why they are introduced in the Wrst place, and, to be plausible, this explanation must accord with the role that names play in our semantic and cognitive lives. The DDN view does not, and I think cannot, oVer us an adequate account of why the name is introduced, and, consequently, it is not the right analysis of descriptive names. Names are vehicles for speaking and thinking about objects in a ‘direct’, psychologically neutral fashion. This is their semantic and psycho-semantic function. The reason we introduce names for objects is to speak and think about such objects in this direct and psychologically neutral way. To speak of O with name N , we do not express any particular mode of presentation of O. Likewise, to think of O with name N , we do not have to think of O via any particular mode of presentation. This does not preclude that in speaking and thinking about O, one may have a mode of presentation of O. It only precludes that in speaking with a name, there is a speciWc mode of presentation of O that is expressed; and in thinking with a name, there is a particular privileged mode of thinking of O—one that must be thought with to think of O. Kripke brought out this fundamental feature of names with his semantic argument. He suggested that one could be fully competent with a name, say ‘Feynman’, yet entirely lack any famous-deed-identifying information about the name’s referent.15 The best one might have is ‘some famous physicist’. One can think thoughts about Feynman by mentally tokening ‘Feynman’, and can refer to Feynman by uttering sentences containing ‘Feynman’. One fully understands one’s words on such occasions. Since one lacks an identifying description, the semantic argument demonstrates the psychological neutrality of ostensive names like ‘Feynman’. I think that the semantic argument is applicable to descriptive names as well. Imagine that Leverrier gives his parents a sketch of his scientiWc research. He tells them that he is trying to telescopically identify Neptune, a new, as yet undiscovered planet that he believes he has reason to think exists. In describing his work, he does not communicate to them the reference-Wxing description (or its rigidiWcation). They brag to their friends that their boy Leverrier is searching for Neptune. We can imagine that the best description they have is ‘some planet’. Yet, just like in the Feynman case, they appear to be linguistically competent with the name.

15 Kripke (1980) writes as if one might entirely lack any identifying information about the individual. I think this is not implausible, but a causal descriptivist could challenge it. I have softened the conclusion to be drawn from the semantic argument, suggesting that one can be competent with the name yet lack any particular ‘famous-deed’-identifying information.

598

VI. Names and Descriptions

We do not even have to change subjects. We can imagine Leverrier himself, ten years down the road, in a somewhat altered epistemic situation. His scientiWc work has not panned out very well. He never discovers Neptune and starts to run around with the wrong sort of crowd. But in wistful moments, he reminisces about his scientiWc glorydays, bringing back the joy he felt while searching for Neptune. He has entirely forgotten the reference-Wxing description. Like his parent in our other scenario, the best he can oVer is ‘some planet’. Again, I think that there is nothing semantically deWcient in his uses of ‘Neptune’, and nothing deWcient in his linguistic understanding. The ‘Feynman’ case and these ‘Neptune’ cases appear to be in parallel in all relevant respects, and consequently appear to demonstrate the same points. The upshot is this: the semantic argument suggests that descriptive names, like ostensive names, are introduced in order to speak and think about objects in a psychologically neutral fashion. A proponent of the DDN view who thinks the semantic argument shows that ostensive names function to secure semantic and cognitive psychological neutrality needs to identify an asymmetry in our cases to avoid the conclusion that descriptive names have the same function. Likewise, the semantic argument tells against Semantic-DDN and Understanding-DDN. A proponent of the DDN view who is convinced by the ostensive name version of the semantic argument as a case against descriptivism needs to identify an asymmetry to preserve the DDN theses.16 In sum, my case against the DDN view is two-pronged: (1) the DDN view cannot oVer a plausible analysis of why the name is introduced; and (2) the descriptive name version of the semantic argument directly casts doubt on two of the central DDN theses.

5. No Reason Response One response to this line of argument is to contend that, yes, the introduction of the descriptive name aVords no semantic or cognitive advantages, and that there is no 16 I am specially targeting theorists like Reimer and Soames who embrace Millianism about ostensive names and appear to regard psychological neutrality as a, or the, fundamental function of names. It will not do to maintain that the cases get diVerent treatment because in the descriptive names case there is an alternative explanation available. Within the dialectic with the descriptivist about ostensive names, the Millian argues directly from the ‘data’ about normal use of the name to the conclusion that one can be competent with the name while lacking identifying information about the referent. This is exactly the argument in play here about descriptive names. The response about alternative explanations should have no force to the Millian because, after all, the neo-Fregean could advocate descriptivism as an alternative explanation of the phenomena in the ostensive names case. And this will (and should) be unacceptable to the Millian. While the argument above is expressly directed to the Millian about ostensive names, there is nevertheless a corresponding challenge to other DDN theorists, like (e.g.) Forbes, who uphold a neo-Fregean position about the semantics of ostensive names. Such neo-Fregeans allow that one may be competent with ‘Feynman’ while lacking a particular privileged mode of presentation of the referent. Yet in upholding DDN, they maintain that one must have a particular mode of presentation of ‘Neptune’—that given by the reference-Wxing description. The challenge is, then, simply to identify an asymmetry in the ostensive and descriptive names versions of the semantic argument.

Jeshion, Descriptive Descriptive Names

599

special reason associated with such advantages for introducing the name. Although descriptive names have exactly the same semantic and cognitive function as the deWnite descriptions with which they are synonymous, this is not a problem. After all, our language is replete with distinct linguistic items that overlap in their linguistic function. Such redundancies do not need to be explained away. So, in fact, we do not need any reason at all for introducing a descriptive name into the language. It is just something that we are always free to do simply by saying or stipulating that N is to refer to the F . To bolster the point, many would reach for Evans’s ‘Julius’ case as an example of free introduction, and, as well, to blunt the force of the semantic argument, for it seems that one is not competent with ‘Julius’ if one does not recognize a priori that Julius is the inventor of the zip. In reply, I want to Wrst acknowledge that languages do contain linguistic items that overlap in their linguistic function. (Consider, for example, pronouns, referential uses of deWnite descriptions, and demonstratives.17) But this, of course, does not itself entail that descriptive names have the same semantic and cognitive function as deWnite descriptions. For the No Reason Response to work, it must be independently plausible that descriptive names are introduced into the language for no particular reason at all, just by virtue of an individual’s linguistic act of saying ‘let N refer to the F ’. This is what I question. Let us reXect on Evans’s Julius example. Now, I don’t know about you, but for me, the Julius example rings hollow. I seriously doubt that Evans introduced any name into our language, or even into his own idiolect, with this ‘stipulation’. I do not have a theoretically neutral argument to establish that cases like Evans’s ‘Julius’ are impossible. But I think there is much to be said to at least make one skeptical of this thesis upon which it rests: ‘Free Introduction’ of Descriptive Names—there are no conditions on descriptive naming apart from the linguistic act of saying/stipulating: let ‘N’ refer to the F. And I will put forward a sketch of an associated theory about introducing names that should be congenial to many that embrace DDN theses, yet which is incompatible with the view that we do not need any reason for introducing descriptive names into the language. Why didn’t Evans bring a name, ‘Julius’, into the language, or even into his own idiolect, when he stipulated that it is to refer to the inventor of the zip? I maintain that he did not do so because he failed to satisfy certain general conditions on introducing names into the language. Naming is a bona-Wde speech act, a declarative illocutionary act. Like other types of illocutionary acts, including assertions, commissives, directives, and expressives, it has associated propositional content, preparatory, and sincerity conditions. These conditions rule out the possibility that names are introduced just by saying ‘Let N refer to the F .’ 17 To say that these types of expressions overlap in their linguistic function is not to say that they have identical linguistic functions or identical semantic analyses.

600

VI. Names and Descriptions

The most important conditions, and those most relevant to our concerns, are those most closely associated with the illocutionary point of the act, and which concern the speaker’s intentions. Consider, for example, another declarative illocutionary act, granting permission. If I grant you permission to take a cookie from the cookie jar, I must intend for my utterance of ‘Sure, take a cookie’ to give you permission to take the cookie. My intention is tied to the essential function of granting permission—to bring into existence a license to do something. The reason why I make my declaration is to bring into existence your cookie-taking-permit.18 Like permission granting, some conditions on naming exhibit the essential function of naming—to speak and think about the named object in a psychologically neutral fashion. More precisely, If agent S aims to introduce a name N into her idiolect by Wxing its referent, S succeeds in doing so only if the following condition obtains: Psychological Neutrality (Ostension) S introduces N for object O because S aims to think about and speak about O by mentally tokening ‘N ’, without necessarily thinking about O via any particular mode of presentation. Psychological Neutrality (Description) S introduces N for the F because S aims to think and speak about the F by mentally tokening N , without necessarily thinking about the F via any particular mode of presentation. Though I have oVered two formulations of Psychological Neutrality, one for ostensive names and one for descriptive names, they both capture the same fundamental idea that obtains for all names.19 The condition states that to introduce a name an agent must have a reason for doing so, one that manifests the function of names as vehicles for speaking and thinking about objects directly, simply by physically or mentally tokening the name. Agents who bring names into the language do so because they wish to use the name to serve its function. They introduce it because they aim to speak and think of the named object in this psychologically neutral fashion, absent any speciWc particular mode of presentation. By introducing the name, an agent thereby makes available a common means of speaking and thinking about the object, and subsequently terminates the special standing of any previously privileged modes of presentations of the object. The namer need not, and usually does not, possess this theoretical analysis of the semantic and cognitive function of names as her own—her internal—reasons for 18 Classical discussions of speech act theory are found in Austin (1962), Searle (1969). For a more contemporary analysis, cf. Vanderveken (1990). 19 ‘Psychological Neutrality’ is a term that I borrow from Recanati (1993) who has skillfully developed the point that the function of names is to enable speaking and thinking about objects in a psychologically neutral fashion. But since he embraces the DDN view, he would probably not approve of my turning this into a condition on descriptive naming.

Jeshion, Descriptive Descriptive Names

601

naming the object. The baptizer might have internal religious reasons for naming baby Vela´zquez. And maybe such an internal reason constitutes additional reason for bringing the name into the language. But I think that, nevertheless, the account I oVered concerning the function of names constitutes a basic self-suYcient (external) reason why any namer introduces a name into the language, even if it is never recognized as such. There are other conditions on or at least loose principles governing naming. One that has been discussed by speech-act theorists and that is pertinent here concerns one’s social standing—whether one has the relevant relation to the named object to be the one to name it. For example, you are not in position to name my son, and I am not in position to christen your yacht. Even if at the time of his birth you screamed at the top of your lungs, ‘I name him ‘‘Lester’’!’, you will not have done so. We have the following: Social Standing If agent S aims to introduce a name N into her idiolect by Wxing its referent, S succeeds in doing so only if S possesses the relevant social standing to be O’s namer. The type of social standing needed and the strictness with which the condition is adhered to will depend upon the type of object named. In the case of ostensive naming of human beings, the social relation is normally very speciWc and it is fairly strictly upheld. Another principle governing naming manifests the function of names to serve as a single common syntactic item for thinking and communicating about the named individual. Single Tagging Fix the reference of a term only if, so far as you know, the named object is not already named, or, so far as you know, you do not possess another name for the named object and, presently, lack the resources for discovering the object’s name.20 Single Tagging admits of exceptions—consider nicknames and pennames. I believe that these exceptions actually contribute to establishing the condition. Still, the basic rule stands in need of some modiWcation. Certainly there can be linguistic communities in which everyone has two names (or maybe three). If you’ve read some Tolstoy or Dostoevsky, you know. But there needs to be some limit here—if not to just one name, something close to that. (If you’ve read some Tolstoy or Dostoevsky, you know!) 20 The formulation given here for Single Tagging diVers from that of Jeshion (2001b), where I made a mistake in oVering only the Wrst disjunct of the present formulation as the whole analysis. But this will not do. Consider ‘Jack the Ripper’. Detectives had every reason to think that the murderer of victims X , Y , and Z had a name. Or consider the circumstances of babies orphaned at, say, one year. Adoption agents have reason to think that such babies were named, but may rationally think they lack the resources to discover it. The second disjunct aims to eradicate this diYculty.

602

VI. Names and Descriptions

There are other conditions on naming.21 But reXection on these three should serve to at least encourage or deepen skepticism about the ‘anything goes’ view expressed by our free introduction thesis, and about whether Evans introduced ‘Julius’ into the language as a name for the inventor of the zip. Evans and those who appeal to the ‘Julius’ case solely in discussions of descriptive names do not satisfy Psychological Neutrality (Description)—for they have no intention of using ‘Julius’ to speak and think about the zip inventor. The Social Standing and Single Tagging conditions are also violated, for they lack the relevant social relation and have good reason to think that the inventor of the zip has already been named. While Evans’s famous example may initially seem to combat the force of the semantic argument, it does so only because of its artiWciality. No name has been introduced into the language.22 One possible move here would be to accept these points as applicable to ostensive names, yet deny them for descriptive names. One would then accept Psychological Neutrality (Ostension), yet deny its correlate for descriptive names, maintaining that descriptive names are a special variety of names that are simply not subject to the same conditions as ordinary names. But this rejoinder sounds somewhat desperate at this stage. It is like saying that in addition to promises, there is a special class of promises, promises —a class known only to philosophers—and for which the promiser need not have any intention of fulWlling the act she promises to do. Few will take this suggestion about promises seriously. I think we should take the same negative attitude about the free introduction of descriptive names.23

6. Modal Response Our basic worry challenges the DDN view to supply a reason why the descriptive name is introduced into the language. A second response to it admits that the DDN view 21 I discuss one other condition in Jeshion (2001b). Primacy of Ostension: Fix the reference of a term descriptively just in case, so far as you know, you cannot do it ostensively. There are, I believe, others, and much more needs to be said here. I intend to deepen this brief account of conditions on naming in future writings. 22 Could it be that, in failing to satisfy these conditions, Evans has merely abused the speech act, and not actually misWred? If so, then, as with one who promises insincerely, he may still have brought a name into the language. I think this cannot be the case. The failure of the social standing preparatory condition rules this out, and the failure of the other conditions may as well be suYcient for showing that the act simply misWres. It may be the case that a name is brought into the language if someone takes the reference-Wxer as satisfying the relevant conditions on the speech act. But then the name is not brought into the language by free introduction. I oVer a brief discussion of cases like this in Jeshion (2001b). 23 I would conjecture that many of us have been initially taken in by the free introduction thesis because we are formally trained and are used to introducing terms as abbreviative deWnitions; and we also routinely think about and construct cases in which we create and name heroes of our thought experiments. Perhaps this encourages us to think of ourselves as always ‘free’ to name objects, just by saying so. And maybe we tend to confuse our freedom to abbreviate any concept or name any object with any name with our (apparent) ‘freedom’ to name an object just by making a statement; and, likewise, to confuse what we do in characterizing the subjects in philosophical thought experiments with what happens when we in fact introduce a name into the language.

Jeshion, Descriptive Descriptive Names

603

must explain why the name is introduced, and oVers the following: the descriptive name is introduced to enable us to express modal facts about the named object. As against my claim above that no semantic or cognitive advantages result from introducing the name on the DDN view, the modal response maintains that the capacity to engage in modal discourse and to think modally about the named object is a semantic and cognitive advantage consistent with DDN.24 My initial point in reply is that the very same semantic and cognitive advantages can be secured with the use of the rigidiWed deWnite description ‘the actual F ’. Nothing is gained by introducing the name as opposed to the rigidiWed deWnite description. To this a proponent of the modal response will probably suggest that, although the same capacity for modal talk and thought can be secured with the rigidiWed description and the descriptive name, most agents are not sophisticated enough to explicitly introduce a rigidiWed description into the language. So the descriptive name has a ‘relational property’—being commonly available to ordinary speakers—that distinguishes it from the rigidiWed description, and which, in turn, enables it to carry through its semantic and cognitive function of allowing modal talk and thought. So the advantage is more pragmatic than theoretical, since it is acknowledged that, in theory, the rigidiWed description could perform the same function. Now, this seems to me a fair reply. There is something right about the idea that ordinary speakers do not explicitly express and think with the actuality concept. So I will grant that there is a pragmatic advantage in introducing descriptive names into the language. This brings me to my second point. Producing an explanation was not all that the basic worry demanded. That is, it is not suYcient to simply point up a function of descriptive names in answering the challenge to explain why the name was introduced. We need an explanation that simultaneously comports with the linguistic data and remains consistent with the DDN. This one does not. Is it plausible that for every descriptive name, reference-Wxers introduced the name because they wished to secure modal thought and talk about the named individual? I think that it is not. According to the DDN proponent of the modal response, detectives introduced the descriptive name ‘Jack the Ripper’ because they wished to make possible discussion of whether the murderer would have killed those prostitutes had his father been kinder, or had he been raised a Quaker, or practiced yoga, or drunk camomile tea on the nights of the murders. This (like camomile tea) is hard to swallow. Detectives’ primary reason for introducing the name into the language was because they wished to communicate and think about the murderer in his or her actual circumstances. It is easy to imagine that they had no interest at all in speaking and thinking about the murderer in counterfactual circumstances. At the very least, such an interest is needed for this to be their reason for coining the name. Of course, it is possible, 24 Some comments by Evans suggest that he might have advanced this line. Stanley (2002) is explicit in advancing it as the reason for introducing the name, and tries to bolster his suggestion with remarks by Stalnaker that the function of names is to enable modal discourse. It is not clear whether Stalnaker himself would advocate this response.

604

VI. Names and Descriptions

maybe even probable, that detectives did have some interest in such counterfactual reasoning. My point is just that they need not have had any such interest and so this need not have been their reason for introducing the name. Note: my point is not that the detectives would not themselves isolate this as their own (internal) reason for introducing the descriptive name. This fact is (to my mind) irrelevant to the debate. My point is that the capacity to speak and think modal thoughts about the named object need not have been their reason for introducing the descriptive name. If you are not already dubious, switching from people to planets may help. Think about Leverrier’s introduction of ‘Neptune’. Thanks to Kripke, we have all been schooled to have intentions to think and speak about Neptune in non-actual circumstances. But why must Leverrier have had these intentions? It seems that this would be quite extraordinary on Leverrier’s part, and at least we can imagine him lacking them. In section 11 below, I shall discuss a series of less well-known examples of descriptive names, including names of objects under construction and names of certain mathematical objects like p. It would be very odd if the reason mathematicians introduced ‘p’ into the language was to speak and think about that number in counterfactual circumstances. In any event, reXection on these descriptive names should make one even more suspicious of the modal response. The advocate of the modal response must maintain that, for every descriptive name, the reference-Wxer introduces the name because she intends to secure modal thought and talk about the named individual. I have been urging that this does not jive with the linguistic and psychological phenomena. To fully appreciate the point, it is important not to confuse the view that descriptive names are introduced to secure modal thought and talk with the idea that, whenever a descriptive name is introduced, the referenceWxer secures an ability to think modal thoughts about the named individual. One might suggest that, in virtue of having such an ability, one thereby has the intentions to think modal thoughts about the named individual. I think that the quick move from ability to intentions is misguided. But even if there is an extended sense in which we have such intentions, this still does not save the modal response. For if reference-Wxers must have intentions to think modal thought about the named individual, this is so because they have a more basic intention to think of that individual in a psychologically neutral fashion. And psychological neutrality is inconsistent with the DDN view.

7. Syntactic Ambiguity Response A related response is this: descriptive names are introduced into the language to avoid ambiguity due to wide and narrow scope readings of deWnite descriptions in subject position. But this too is implausible. There are numerous occasions in which we encounter ambiguities due to scope when using a deWnite description in subject position. Yet to avoid such confusions, we do not therewith introduce a name into the

Jeshion, Descriptive Descriptive Names

605

language. We have ordinary, well-known resources for clarifying our intended meaning and avoiding any misinterpretations due to scope ambiguities that may arise. The point generalizes: though there are numerous occasions in which we encounter (general) syntactic or semantic ambiguities, we do not introduce a new word every time we anticipate confusion. We do not need to because we have alternative, linguistically economical resources for avoiding such misunderstandings.

8. Abbreviation Response To my question ‘why introduce the descriptive name?’ a DDN theorist might bite my rhetorical answer—‘To save breath?’—and soberly suggest that the name is in fact introduced to abbreviate a long description. The idea is that Leverrier tires of saying and writing ‘the planet causing perturbations in the orbits of the other planets’ and so introduces ‘Neptune’ not only to save breath, but also to save time, trees, and ink. Indeed, he introduces the name to do everything that ordinary abbreviations do. This conservationist line is charming, but it makes me want to ask—I cannot resist—why did Leverrier choose ‘Neptune’, and not, say, ‘Nep?’25 That would be even more environmentally sound. And does this mean that ‘Johnjacobjingleheimerschmidt’ is ruled out, automatically, as a descriptive name for any description with fewer than seven syllables? That seems unfair. Seriously, this view puts limitations on the amount of phonemes and syntactic units that may be used in a descriptive name. The descriptive name must have less of each than those in the reference-Wxing description. But surely we think that these considerations should not prevent us from introducing, say, ‘Geremiah the Gigolo’ as a descriptive name for Pam’s lover (seven syllables versus three). I think that what I say shortly against the rarity of descriptive names should help further dispel this response, for there are an enormous amount and variety of descriptive names in natural language, some of them, in fact, quite lengthy. Even if the foregoing points do not themselves entirely put this response to rest, we should keep in mind that the semantic argument still needs a rejoinder. A, perhaps the, central idea of our version of the semantic argument is that descriptive names do not function semantically or in thought like abbreviations. One of the central properties of abbreviations that stand for properties or concepts26 is that a competent user of such an 25 Despite appearances, I do realize that Leverrier had a lot of good reasons for choosing ‘Neptune’ over ‘Nep’, e.g. continuing the tradition of choosing gods of Roman mythology as names of planets. But my point is a general one. 26 Competence with abbreviations of names is a diVerent matter entirely. You can be a competent user of the acronym ‘YMCA’ without knowing what the letters stand for. But this cannot help the proponent of the abbreviation response. To successfully avoid our challenge, descriptive names cannot be modeled on abbreviations of ostensive names—after all, then we should expect that there is a name that the descriptive name abbreviates.

606

VI. Names and Descriptions

abbreviation knows what the abbreviation abbreviates. If one used ‘w/’ to abbreviate ‘with’, understanding one’s own words requires that one be competent with ‘with’ and, moreover, that one knows that ‘w/’ abbreviates ‘with’. Yet no such knowledge appears to be required for competent use of descriptive names. The reason we introduce descriptive names is not for linguistic economy.

9. Anticipation Response Another response acknowledges that the DDN view must explain why the name is introduced, and, in fact, oVers up an explanation that is sensitive to considerations about psychological neutrality. The view maintains that names are essentially vehicles for thinking and speaking about objects in a psychologically neutral fashion. It holds that descriptive names are names that aim to fulWll this function, but do not in fact fulWll it. Descriptive names are introduced in anticipation of a time when one will be in position to speak and think of the named object in a psychologically neutral fashion. One introduces descriptive names into the language precisely because one expects that at some point in the future—once acquaintance relations are established—one will be able to directly refer to and think about the named object.27 While I think that this response is right to isolate psychological neutrality as the function of names, the oVered analysis of why descriptive names are introduced into the language is oV the mark. For it fails to explain why the reference-Wxer introduces the name now for the way in which it now functions in language and thought, that is, according to the DDN theorist, descriptively. Yet this is precisely what needs explaining. Furthermore, the suggestion itself is implausible. Descriptive names are not introduced because one anticipates a future time in which one will be speaking and thinking about the named object in a psychologically neutral fashion. It is rather to begin (now—with the introduction of the name) speaking and thinking of the object directly, with no privileged mode of presentation. Think about the introduction of ‘Jack the Ripper’—or, better yet, take the less dated descriptive name ‘Unabomber’. The federal authorities did not introduce the name because they were anticipating that at some point, when they discover who was responsible for the mail-bombings, they would be in position to speak and think about that individual nondescriptively, just by tokening the name. They introduced it so that they could speak and think (or perhaps because they were already speaking and thinking) about the bomber in a de re way, without any particular privileged mode of presentation. Those of us who used the name ‘Unabomber’ prior to Kozinski’s apprehension were also thinking and speaking in a de re way. 27 Recanati’s (1993) interesting discussion of descriptive names contains the anticipation response as a possible explanation of descriptive names’ introduction, though of course he is not explicitly responding to my challenge.

Jeshion, Descriptive Descriptive Names

607

We were not speaking and thinking descriptively, anticipating that it would come to be de re. It already was.

10. Multiple Function Response Let us consider one Wnal response to our central worry. The argument adduced above is misguided because it assumes that there is one basic linguistic and cognitive function of proper names. There are, in fact, many such functions, and consequently many reasons for introducing descriptive names. Some descriptive names are introduced to pave the way for modal thought and talk, others to skirt syntactic ambiguities, and yet others to save breath. To this objection, three points. First, I doubt that, for all descriptive names, we can pinpoint a plausible analysis of the function of that name without appealing to psychological neutrality. This is all that I need to defeat the DDN view. I do acknowledge that perhaps none of the descriptive names discussed so far will stand as a single knockdown example. But, given that the considerations I raised against each alternative analysis of the function of names are broadly applicable, it is worth considering each descriptive name one-by-one to see if any other alternative analysis is really plausible as the reason for the name’s introduction. In any event, much of what I shall say in a moment about the vast amount and variety of descriptive names will encourage skepticism on this point. Second, nothing in this response has been oVered to suggest that psychological neutrality is not a function of descriptive names. (I believe that the semantic argument shows that it is.) It is compatible with my view that there are, in addition to psychological neutrality, further functions that descriptive names may serve—like saving breath, securing a simple means for engaging in modal talk. But the DDN view must deny psychological neutrality. Last, though it is perfectly consistent to maintain that descriptive names have multiple functions, this analysis is, other things equal, considerably less attractive than one that oVers a unifying function.

11. Dime a Dozen Many philosophers of language and mind think: Rarity: descriptive names occur extremely rarely in natural language. This thesis is sometimes explicit, as in Evans (1979), sometimes implicit, as witnessed by the handful of descriptive names that are recycled as examples that occur in natural

608

VI. Names and Descriptions

language: ‘Jack the Ripper’, ‘Neptune’, ‘Deep Throat’, ‘Vulcan’. And it is apparent not only from the words of Evans, but from the general thrust of the DDN view, that most all DDN theorists embrace Rarity. After all, these theorists typically regard descriptive names as very special cases, and this drives them to suggest that the way in which we refer and think with such names diVers dramatically from the standard way of thinking with names, that is, ostensive names.28 I think that Rarity is mistaken. What is more, I think that appreciating the amount and variety of descriptive names in natural language pushes us toward abandoning the acquaintance condition on de re thought. Let us start oV by considering the standard examples: ‘Jack the Ripper’, ‘Neptune’, ‘Vulcan’, ‘Deep Throat’. Of course, we can easily add to this class other names of famous murderers: ‘Unabomber’, ‘Son of Sam’, and so on. The scientiWc examples are less common, though more current examples exist. I was recently reading an article about a physicist at the University of Chicago who said that his major goal in the last few years has been to discover the cause of the expansion of the universe, yet that all he has accomplished in this task is to give it the name ‘Dark Energy’. What uniWes this class of descriptive names is that they name individuals we think exist but we are presently unable to identify, in the sense that we lack a certain variety of ‘knowledge-who’ or ‘knowledge which’, yet, and this is the important point, we (or others in our community) are currently planning activities and actions in response to the named individual. We introduce the name because we aim to catch, or avoid, or discover the named individual. We also introduce descriptive names for objects that we are currently, or soon intend to be, constructing. Imagine Gaudı´ in 1887 with plans nowhere but in his head for an enormous cathedral with fantastic spires, coining a name ‘Sagrada Familia’ for that cathedral that will result from this seed of an architectural plan. Acquaintance with the object named cannot have played a role in the reference-Wxing since the named object did not yet exist. The name is introduced with a deWnite description used attributively—that cathedral, whatever it is, that is the result of these architectural-plans-inthe-head. This class also contains cases in which an artist names a painting he intends to begin, but for which he has only done preliminary sketches. Names of movies are oftentimes (perhaps usually) introduced descriptively: Psycho is the movie that we will construct based on this script. Maybe you have a name of a paper you have not yet completed writing—in fact, maybe you’ve written none of it—though the basic ideas, hazy as they may be, are in your head. This was surely the state of things for me with respect to ‘Descriptive Descriptive Names’ back in February 2001. I think that the names of Wctional characters and Wctional works may belong to this group. We can imagine Kundera coining the name ‘Teresa’ for a certain Wctional character he was constructing via the process of writing The Unbearable Lightness of 28 DDN theorists are not the only ones who embrace Rarity. Plenty of purist Millians (Millians about the semantics of both ostensive and descriptive names) do as well.

Jeshion, Descriptive Descriptive Names

609

Being. The naming need not have occurred explicitly, but if the character was named during (and not after) the construction of the novel, it was likely descriptive. The name of the novel itself was probably descriptive as well.29 Sometimes we name events we are constructing, or planning for, yet which do not currently exist. Think of the person who introduced the descriptive name ‘Woodstock Arts and Music Fair’ to name that festival-in-the-making that would occur from August 15–17, 1969 on Max Yasgur’s farm. If you have organized a conference, you have probably done it too: ‘Barcelona Conference on the Theory of Reference’, ‘Metaphysics Mayhem’. Twenty years ago or so, my father planned on starting oV a business venture delivering cakes to restaurants. Even before making his Wrst investment in a refrigerated truck, he coined a name for his future business: ‘Dessert Sensations’. There was no object with which he was acquainted that would have enabled him to engage in ostensive reference-Wxing. It did not exist yet. But he was planning its construction and so introduced ‘Dessert Sensations’ to name his (then) future cake-delivering business. Lots of business names begin this way—and they are every bit as legitimate names as ostensive names of individuals. The White Pages are full of ostensive names; the Yellow Pages are full of descriptive names. What characterizes this class of cases is that we introduce a name of an object we are currently constructing or intend to construct. So it is an object we believe will exist in the future. As in our last class of cases, the name is introduced because we are, or will be, planning certain activities and actions in connection with this to-be-constructedindividual. Here, the actions and activities are not aimed at discovering, catching, or avoiding the named individual. They are, rather, aimed at shaping—at constructing— that individual. Another class of descriptive names is rather special, but important. These are names of numbers or other mathematical objects with which we lack acquaintance. The name ‘p’ is a good candidate here. The name was introduced descriptively to refer to the ratio of the circumference to the diameter of a circle. ‘p’ names an irrational number— a number that, quite plausibly, we could not be acquainted with.30 That is, we could not have any variety of Kantian intuition of p. It is not out of the question that all of the names of natural numbers are descriptive names. Or maybe we have Kantian intuitions of, hence a variety of acquaintance with, some natural numbers—say 1, 2, 3 and perhaps some others. Then the numerals ‘1’, ‘2’, ‘3’ are ostensive names. Still, it may well be that we lack acquaintance with larger natural numbers. Maybe at around 17 we lack acquaintance. If so, then the numerals ‘17’, ‘18’ ‘19’, . . . ‘a google’ are perhaps descriptive names. 29 I am hedging here because I recognize the room for dispute due to certain views about the metaphysics of Wctional characters and Wctional works. 30 Though (on some views) we may be acquainted with the circumferences and the diameters of particular circles, in my view we are never acquainted with their ratio. Our access to ratios of this sort lacks the relevant directness of a (Kantian) intuition.

610

VI. Names and Descriptions

I will not theorize too much about this class of cases because I cannot here go into the associated issues in philosophy of mathematics.31 But I do want to at least tentatively suggest the following: whatever the extension of the class of mathematical descriptive names (only the names of irrationals, or also the names of large naturals, or also the names of all the naturals, or . . . ), they were introduced because we wish to be able to think about any number in the way in which we think about those objects with which we have acquaintance. We aim to think of them in an object-like fashion, as opposed to descriptively. I think that this survey of classes of descriptive names shows quite clearly that Rarity is false. Descriptive names are not rare birds.32 What unites all the varieties of descriptive names is this: descriptive names are introduced precisely because we have the intention or desire to speak and think (now) of the F in a direct, object-like fashion. We think or desire to speak and think of the F in a psychologically neutral way—the way in which we speak and think of objects by using their ostensive names. The DDN view conXicts with these points. We need a theory of the semantics and psycho-semantics of descriptive names that respects them.

12. An Alternative I want to use a critical examination of the two motivations for the DDN view (considered in §. 3) as a springboard for a new analysis of the psycho-semantics of descriptive names.33 The Wrst, semantic, motivation is rooted on the diVerence between the way that the descriptive names and ostensive names are introduced. It privileges the ‘descriptive component’ of descriptive names—that their reference is Wxed with a deWnite description—but at the expense of the ‘name’ component of the descriptive name. The problem that this engenders is precisely the one we have been rehearsing, namely that the DDN view cannot plausibly explain why a name is being introduced into the language in the Wrst place. Our analysis needs to account for the descriptive name’s ‘nameness’. The second, epistemic motivation says that, without acquaintance with the named object, any thought that the stipulator has by mentally tokening ‘N is G’ must be 31 Matters about the status of terms for numbers and other mathematical objects are extremely complex, and require separate attention. It seems that, while the names of some irrational numbers have their reference Wxed descriptively, they may function more like natural kind terms and artifact terms than ordinary proper names. 32 To be sure, my argument by appeal to examples here trades on my identifying certain names as ostensive and some as descriptive. But one might have an epistemic worry: how do you tell whether a name is ostensive or descriptive? This is an interesting and legitimate epistemological question, one that should receive extended discussion. But I do not take seriously the idea that we have no good grounds for classifying certain names as ostensive, others as descriptive. Still, in particular cases it may be extremely diYcult, maybe even impossible, to determine whether the name is one or the other (or neither). 33 Nothing I have said in this chapter entails, or constitutes an argument for, the acquaintanceless position I favor. I am here only clearing away barren pastures and preparing the way for a more fertile theory.

Jeshion, Descriptive Descriptive Names

611

exactly the same as she has by mentally tokening ‘The (actual) F is G’. The reason why the thoughts must be the same is that there is no change in the stipulator’s information about the object. The problem here is that, although descriptive reference-Wxing plainly generates no new information about the object, it is far from clear that it does not eVect a psychological change. I think it does. Consider two subjects, A1 and A2 who are in exactly the same epistemic informational state. They both think some object O exists that is the F . Both (think they) are not acquainted with the object that is the F . Suppose that, initially, both have only de dicto beliefs about the object that is the F . But at some point, A 1 aims to collect further information about the F , and so stipulates that N names the F . A 1 satisWes Psychological Neutrality (Description), and so aims to think about the F directly, non-descriptively, and in an object-like fashion. A2 does not intend to think about the F in a psychologically neutral way at all. I think that there is an important psychological diVerence between A 1 and A2 , one which is independent of their informational states. A 1 ’s thought about the referent of N is now de re while A2 ’s is merely de dicto. I am not here trying to prove this point with an a priori argument—it isn’t that sort of thesis. In particular, my point is not that it follows immediately (analytically), from one’s satisfying Psychological Neutrality (Description), that one will in fact think of O in a direct object-like fashion. I think it follows because of an empirical fact about our psychologies. Our minds are sensitive to our basic intensions to think about things in an object-like fashion. In fact, I would hold that alteration in the structure of our thought due to this sensitivity is not under our voluntary control. I think it happens to us. It might well be that the best we can do to support this claim is to encourage reXection on cases and one’s own psychology in assessing its worth.34 The literature on de re belief frequently makes use of the idea of the mental Wle folder. For the many objects with which we are acquainted and for which we think about by mentally tokening an ostensive name, we have mental Wle folders that, in eVect, organize our information about the object. The name of the object is conceived as the label on the folder and serves the ordinary organizational function of labels on Wles—it enables us to eYciently organize and retrieve information about the named object. Within the bulk of this literature, acquaintance with the named object is a necessary condition on having de re thought about it. In my view, we need to shed the acquaintance condition, and recognize that having a mental Wle folder itself is suYcient for de re thought. We generate mental Wle folders and de re thought by introducing descriptive names into the language. In introducing the descriptive name, A1 opens and labels a new mental Wle folder as a repository of information about the F . Initially, she had a series of de dicto thoughts that were

34 I am not ruling out the possibility of supporting the view with a transcendental argument. Perhaps one could be developed. But nothing of the sort is what I am after here.

612

VI. Names and Descriptions

unorganized in cognition. The creation of this mental Wle folder itself alters this agent’s cognitive architecture so that her thought about O is now de re. Clearly this view stands in need of development. An enormous amount of detail and potholes need Wlling.35 But the examples considered in suggesting that there are constraints on introducing descriptive names and that they are not so terribly rare should make this alternative tantalizing. Agents’ thoughts about Jack the Ripper and Woodstock and Dessert Sensations were de re simply because they had the same organizational and structural role in cognition as their thoughts about Vela´zquez. 35 For a somewhat fuller discussion, see Jeshion (2001b). The view is much more carefully spelt out in Jeshion (2003).

19 Descriptively Introduced Names Marga Reimer

1. Preliminaries Consider the following two scenarios. Scenario 1 Nineteenth-century astronomer Leverrier discovers unexplained perturbations in the orbit of Uranus. He hypothesizes that there is a planet causing these perturbations—a planet yet to be identiWed telescopically. He introduces the name ‘Neptune’ into the language by stipulating that ‘Neptune’ refers to the planet hypothesized to be causing the perturbations in question. Scenario 2 Philosopher of language Gareth Evans (1979, 1982) is interested in theorizing about names like ‘Neptune’—names introduced into the language by way of stipulative acts of descriptive reference-Wxing. He wants an example of his own and so introduces the name ‘Julius’ by stipulating that ‘Julius’ refers to whoever it was that invented the zipper. ‘Neptune’ and ‘Julius’ are examples of what I will refer to as ‘descriptively introduced names’. Such names (‘d-names’ for short) are expressions that ordinary folk would classify as names and yet, unlike the vast majority of so-called names, d-names are not introduced into the language by way of anything remotely like a baptism. In the case of an ordinary (non-d) name, there is typically some yet to be named entity x with which the namer is ‘acquainted’ in an intuitive and non-technical sense of that locution. The namer wishes to speak about x and so introduces a name N for this very purpose, dubbing x N . Because such dubbings are often accompanied by ostensive acts of some sort, such names might be called ‘ostensive’ names. D-names are diVerent. In the case of a d-name, the namer believes—perhaps mistakenly—that there is (was or will be) some entity x that satisWes a particular deWnite description, and he wishes to talk about I would like to thank Kent Bach, Dave Chalmers, Michael Devitt, Robin Jeshion, Ernie Lepore, and Mark Richard for helpful comments on earlier drafts of this chapter.

614

VI. Names and Descriptions

this entity. But because he is not acquainted with x (at least not qua the F ), he cannot dub x in the usual way, by ostension. So he introduces a name N and stipulates that N is to refer to whatever it is that satisWes the deWnite description in question. D-names appear to be rather uncommon, though without question they exist in natural language. The expressions ‘Neptune’, ‘Vulcan’, ‘Jack the Ripper’, and ‘Deep Throat’ are all examples of such expressions. And, of course, artiWcial cases can be found in the philosophical literature on names—we have Kaplan’s (1969) ‘Newman 1’, Evans’s (1979, 1982) ‘Julius’, Devitt’s (1981a) ‘Oscar’, and Jeshion’s (2001a) ‘Oldman 1’. Uncommon though they may be, d-names constitute an interesting and important class of expressions. They look like ordinary names—indeed, we call them ‘names’. It is thus natural to suppose that the semantic content (or ‘propositional contribution’) of a d-name will be similar in kind to that of an ordinary name. As Salmon (1986), Soames (1989a), and others have argued, it is plausible to suppose that the semantic content of an ordinary name like ‘Kripke’ is simply the name’s bearer: Kripke. (Such theorists endorse Mill’s (1843) view of proper names, and are thus often referred to as ‘Millianians’.) Because the expressions ‘Neptune’ and ‘Julius’ look rather like the expression ‘Kripke’, one naturally expects their semantic contents to be similar in kind—Neptune and Julius, respectively. And yet d-names seem to refer—or rather to denote—just like deWnite descriptions do. Such expressions (at least when Wrst introduced into the language) denote the entities they do in virtue of the fact that those entities ‘satisfy’ a particular deWnite description. ‘The chair of Princeton’s philosophy department’ denotes the one and only chair of Princeton’s philosophy department. Similarly, ‘Jack the Ripper’ (as introduced by the English of the late 1880s) denotes the one and only murderer of certain London prostitutes; ‘Julius’ (as introduced by Gareth Evans 1979, 1982) denotes the one and only inventor of the zipper. What is one to make of these interesting expressions? It is not implausible to suppose that d-names are what Kripke (1980) has called ‘rigid designators’: expressions that refer to the same object/individual in every possible world in which that object/ individual exists. (See Soames 1998 for an alternative view.) Intuitively, ‘Jack the Ripper’ refers to the actual murderer of certain London prostitutes, even in possible worlds where he never murders anyone. If I say that Jack the Ripper might never have met the said prostitutes, I say nothing that is not perfectly coherent. The diYculty is that the putative rigidity of d-names is consistent with both the view that they are Millian and the view that they are rigidiWed deWnite descriptions: expressions of the form the F , interpreted as denoting the entity (if any) that is uniquely F in the actual world, in every possible world in which that entity exists. Thus, the rigidity of such expressions will not enable us to adjudicate between these two competing views. But this is precisely the debate that I would like to adjudicate. Accordingly, the question that I will be attempting to answer in what follows is:

Reimer, Descriptively Introduced Names

615

What is the semantic content of a d-name? What (in other words) does such an expression contribute to the proposition expressed (to ‘what is said’) by the sincere utterance of a sentence containing it? I will begin by considering two responses, both of which treat d-names as rigid designators. According to the Millian response, the semantic content of such an expression is the denotation (if any) of the description that Wxes its reference, at the time that the reference is initially Wxed. There is no semantic content if there is nothing that satisWes the description in question. According to what I will call ‘R-Descriptivism’, the semantic content of a d-name is that of the description that Wxes its reference, interpreted rigidly: that is, interpreted as denoting the entity (if any) that is uniquely F in the actual world, in every possible world in which that entity exists. The particular version of R-Descriptivism that I will be discussing employs a Russellian (quantiWcational) analysis of deWnite descriptions. Although one need not be a Russellian about descriptions to be a descriptivist about names, I am assuming a Russellian analysis of descriptions simply because it is the bestknown and arguably most promising account of such expressions (see Neale 1990). I will conclude that neither Millianism nor R-Descriptivism provides an adequate account of d-names, as neither provides a plausible account of all of the relevant data. I will go on to suggest a hybrid account, according to which the R-Descriptive analysis is correct so long as acquaintance-based uses of the d-name have not become established. (My notion of ‘acquaintance-based’ uses will be detailed below, in s. 4.) If and when acquaintance-based uses of the name do become established, a Millian analysis is correct.

2. The Data Before looking at Millian and R-Descriptive accounts of d-names, let us have a quick look at the data, which concern expressions such as the following: Neptune Vulcan Deep Throat Jack the Ripper Julius Newman 1 What do these expressions have in common? They all look like names: they are singular terms, terms that purport to refer to some particular entity, marked as names via the convention of capitalization. And yet their introduction into the language is diVerent than in the case of other expressions we call ‘names’. In particular, there is no baptismal ceremony, however informal, at which some perceived object or individual

616

VI. Names and Descriptions

is dubbed N . Rather, d-names are introduced via stipulative acts of descriptive reference-Wxing. Sometimes the reference-Wxing is by Wat—as when Kaplan (1969) declared that ‘Newman 1’ was to name the Wrst child born in the twenty-Wrst century. Sometimes, the reference-Wxing is not a single, explicit act but a gradual process, often involving a community of speakers, as when the English of the late 1880s came to refer to the unknown slayer of assorted London prostitutes as ‘Jack the Ripper’. As noted above, the ordinary use of such expressions suggests that they are rigid designators. We say, for instance, such things as that Jack the Ripper might not have committed the murders had he come from a loving home; or that Julius might never have invented anything had he lacked the ingenuity. That d-names are rigid designators is consistent with their being either Millian or R-Descriptive. So, which (if either) are they? Let us consider each of these two possibilities, together with the considerations that respectively motivate them.

3. The Views, the Arguments, and the Counter-Arguments The Views According to the Millian account, the semantic content of a d-name is the denotation (if any) of the description that Wxes its reference. The proposition expressed by an assertive utterance of ‘N is G’, where N is a d-name, is x is G, where x is the F . Thus, an assertive utterance of ‘Neptune is a planet’ expresses a ‘singular’ proposition to the eVect that x is a planet, where x is Neptune itself. According to the R-Descriptive account, the semantic content of a d-name is the semantic content of the deWnite description that Wxes its reference, interpreted rigidly. Thus, the proposition expressed by an assertive utterance of ‘N is G’, where N is a d-name, is identical to the proposition expressed by an assertive utterance of ‘the F is G’, where the description is interpreted as denoting the entity (if any) that is uniquely F in the actual world, in every possible world in which that entity exists. Thus, an assertive utterance of ‘Newman 1 is now 3 years old’ expresses a ‘general’ (or ‘object-free’) proposition to the eVect that the Wrst person born in the twenty-Wrst century (whoever that may be) is now 3 years old.

The Arguments and the Counter-Arguments The considerations that motivate the view that d-names are Millian are the same sorts of considerations that motivate the view that names, generally speaking, are Millian. However, it appears as though there are plausible responses that the R-Descriptivist might make to these considerations. Moreover, there are well-known arguments that support the view that names, generally speaking, are descriptive and some of these can arguably be extended to support the view that d-names are R-Descriptive. But, perhaps

Reimer, Descriptively Introduced Names

617

not surprisingly, there are plausible responses that the Millian might make to these arguments. The result is a stalemate of sorts. To see this, let us begin by looking at the considerations that support a Millian account of d-names, and then turn to the responses likely to be made by the R-Descriptivist. Afterward, we can have a look at the sorts of considerations that an R-Descriptivist might oVer on behalf of his own positive view of d-names. The argument from non-analyzability Intuitively, d-names do not appear to have descriptive meanings. Contrast the d-name ‘Julius’ with the deWnite description ‘the inventor of the zipper’. Intuitively, the latter has a meaning (a semantic content)—one that is analyzable in terms of the meanings of its parts—while the former no more has a meaning than does ‘Kripke’. To the extent that ‘Julius’ does have a meaning, that meaning would appear to be the individual the name refers to: the inventor of the zipper (whoever he is). Now although some d-names do have ‘parts’ of sorts, these parts are not semantically signiWcant. For the meaning of such names (whatever it is) is clearly not analyzable in terms of the meanings of these parts. Hence, whatever the meaning of ‘Deep Throat’, it is not one analyzable in terms of the meanings of ‘deep’ and ‘throat’. Whatever the meaning of ‘Newman 1’, it is not analyzable in terms of the meanings of ‘new’, ‘man’, and ‘1’. Here, the R-Descriptivist might respond as follows. It is certainly true that d-names do not obviously have complex, analyzable meanings. However, nor do other expressions which arguably do have such meanings. Consider the expression ‘bachelor’ or the artifactual kind term ‘pencil’. It is not implausible to suppose that the meaning of the Wrst is that of the description ‘unmarried adult male’, while the meaning of the second is (something like) that of the description ‘graphite or lead Wlled writing implement’. Thus, to understand the meaning of ‘bachelor’ one must understand the meaning of ‘male’, and to understand the meaning of ‘pencil’ one must understand the meaning of the verb ‘to write’. And yet because the expressions ‘bachelor’ and ‘pencil’ are syntactically simple, one is initially disinclined to suppose that their semantics could be complex. Similarly, it might be claimed that to understand the meaning of ‘Julius’, one must understand (inter alia) the meaning of the word ‘inventor’—which is not quite the same as understanding that ‘inventor’ means what it does. It is more like knowing what an inventor is. But it is only this latter sort of conceptual (non-meta-linguistic) knowledge that is required on an R-Descriptive account. To know the meaning of ‘Newman 1’ is to know the meaning of ‘the Wrst person to be born in the twenty-Wrst century’—it is not to know of the reference-Wxing description itself, that it has the particular meaning that it does. The semantic argument The Millian view of d-names would appear to be reinforced by the fact that one can be a competent user of a d-name such as ‘Neptune’ even if one is unaware of its referenceWxing description. Indeed, one can be a competent user of the name even if one

618

VI. Names and Descriptions

associates with the name a description that is satisWed by some object other than Neptune (e.g. ‘the planet between Mars and Saturn’). The competent use of a d-name contrasts sharply with the competent use of its reference-Wxing description. To be a competent user of the expression ‘the planetary cause of Uranus’s orbital perturbations’ one must know, among other things, what the expression ‘orbit’ means. But I can be a competent user of the name ‘Neptune’ even if I do not know this. (This argument amounts to an extension of Kripke’s (1980) ‘semantic’ argument against what he calls the ‘Frege–Russell’ theory of proper names.) But here too, there seems to be a plausible response available to the R-Descriptivist. Such a theorist might argue as follows. A speaker who is unaware of the descriptive content of a d-name is unaware of its meaning. This does not mean that such a speaker cannot succeed in referring by means of the name. The analogy would be to a speaker who accurately reports that the problem with his car is that the carburetor needs to be replaced, without knowing what the expression ‘carburetor’ means. The intuition that the user of ‘carburetor’ knows what he is talking about is an illusion created (in part) by the fact that failure to know the meaning of an expression does not prevent one from using it to refer successfully. Similarly for the user of a d-name who does not know the descriptive content of that name. This does not mean that he cannot refer by means of it, nor does it mean that reference is not achieved via satisfaction of the associated reference-Wxing description. What if the speaker has a partial knowledge of the descriptive content of a d-name— what if, for instance, he knows that the individual named ‘Julius’ invented something? The R-descriptivist would claim that he has a partial grasp of that name’s meaning— which is just what our intuitions would seem to suggest. (Similarly, someone who— like myself—knows only that a carburetor is part of a car’s engine, has only a partial knowledge of the meaning of ‘carburetor’.) However, even with only a partial grasp of an expression’s descriptive content, reference by means of that expression is still possible—provided the name’s content is (uniquely) satisWed by some object or individual. Moreover, the R-Descriptivist can rightly insist that the description alleged to constitute the semantic content of a d-name is guaranteed to ‘get the reference right’. There is no possibility of anything like Kripke’s (1980) Go¨del/Schmidt case arising with respect to a d-name. For in the case of a d-name, it is stipulated that the name refers to whatever (if anything) is in fact the F . The argument from contingency Consider the following three sentences: (1) (2) (3)

Neptune (if it exists) causes perturbations in Uranus’s orbit. Jack the Ripper (if there was such a person) targeted prostitutes. Deep Throat (if there was such a person) was an informant.

Some might claim that the truths expressed by assertive utterances of sentences such as these are contingent, but they should be necessary if d-names are semantically equivalent

Reimer, Descriptively Introduced Names

619

to the descriptions that Wx their reference. In this respect, sentences (1) through (3) are more like (4) than like (5): (4) (5)

Kripke (if there is such a person) authored Naming and Necessity. Squares (assuming there are such things) have exactly four sides.

For while the truth expressed by (4) appears to be contingent, that expressed by (5) appears to be necessary. (Here, we have what amounts to an extension of Kripke’s (1980) ‘epistemological’ argument against the ‘Frege–Russell’ theory of proper names.) What might the R-Descriptivist say in response? He might claim that the apparently contingent nature of the truths imparted by (1) through (3) is easily explained. The intuitor is simply unaware of the name’s descriptive content. Once the intuitor were supplied with the relevant descriptive content, the necessary nature of the truths imparted by these sentences would become obvious. An analogous case would involve a speaker who Wnds the truth that carburetors (if they exist) mix gasoline and water, to be contingent. Such a speaker simply does not know what the expression ‘carburetor’ means. Thus, for such a person, the truth in question will sound contingent. The argument from error Imagine the following scenario. Several years after the telescopic identiWcation of the planet we now call ‘Neptune’, it is discovered that the planet so-called is not (contrary to what had been supposed) the cause of Uranus’s orbital perturbations. We would probably not conclude that the planet in question was not Neptune after all. Moreover, we would probably continue to refer to that planet as ‘Neptune’. This seems at odds with the view that the name is semantically equivalent to the description ‘the planetary cause of Uranus’s orbital perturbations’. After all, after our hypothetical discovery, we would never think of applying that description to the planet we call ‘Neptune’. There are two responses the R-Descriptivist might make here. First, he might counter with a tu quoque: Suppose that, just hours or days after the telescopic identiWcation of the planet we now call ‘Neptune’, it is discovered that the planet so-called is not in fact the cause of Uranus’s orbital perturbations. Would that planet still be called ‘Neptune’? Probably not; we would probably discontinue to apply the name ‘Neptune’ to the planet in question on the grounds that the planet was not the cause of the perturbations in question. This would be just what one would expect—assuming ‘Neptune’ to be semantically equivalent to the description that Wxes its reference. Second, one might admit that, although we would still call the planet we have been calling ‘Neptune’ ‘Neptune’ in the case where the discovery of our error is made long after the telescopic identiWcation of Neptune, that only shows that the meaning of the name would eventually shift from that of the rigidiWed description to the planet itself. (This is roughly correct on the view for which I go on to argue.)

620

VI. Names and Descriptions

The argument from emptiness So much for the R-Descriptivist’s responses to the arguments for Millianism. What sorts of considerations might such a theorist adduce on behalf of his own positive view? As is well-known, intuitive data surrounding identity statements, propositional attitude attributions, and empty names lend credence to a broadly descriptive (Frege– Russell) account of proper names. These same sorts of considerations might be invoked on behalf of an R-Descriptive account of d-names. Let us focus on those having to do with empty names (including true negative existentials), as the anti-Millian intuitions here are especially compelling. Suppose that, as it turns out, the various prostitutes whose murders were attributed to Jack the Ripper were actually murdered by their own husbands. In that case, sentences like (6) and (7) would nevertheless appear to be meaningful—to express ‘propositions’. (6) (7)

Jack the Ripper was crazy. Jack the Ripper never really existed.

These intuitions would be preserved on an R-Descriptive account of d-names. In contrast, a Millian who maintains that ‘Jack the Ripper’ has no reference (Salmon 1998 is an exception), would seem forced to explain away the intuitions of truthevaluability and meaningfulness. In particular, such a theorist would have to claim that, while something truth-evaluable might well be conveyed by assertive utterances of these sentences, nothing at all would be said (or ‘literally expressed’). The confusion would not be diYcult to characterize, as it is arguably just the sort of confusion made by descriptivists with respect to indexical expressions. Such theorists claim (for instance) that the meaning of ‘I’ is that of ‘the speaker’ and the meaning of ‘here’ is that of ‘the location of the speaker’. On the sort of view made famous by Kaplan (1989a), the confusion is the result of mistaking the linguistic ‘character’ of an indexical for its truth-conditional ‘content’. The Millian analysis of empty d-names is, however, less intuitive than that provided by the R-Descriptivist. For it eVectively claims that certain strong and uniform intuitions regarding truth-evaluability and meaningfulness are misleading. (I will return to the argument from emptiness below, in s. 7.)

4. Assessment of the Arguments/Responses The foregoing arguments and responses all suVer from the same basic defect: the intuitions appealed to are selective in a way that is clearly self-serving. The Millian arguments and responses tend to appeal to intuitions surrounding ostensive names and d-names the uses of which have become acquaintance-based. The R-descriptive arguments and responses tend to appeal to intuitions surrounding d-names the uses of

Reimer, Descriptively Introduced Names

621

which have not become acquaintance-based. In calling the use of a name ‘acquaintance-based’ I mean that it involves reference to an object with which the speaker has acquaintance (qua bearer of that name) or that it is part of a Kripkean causal chain originating in such uses. Moreover, when the use of a name is ‘acquaintance-based’ (in my sense of that locution), it is applied to the object of acquaintance without regard to the properties of that object; in particular, it is applied without regard to whether the object has the properties speciWed by the description that originally Wxed the name’s reference. The intuitions appealed to in the R-descriptivist arguments concern d-names whose current use is primarily descriptive. In saying that the current use of a d-name is primarily descriptive, I mean to say that the majority of uses of the name are not acquaintancebased. The diYculty is that what is true of d-names like ‘Julius’ is arguably not true of d-names like ‘Neptune’. I think this is precisely what the preceding arguments and responses tend to show. D-names like ‘Julius’ are arguably R-Descriptive; d-names like ‘Neptune’ are arguably Millian. But we need an account of the semantics of d-names that explains these facts. It is not enough to simply say that d-names currently grounded in acquaintance-based uses are Millian, while those currently grounded in descriptive uses are R-Descriptive.

5. An Alternative Hybrid Account When a d-name is introduced into the language, whether by Wat or through some gradual, community-based process, its semantic content is that of the description that Wxes its reference. This enables us to explain the intuitions behind the arguments for R-Descriptivism (which focus on names like ‘Julius’). Not only does the hypothesis explain these intuitions, it makes a good deal of sense given an intuitive understanding of linguistic meaning. To be a competent user of a linguistic expression is to ‘know the meaning of ’ that expression—whatever this amounts to. To know the meaning of an expression is arguably to know that meaning in an ‘acquaintance’ sense of knowledge. This is why we often talk in terms of a speaker’s ‘grasping’ the meaning of an expression. ‘Grasping’ something implies a kind of acquaintance with what is grasped. But, by hypothesis, when a d-name is introduced into the language, there is no acquaintance with its referent (at least not qua denotation of the reference-Wxing description). Nevertheless, if competence in using the d-name is to be even possible, the speaker must be in a position to grasp something that might plausibly be considered to be the name’s meaning. The content of the description that Wxes the name’s reference is something that the competent speaker might well be in a position to grasp, as the act of reference-Wxing is presumed to be a public act. This does not of course suggest that the competent speaker will have actually witnessed the reference-Wxing act, only that he will be aware of its eVects—that (in other words) he will be aware of the fact that N refers to whatever is the actual F . At any rate, that a d-name N refers to the F is

622

VI. Names and Descriptions

arguably something that the competent user of that name will know. Moreover, the descriptive content is arguably relevant to the (truth-conditional) meaning of the expression, as it determines its reference. Since the nominatum is not a candidate for a graspable meaning, the content of the name’s reference-Wxing description is a natural—indeed, an obvious—candidate for that meaning. If and when the denotation of the reference-Wxing description is (perceptually) identiWed, the possibility of acquaintance-based applications becomes a reality. If and when such applications become established, the descriptive name acquires a Millian semantic content. This enables us to explain the intuitions behind the arguments for Millianism (which focus on names like ‘Neptune’). Not only does the hypothesis explain these intuitions, it also coheres with the plausible if not obvious view that—in some sense at least—the meaning of an expression is determined by its use. If an expression that means what some description means is used to refer to the denotation of that description without regard to the speciWed properties, then if this sort of use becomes established, the name no longer refers by description. Rather, it refers via some sort of acquaintance-based (and presumably causal) relation between the speaker and the referent. No doubt, when such meaning shifts do occur, they are gradual: there is no particular point at which one can truly say: OK, now the name is Millian. Indeed, it is likely that during some phase of the transition from R-Descriptive to Millian semantic content, there is simply no fact of the matter as to what the expression’s semantic content is.

6. Further Evidence for the Proposed View Before considering some possible objections to the proposed view of d-names, I would like to draw attention to two further considerations that favor the view. The Wrst consideration has to do with the generalizability of the proposed view; the second has to do with a philosophical problem Wrst noted by Kripke (1980)—the so-called problem of the contingent a priori.

Natural Kind Terms Sometimes natural kind terms are descriptively introduced into the language before the existence of the kind satisfying the description is conWrmed. This was surely the case with the kind term ‘atom’. We now know that atoms exist—indeed, we can literally see them with the assistance of an electron microscope. But we knew that atoms existed only after the term was introduced. Before such entities were microscopically identiWed, the term ‘atom’ presumably had a meaning. What was that meaning? Originally, it was (something like): indivisible particle. But clearly, that is no longer the

Reimer, Descriptively Introduced Names

623

term’s meaning. After all, we now know that atoms are not indivisible, but contain ‘subatomic’ particles. What, then, does the term ‘atom’ mean? What is its ‘semantic content’? A Millian about natural kind terms would argue that the term’s semantic content is simply the kind that it refers to. Suppose that this is so. How is it possible for a term that once had a descriptive meaning to come to mean a natural kind that no longer even satisWes the term’s original descriptive meaning? This question is easily answered by simply extending the proposed view. The term ‘atom’ was descriptively introduced into the language (probably not by stipulation, but through a gradual process). Its meaning, at the time of its introduction, was that of its reference-Wxing description. This meaning changed with the establishment of acquaintance-based applications of the term to what we now call ‘atoms’. Its meaning, its semantic content, became the kind itself. But note: the current meaning is not the kind actually denoted by the original reference-Wxing description: for ‘atom’ denotes a kind that is divisible. The proposed view, extended to kind terms, easily explains these facts. The kind term originally meant what the reference-Wxing description meant, but acquired a Millian meaning as soon as acquaintance-based applications of the term— applications to something that happened not to Wt the description in question— became the norm. (Millians have of course oVered alternative accounts of the same data. See e.g. Donnellan 1979; Jeshion 2001a.)

The Problem of the Contingent A Priori It seems reasonable to suppose that—prior to the telescopic identiWcation of Neptune—Leverrier (who introduced the name ‘Neptune’) knew what was expressed by the following sentence: (8)

Neptune (if it exists) is the cause of Uranus’s orbital perturbations.

Not only is it plausible to suppose that Leverrier knew what the sentence expressed, it is plausible to suppose that he knew what it expressed to be true. But although what (8) expresses would appear to be a contingent truth, it was a truth that Leverrier seemingly knew a priori—indeed, he presumably knew it as soon as he introduced the name ‘Neptune’ into the language, stipulating (in eVect) that it was to refer to whatever was the planetary cause of Uranus’s orbital perturbations. It would thus seem that there are contingent truths that can be known a priori! This has seemed highly counter-intuitive to many. The proposed view of d-names can easily account for this counter-intuitiveness. The truth grasped by Leverrier is indeed both contingent and a priori—it is the truth that the planet actually causing such-and-such perturbations is the planet causing suchand-such perturbations. It is a trivial truth and so its a priori nature is not surprising. Moreover, it is contingent simply because the actual F might (had the world been diVerent in the relevant sorts of ways) not have been the F . Moreover, the proposed view easily explains the fact that—to contemporaries—the truth expressed by (8)

624

VI. Names and Descriptions

seems a posteriori—and far from trivial. This is because, once Neptune was telescopically identiWed, acquaintance-based applications of the name became possible, and once such applications became established, the name thereby acquired a Millian meaning. Now the sentence arguably expresses a substantive a posteriori truth to the eVect that x is the cause of such-and-such perturbations, where x is the entity we call ‘Neptune’.

7. Objections and Replies There are, I am sure, any number of objections that one might attempt to level against the proposed view of d-names. Here, I will consider half a dozen of the more potentially damaging objections to that view—all of which can, I believe, be adequately responded to.

A Semantic or Epistemic Distinction? Kent Bach (pers. comm., April 2001) has claimed that the distinction I draw between d-names like ‘Neptune’ and those like ‘Julius’ appears to be an epistemic—rather than semantic—distinction. The point is basically this. Even if no one is acquainted with Neptune, Neptune is still arguably the semantic content of the name ‘Neptune’. What follows from the assumption that no one is acquainted with Neptune is simply that no one is in a position to have de re thoughts about it: thoughts whose contents are singular propositions of the form < Neptune, F >. But none of this keeps a certain celestial object from constituting the semantic content of ‘Neptune’. In sum, I seem to be assuming that ‘Neptune’ can have Neptune as its semantic value only if we can have de re thoughts about it. But why assume that? Although I freely admit that the proposed distinction is an epistemic one, I believe that the epistemic distinction underwrites a semantic one. Let me explain. A name N can have its denotation x as its semantic content only if that name denotes x without regard to the intrinsic properties of x. For only in such cases can one use a name to talk about (or ‘express a proposition’ about) ‘the thing itself ’—as opposed to the thing in so far as it satisWes some description or other. It is thus only in such cases that one can speak accurately of using a name as a mere ‘tag’. Now if N denotes x in virtue of the fact that x satisWes the name’s reference-Wxing description, then obviously that name does not denote x without regard to x’s intrinsic properties; the name cannot be functioning as a mere ‘tag’. Under what conditions is it possible to use a name N to denote x without regard to intrinsic x’s properties—and thus as a mere ‘tag’? Only when the speaker’s relation to x is not mediated by description. When do such conditions obtain? It seems to me that they obtain only when the speaker is acquainted with x (qua bearer of N ) either directly or indirectly. The speaker has a direct acquaintance

Reimer, Descriptively Introduced Names

625

with x when he/she has perceptual contact with that thing, qua bearer of the name ‘N ’; and the speaker has indirect acquaintance with x when he/she is a member of a Kripkean causal chain originating in such acquaintance. For only in such cases is there a relation to x that is not mediated by description, but causally. Thus, only in such cases does it make sense to suppose that x is the semantic content of N . The proposed view thus does indeed entail (as Bach realizes) that the semantic content of ‘Neptune’ can be Neptune only if we can have de re thoughts about Neptune: thoughts about the planet itself—not the planet in so far as it satisWes some description or other. For if we cannot have de re thoughts about Neptune, then reference via the name ‘Neptune’ cannot be achieved by way of acquaintance (whether direct or indirect). In that case, the name must refer to the planet by way of description. And, in that case, its use does not enable us to talk about Neptune itself: that is, without regard to any of its intrinsic properties.

Epistemic Relations to Nominata Mark Richard (pers. comm., March 2001) worries that the proposed view makes unintuitive predictions regarding the epistemic relations a speaker has to the referents of the names in his idiolect. Consider a contemporary speaker whose only ‘contact’ with Thales consists of being at the end of a long chain of vocabulary transmission. Why should such a person be thought to have a more interesting epistemic relation to Thales, in virtue of such ‘contact’, than Leverrier did to Neptune, in virtue of his justiWcation for thinking that he had successfully named a planet? It is perhaps misleading to describe the contemporary speaker’s relation to Thales as one of ‘acquaintance’—for this suggests a kind of intimacy that is surely lacking. But the point is that this relation is diVerent in kind from that which Leverrier bore to Neptune in virtue of his introduction of the name ‘Neptune’ into the language. Leverrier no doubt had a more intimate relation to the planet than the contemporary speaker has to Thales, but the point is that Leverrier did not have this relation in virtue of his descriptive use of the name ‘Neptune’. He had it in virtue of his observations of the eVects that Neptune had on the orbit of Uranus. But the relation that he bore to Neptune in virtue of his descriptive act of reference-Wxing was an a-causal one. It is the same sort of relation that I bear to Smarty Pants in virtue of stipulating that ‘Smarty Pants’ is to name the smartest person in the world (whoever he/she is). But the contemporary speaker has a causal—albeit remote—relation to Thales simply in virtue of his being a member of a causal chain originating in acquaintance-based applications of the name to Thales himself.

The Semantic Argument Revisited According to Robin Jeshion (pers. comm., March 2001) there are diYculties with my attempt to circumvent Kripke’s semantic argument. In particular, it is unclear how the

626

VI. Names and Descriptions

special properties of d-names are supposed to make them immune to the semantic argument. After all, I agree that one could use a d-name without having the referenceWxing description in mind. Indeed, I claim that a speaker can successfully refer to the nominatum even though he lacks access to the name’s meaning. But then it seems that I must be giving up some feature of descriptivism. In particular, I must be giving up either (i), the view that the deWnite description expresses the agent’s mode of presentation of the referent, or (ii), the view that the object satisfying the deWnite description is the referent of N on a given occasion of use. In fact, by appealing to the ‘carburetor’ analogy, it seems that I am invoking Putnam’s view that kind terms function in part like proper names in the sense that the theses of descriptivism do not hold even for them. So, it seems that the proposed view cannot be a bona-Wde descriptivism about d-names (prior to acquaintance qua the F ). In order to respond to Jeshion’s concern, I will Wrst have to provide a bit of background information. Many believe that Putnam (1973) showed that descriptivism is false for natural kind terms. Suppose (if only for the sake of argument) that this is so. It does not follow that Putnam has shown that descriptivism is false for artifactual kind terms—expression like ‘pen’, ‘chair’, and ‘carburetor’. To see this, consider the descriptions associated with kind terms by the relevant sorts of experts. Might not such descriptions provide the key to the reference of natural kind terms? Even when culled from the ken of the relevant experts, the associated descriptions might ‘pick out’ the wrong kind. Thus, what we call ‘water’ might turn out to be something other than H2 O, and so the sense of ‘H2 O’ cannot be the meaning of ‘water’. (More picturesquely, for all we know, Earth might be Twin Earth.) But similar considerations do not seem to apply in the case of artifactual kind terms. Could it really turn out that we ordinary folk (the relevant experts) are wrong in supposing that pens are ink-Wlled writing implements or that chairs are designed for sitting in? It seems not. The reason for this is not diYcult to discern: artifactual kinds are individuated—not by some ‘underlying structure’ the nature of which is never really transparent to us—but (at least partly) by their ‘function’—which is as transparent as anything can be, since we determine it. Just as it could not turn out that chairs were not really designed for sitting in, it could not turn out that Jack the Ripper killed no one—not as long as we regard the name’s reference as Wxed by description. These considerations suggest, I think, that d-names (prior to the establishment of acquaintance-based uses) are more akin to artifactual kind terms than to natural kind terms with an established history of acquaintance-based uses. Both d-names (prior to establishment of acquaintance-based uses) and artifactual kind terms are ‘descriptive’ in the following sense. Their semantic content is that of some description; this description determines the reference of the expression; and the sense of the description expresses the mode of presentation of the relevant experts (Leverrier in the case of ‘Neptune’, ordinary folk in the case of ‘pen’, and mechanics in the case of ‘carburetor’).

Reimer, Descriptively Introduced Names

627

The Argument from Emptiness Revisited I suggest that the argument from emptiness supports the view that d-names are—prior to the establishment of acquaintance-based uses—semantically equivalent to rigidiWed descriptions. And yet, I am sympathetic to Millianism with regard to ordinary names. Do I not Wnd the argument from emptiness compelling when applied to empty names that are not d-names? My response here is a simple one. It seems to me that most, if not all, empty (nondenoting) names do have meanings of some sort. This is suggested by the apparent fact that sentences containing empty names can be used to express meaningful propositions. But perhaps what this suggests is that empty names, generally speaking, are d-names. After all, such names cannot really be ‘ordinary’ names: for in ordinary cases, names are introduced by way of something akin to a baptism at which some perceived entity x is dubbed N . Empty names are introduced into the language via something very much like stipulative descriptive reference-Wxing. A name like ‘Santa Claus’ is introduced for the purpose of talking about a unique ‘individual’ constitutionally incapable of being the object of anyone’s ‘acquaintance’. The name is eVectively stipulated to mean (something like): the jolly, red-suited, white-bearded bearer of Christmas gifts. Similar considerations would arguably apply to the names of characters from Wction—names like ‘Sherlock Holmes’. This sort of view has the additional advantage of accounting for the apparent triviality of sentences involving names of mythical or Wctional entities, such as ‘Santa Claus is jolly’ and ‘Sherlock Holmes is a brilliant detective’, which contrasts with the non-triviality of such sentences as ‘Kripke is a philosopher’ and ‘Bush is now President’.

Meaning Shifts Contrary to the proposed account, it does not seem plausible to suppose that once acquaintance-based uses of a d-name become established, the name invariably acquires a Millian meaning. Consider the following situation. Physical anthropologists introduce the name ‘Eve’ and stipulate that it is to refer to the Wrst (female) human being. Suppose that anthropologists Wnd the skeletal remains of what they believe be the Wrst (female) human being. For decades, they refer to the remains (e) as ‘Eve’. Photographs and descriptions of Eve are found throughout physical anthropology journals and textbooks. Acquaintance-based uses of the name have, by this point, surely become established. However, the remains of an earlier female human (e1) are subsequently discovered, and anthropologists throughout the world declare that they have now found Eve—and that the remains of the other female previously called ‘Eve’ are not in fact those of Eve. The question is, why hasn’t the name ‘Eve’ acquired a Millian content (namely, e), given the establishment of acquaintance-based uses? Why (in other words)

628

VI. Names and Descriptions

is it not simply false to say that Eve was discovered decades after anthropologists were convinced that they had found her? In response, I would simply deny that the uses in question are acquaintance-based in the relevant sense. For they are applied to e only because e is believed to satisfy a certain description: Wrst female human. Thus, the name is never really used nondescriptively. It is used to apply to an individual with whom the speakers have acquaintance, but this does not mean that the uses are genuinely acquaintance-based. For the application of a d-name to be genuinely acquaintance-based, the name must be applied without regard to whether the intended referent has the properties speciWed by the reference-Wxing description. What I mean by this is simply that application of the name to the intended referent (x) would not be regarded (by the speaker) as having been in error, were it discovered that x lacked the properties in question. It is of course possible that the speaker, upon discovery of such an error, would not be comfortable with either the idea that the application was in error or that it was not. What I think this would suggest would be that (at least in the idiolect of the speaker), the name was neither R-Descriptive nor Millian. Rather, it was somewhere in between. Thus, to echo a point made earlier (in s. 5), there are no doubt cases where there is simply no answer to the question: Is the d-name R-Descriptive or Millian?

Why Introduce D-Names? A Wnal criticism has been voiced by Robin Jeshion (this volume). Jeshion claims that the proposed view, along with other descriptive accounts of d-names, ‘fails to account for why the [d-name] was introduced in the Wrst place’ (Jeshion calls such names ‘descriptive names’). As she goes on to explain, any account of the sort in question, states that the descriptive name will be semantically equivalent to a certain deWnite description. There is, consequently, no semantic advantage gained by introducing the name. The same goes for [understanding such names]. One will be competent with the descriptive name just in case one is competent with a certain deWnite description . . . [and] all thoughts and attitudes with the descriptive name are identical to thoughts and attitudes with the corresponding deWnite description. So there is no cognitive advantage secured by introducing the name. The [view in question] gives descriptive names the same semantical and cognitive status as a deWnite description. Why, then, introduce the descriptive name?

My response to Jeshion’s concern is simple. D-names are introduced into the language for the same reason other expressions are so introduced—to enhance communicative eYciency. ‘Neptune’ was introduced to make talk of the hypothesized planet easier; the same goes for the term ‘bachelor’ which was no doubt introduced into the language to make talk of never-married adult human males easier. Such talk is easier in the sense that it makes communicating what one wants to communicate less arduous than it would otherwise be. Jeshion (this volume) considers a response similar to the foregoing, according to which the purpose of introducing d-names into the language is mere abbreviation. Such a view, according to Jeshion, imposes implausible constraints

Reimer, Descriptively Introduced Names

629

on the length and complexity of descriptive names. How, for instance, could such a view make sense of the perfectly natural introduction of ‘Geremiah the Gigolo’ as a descriptive name for Pam’s lover? Jeshion’s response may well have force against the view that abbreviation is the sole reason for introducing d-names. My view is, however, importantly diVerent. I claim that d-names are introduced into the language in order to enhance communicative eYciency, something which can be eVected not only by abbreviation, but by word choice as well. Thus, naming Pam’s lover ‘Geremiah the Gigolo’ communicates quite a bit—in particular, that the lover is a Xamboyant ‘ladies man’ whose demeanor is subject to ridicule. Such information is not communicated via the description ‘Pam’s lover’. There is thus a clear communicative advantage to introducing the d-name ‘Geremiah the Gigolo’. In the case of the d-names considered in this chapter, all are arguably introduced for communicative eYciency, but some of these are introduced for more than purposes of abbreviation. ‘Deep Throat’ communicates something about the nature and seediness of what the nominatum was accused of doing; ‘Jack the Ripper’ communicates something about the nature of the crimes committed by the nominatum; ‘Newman 1’ and ‘Oldman 1’ communicate something about the age of their respective nominata. Even ‘Julius’ communicates something: the gender of the nominatum. There is no case I can think of involving a d-name (or any other natural language expression for that matter) introduced into the language for purposes other than communicative eYciency.

8. Concluding Remarks I have proposed what amounts to a hybrid account of d-names. Such accounts are rather rare in philosophy of language. (Devitt (1981), who recommends a hybrid view, is an exception.) Why is that? Perhaps because it is only natural to suppose that what is true of ordinary names should be true of d-names. If ordinary names are Millian, then so are d-names. If ordinary names are descriptive, then so are d-names. But this is not plausible on reXection, as the arguments for the competing views of Millianism and R-Descriptivism seem to show. It appears to be based on a misguided quest for theoretical simplicity and uniformity: these virtues are pursued in spite of the linguistic data. Moreover—and no less importantly—the proposed account is eVectively uniWed by a simple account of knowledge of meaning, according to which a speaker ‘knows the meaning of ’ an expression e just in case he knows the semantic content of e in an acquaintance-based sense of ‘knows’. This contrasts rather sharply with the competing Millian and R-Descriptive accounts of d-names, neither of which (so far as I can see) appears consistent with any intuitively plausible notion of ‘knowing the meaning of ’ a linguistic expression.

References

Abbott, B. (1999). ‘Support for a Unique Theory of DeWnite Descriptions’, in T. Matthews and D. Strolovitch (eds.), Proceedings of the Ninth Conference on Semantics and Linguistic Theory (Ithaca NY: Cornell University), 1–15. —— (forthcoming). ‘DeWnites and IndeWnites’, in L. Horn and G. Ward (eds.), Handbook of Pragmatics (Oxford: Blackwell). Abney, S. (1987). ‘The English Noun Phrase in its Sentential Aspect’, Ph.D. dissertation, MIT. Abusch, D. (1994). ‘The Scope of IndeWnites’, Natural Language Semantics, 2: 83–136. —— and Rooth, M. (forthcoming). ‘Empty-Domain EVects for Presuppositional and NonPresuppositional Determiners’, in H. Kamp and B. Partee (eds.), Context-Dependence in the Analysis of Linguistic Meaning (Dordrecht: Elsevier). Almog J. (1984). ‘Semantical Anthropology’, Midwest Studies in Philosophy, 9: 479–90. —— (1998a). ‘The Subject Verb Object Class I’, Philosophical Perspectives, 12: 39–76. —— (1998b). ‘The Subject Verb Object Class II’, Philosophical Perspectives, 12: 77–104. —— (forthcoming). ‘The Language to Describe our Cognitive Lives’. Asher, N. (1993). Reference to Abstract Objects in Discourse (Dordrecht: Kluwer). —— and Lascarides, A. (1998a). ‘Questions in Dialogue’, Linguistics and Philosophy, 21: 237–309. —— and —— (1998b). ‘The Semantics and Pragmatics of Presupposition’, Journal of Semantics, 15: 239–300. Atlas, J. D. (1975). ‘Frege’s Polymorphous Concept of Presupposition and its Role in a Theory of Meaning’, Semantikos, 1: 29–44. —— (1988). ‘What are Negative Existence Statements About?’, Linguistics and Philosophy, 11: 373–94. —— (1989). Philosophy without Ambiguity, (Oxford: Clarendon Press). —— (1991a). ‘Aboutness and Quantifying into Intentional Contexts’, Symposium on Reference, Duisburg, Germany. —— (1991b). ‘Topic/Comment, Presupposition, Logical Form, and Focus Stress Implicatures: The Case of Focal Particles Only and Also’, Journal of Semantics, 8: 127–47. —— (forthcoming). ‘Presupposition’, in L. Horn and G. Ward (eds.), Handbook of Pragmatics, (Oxford: Blackwell). —— and Levinson, S. C. (1981). ‘It–clefts, Informativenesss, and Logical Form: Radical Pragmatics (Revised Standard Version)’, in P. Cole (ed.), Radical Pragmatics (New York: Academic Press), 1–61. Austin, J. (1950/1970). ‘Truth’, reprinted in his Philosophical Papers (Oxford: Clarendon Press), 117–33. —— (1962). How to Do Things with Words (Oxford: Clarendon Press). Bach, K. (1981). ‘Referential/Attributive’, Synthese, 49: 219–44. —— (1987/1994). Thought and Reference (Oxford: Clarendon Press).

References

631

—— (1994). ‘Conversational Impliciture’, Mind and Language, 9: 124–62. —— (1995). ‘Standardization vs. Conventionalization’, Linguistics and Philosophy, 18: 677–86. —— (1997). ‘Do Belief Reports Report Beliefs?’, PaciWc Philosophical Quarterly, 78: 215–41. —— (1999a). ‘The Semantics–Pragmatics Distinction: What it is and Why it Matters’, in K. Turner (ed.), The Semantics–Pragmatics Interface from DiVerent Points of View (Oxford: Elsevier), 65–84. —— (1999b). ‘The Myth of Conventional Implicature’, Linguistics and Philosophy, 22: 327–66. —— (2000). ‘QuantiWcation, QualiWcation and Context: A Reply to Stanley and Szabo´’, Mind and Language, 15: 262–83. —— (2001a). ‘You Don’t Say?’ Synthese, 128: 15–44. —— (2001b). ‘Speaking Loosely: Sentence Nonliterality’, Midwest Studies in Philosophy, 25: 249–63. —— (2002a). ‘Seemingly Semantic Intuitions’, in J. K. Campbell, M. O’Rourke, and D. Shier (eds.), Meaning and Truth (New York: Seven Bridges Press), 21–33. —— (2002b), ‘Giorgione was So-Called Because of his Name’, Philosophical Perspectives, 16: 73–103. —— and Harnish, R. (1979). Linguistic Communication and Speech Acts (Cambridge, Mass.: MIT Press). Barker, C. (1991). ‘Possessive Descriptions’, Ph.D. dissertation, University of California at Santa Cruz. —— (1993). ‘A Presuppositional Account of Proportional Ambiguity’, Proceedings of the Third Conference on Semantics and Linguistic Theory (Ithaca, NY: Cornell University Press), 1–18. —— (1996). ‘Presuppositions for Proportional QuantiWers’, Natural Language Semantics, 4: 237–59. Barker, S. J. (1997). ‘E-Type Pronouns, DRT, Dynamic Semantics and the QuantiWer/VariableBinding Model’, Linguistics and Philosophy, 20: 195–228. Barwise, J. (1988). The Situation in Logic (Stanford, Calif.: CSLI Publications). —— and Cooper, R. (1981). ‘Generalized QuantiWers and Natural Language’, Linguistics and Philosophy, 4: 159–219. —— and Perry, J. (1983). Situations and Attitudes (Cambridge, Mass.: MIT Press). Beaver, D. (1997). ‘Presupposition’, in J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language (Amsterdam: Elsevier), 939–1008. Bennett M. (1974). ‘An Extension and Variation in Montague Grammar’, Ph.D. dissertation, University of California at Los Angeles. Benthem, J. van (1987). ‘Semantic Automata’, in J. Groenendijk, D. de Jongh, and M. Stokhof (eds.), Studies in Discourse Representation Theory and the Theory of Generalized QuantiWers (Dordrecht: Foris), 1–25. Bertolet, R. (1980). ‘The Semantic SigniWcance of Donnellan’s Distinction’, Philosophical Studies, 37: 281–8. —— (1987). ‘Referring, Demonstrating, and Intending’, Philosophical Research Archives, 12: 251–60. Bezuidenhout, A. (1997). ‘Pragmatics, Semantic Underdetermination, and the Referential/Attributive Distinction’, Mind, 106: 375–410. Birner, B., and Ward, G. (1994). ‘Uniqueness, Familiarity, and the DeWnite Article in English’, Proceedings of the Berkeley Linguistics Society, 20: 93–102.

632

References

Blackburn, P., and Bos, J. (1999), Representation and Inference for Natural Language: A First Course in Computational Semantics. ‘Working with Discourse Representation Structures’, University of Saarland (MS). Blackburn, S. (1984). Spreading the Word: Groundings in the Philosophy of Language (Oxford: Clarendon Press). Blackburn, W. (1988). ‘Wettstein on DeWnite Descriptions’, Philosophical Studies, 53: 263–78. Blakemore, D. (1987). Semantic Constraints on Relevance (Oxford: Blackwell). —— (2002). Relevance and Linguistic Meaning: The Semantics and Pragmatics of Discourse Markers (Cambridge: Cambridge University Press). Boe¨r, S. E., and Lycan, W. G. (1986). Knowing Who (Cambridge, Mass.: MIT Press). Borg, Emma. (2000). ‘Complex Demonstratives’, Philosophical Studies, 97: 229–49. Bradley, F. H. (1883). Principles of Logic (London: Oxford University Press). Braun, David (1994). ‘Structured Characters and Complex Demonstratives’, Philosophical Studies, 74: 193–219. Breheny, R. (1999). ‘Context Dependence and Procedural Meaning: The Semantics of DeWnites’, Ph.D. thesis, University College, London. —— (2000). ‘Maximality, Negation and Plural DeWnites’, RCEAL, University of Cambridge (MS). —— (2002a). ‘The Current State of (Radical) Pragmatics in the Cognitive Sciences’, Mind and Language, 17: 169–87. —— (2002b). ‘Non–Dynamic Analyses of Anaphoric Pronouns: Do Things Look Better in 2-D?’, in G. Katz, S. Reinhard, and P. Reuter (eds.), Proceedings of Sinn und Bedeutung, 6. —— (2002c). ‘On the Dynamic Turn in the Study of Meaning and Interpretation’, in J. Peregrin (ed.), Meaning: The Dynamic Turn (Dordrecht: Elsevier). Brinton, A. (1977). ‘DeWnite Descriptions and Context Dependence’, Nouˆs, 11: 397–407. Burge, T. (1974). ‘Truth and Singular Terms’. Nouˆs, 8: 309–25. —— (1977). ‘Belief De Re’, Journal of Philosophy, 74: 338–62. Bu¨ring, D., and Hartmann, K. (1997). ‘The Kayne Mutiny’, in D. Beerman, D. LeBlanc, and H. van Riemsdijk (eds.), Rightward Movement (Amsterdam: John Benjamins), 59–80. Burton-Roberts, N. (1989). The Limits to Debate: A Revised Theory of Semantic Presupposition (Cambridge: Cambridge University Press). Cann, R., Kempson, R., Marten, L., and Otsuka, M. (2004). ‘On the Left and on the Right’ in D. Adger, C. de Cat, and G. Tsoulas (eds.), Peripheries (Dordrecht: Kluwer). Cappelen, H., and Lepore, E. (1997). ‘On an Alleged Connection between Indirect Quotation and Semantic Theory’, Mind and Language, 12: 278–96. —— (2002). ‘Indexicality, Binding, Anaphora and A Priori Truth’, Analysis, 62: 271–81. —— (forthcoming). ‘Does Meaning Determine Truth Conditions?’, in Z. Szabo´ (ed.), Semantics and Pragmatics (Oxford: Oxford University Press). Cardinaletti, A. (1994). ‘On the Internal Structure of Pronominal DPs’, Linguistic Review, 11: 195–219. —— and Starke, M. (1999). ‘The Typology of Structural DeWciency: A Case Study of the Three Classes of Pronouns’, in H. van Riemsdijk (ed.), Clitics in the Languages of Europe (Berlin: Mouton), 145–233.

References

633

Carlson, G. (1977a). ‘Reference to Kinds in English’, Ph.D. dissertation, University of Massachusetts, Amherst. —— (1977b). ‘A UniWed Analysis of the English Bare Plural’, Linguistics and Philosophy, 1: 413–57. Carminati, M., Frazier, L., and Rayner, K. (2002). ‘Bound Variables and C-Command’, Journal of Semantics, 19: 1–34. Carston, R. (1988). ‘Implicature, Explicature and Truth-Theoretic Semantics’, in R. Kempson (ed.), Mental Representations (Cambridge: Cambridge University Press), 155–81. —— (1993). ‘Conjunction, Explanation and Relevance’, Lingua, 90: 27–48. —— (1998). ‘The Semantics/Pragmatics Distinction: A View from Relevance Theory’, in K. Turner (ed.), The Semantics/Pragmatics Interface from DiVerent Points of View (Dordrecht: Elsevier), 85–125. —— (2002). Thoughts and Utterances: The Pragmatics of Explicit Communication (Oxford: Blackwell). Chafe, W. (1976). ‘Givenness, Contrastiveness, DeWniteness, Subjects, Topics, and Points of View’, in C. Li (ed.), Subject and Topic (New York: Academic Press), 27–55. Chao, W. (1983). ‘Pseudo–Anaphora’, University of Massachusetts, Amherst (MS). Chastain, C. (1975). ‘Reference and Context’, in K. Gunderson (ed.), Minnesota Studies in the Philosophy of Science, vii. Language, Mind, and Knowledge (Minneapolis: University of Minnesota Press), 194–269. Chierchia G. (1992). ‘Anaphora and Dynamic Binding’, Linguistics and Philosophy, 15: 111–83. —— (1995). Dynamics of Meaning: Anaphora, Presupposition, and the Theory of Grammar (Chicago: University of Chicago Press). Chomsky, N. (1957). Syntactic Structures (The Hague: Mouton). —— (1964) Current Issues in Linguistic Theory (The Hague: Mouton). —— (1965). Aspects of the Theory of Syntax (Cambridge, Mass.: MIT Press). —— (1972). ‘Deep Structure, Surface Structure, and Semantic Interpretation’, in H. Brekle and P. LuelsdorV (eds.), Studies in Semantics and Generative Grammar (The Hague: Mouton), 62–119. —— (1975). ReXections on Language (New York: Pantheon). —— (1976). ‘Conditions on Rules of Grammar’, Linguistic Analysis, 2: 303–51. —— (1980). Rules and Representations (New York: Columbia University Press). —— (1981). Lectures on Government and Binding (Dordrecht: Foris). —— (1986). Knowledge of Language (New York: Praeger). —— (1995). The Minimalist Program (Cambridge, Mass.: MIT Press). —— (2000). New Horizons in the Study of Language and Mind (Cambridge: Cambridge University Press). Christopherson, P. (1939). The Articles: A Study of their Theory and Use in English (Copenhagen: Munksgaard). Clark, H. H. (1975). ‘Bridging’, in R. C. Schank and B. L. Nash-Webber (eds.), Theoretical Issues in Natural Language Processing (New York: Association for Computing Machinery). Cohen, L. J. (1971). ‘Some Remarks of Grice’s views about the Logical Particles of Language’, in Y. Bar-Hillel (ed.), Pragmatics of Natural Languages (Dordrecht: Reidel), 50–68.

634

References

Cohen, S. (1999). ‘Contextualism, Skepticism, and the Structure of Reasons’, Philosophical Perspectives, 13: 57–89. Cooper, R. (1979). ‘The Interpretation of Pronouns’, in F. Heny and H. Schnelle (eds.), Syntax and Semantics, 10. Selections from the Third Gro¨ningen Round Table (New York: Academic Press), 61–92. —— (1983). QuantiWcation and Syntactic Theory (Dordrecht: Reidel). —— (1996). ‘The Role of Situations in Generalized QuantiWers’, in S. Lappin (ed.), Handbook of Contemporary Semantic Theory (Oxford: Blackwell), 65–86. —— (1998). ‘Mixing Situation Theory and Type Theory to Formalize Information States in Dialogue Exchanges’, Paper presented at the 2nd Workshop on Formal Semantics and Pragmatics of Dialogue, University of Twente, the Netherlands, 13–15 May. Copestake, A., Flickinger, D., and Sag, I. (1999). ‘Minimal Recursion Semantics: An Introduction’, Stanford University (MS). Cormack, A. and Kempson, R. (1990). ‘On SpeciWcity’, in J. GarWeld and M. Kiteley (eds.), Meaning and Truth (New York: Paragon House), 546–82. Cresti, D. (1995). ‘IndeWnite Topics’, Ph.D. dissertation, MIT. Creswell, M. (1973). Logics and Languages (London: Methuen). —— (2002). ‘Static Semantics for Dynamic Discourse’, Linguistics and Philosophy, 25: 545–71. Crummins, M. (1992). Talk about Belief (Cambridge, Mass.: MIT Press). —— and Perry, J. (1989). ‘The Prince and the Phone Booth: Reporting Puzzling Beliefs’, Journal of Philosophy, 86: 685–711. Dalrymple, M., Shieber, S., and Pereira, F. (1991). ‘Ellipsis and Higher-Order UniWcation’, Linguistics and Philosophy, 14: 399–452. Davidson, D. (1980a). ‘The Logical Form of Action Sentences’, in Essays on Actions and Events (Oxford: Clarendon Press), 105–22. —— (1980b). ‘Reply to Cargile’, in Essays on Actions and Events (Oxford: Clarendon Press), 137–48. —— (1984). Inquiries into Truth and Interpretation (Oxford: Clarendon). —— (1986). ‘A Nice Derangement of Epitaphs’, in E. Lepore (ed.), Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson (Oxford: Blackwell), 433–46. Davies, M. (1981). Meaning, QuantiWcation, Necessity (London: Routledge & Kegan Paul). —— and Humberstone, L. (1980). ‘Two Notions of Necessity’, Philosophical Studies, 38: 1–30. Dekker, P. (1993). ‘Existential Disclosure’, Linguistics and Philosophy, 16: 561–87. —— (1994). ‘Predicate Logic with Anaphora (Seven Inch Version)’, in L. Santelmann and M. Harvey (eds.), Proceedings of the Fourth Conference on Semantics and Linguistic Theory (Ithaca, NY: DMLL Publications, Cornell University). —— (2001). ‘Dynamics and Pragmatics of ‘‘Peirce’s Puzzle’’ ’, Journal of Semantics, 18: 211–41. —— (2002a). ‘Meaning and Use of IndeWnite Expressions’, Journal of Logic Language and Information, 11/2: 141–94. —— (2002b). ‘Pronouns in a Pragmatic Semantics’, Journal of Pragmatics, 34: 815–27. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research, 52: 913–29. —— (1995). ‘Solving the Skeptical Problem’, Philosophical Review, 104: 1–52. Devitt, M. (1974). ‘Singular Terms’, Journal of Philosophy, 71: 183–205.

References

635

—— (1981a). Designation (New York: Columbia University Press). —— (1981b). ‘Donnellan’s Distinction’, Midwest Studies in Philosophy, 6: 511–24. —— (1985). ‘Critical Notice of Evans 1982’, Australasian Journal of Philosophy, 63: 216–32. —— (1996). Coming to our Senses (Cambridge: Cambridge University Press). —— (1997a). ‘Meanings and Psychology: A Response to Mark Richard’, Nouˆs, 31: 115–31. —— (1997b). ‘Responses to the Maribor Papers’, in D. Jutronic (ed.), The Maribor Papers in Naturalized Semantics (Maribor: Pedagoska Fakulteta), 353–411. —— (2001). ‘A Shocking Idea about Meaning’, Revue Internationale de Philosophie, 208: 449–72. —— and Sterelny, K. (1999). Language and Reality: An Introduction to the Philosophy of Language (2nd edn. Oxford: Blackwell). Diesing, M. (1992). IndeWnites (Cambridge, Mass.: MIT Press). Donnellan, K. (1966). ‘Reference and DeWnite Descriptions’, Philosophical Review, 75: 281–304. —— (1968). ‘Putting Humpty Dumpty Together Again’, Philosophical Review, 77: 203–15. —— (1972). ‘Proper Names and Identifying Descriptions’, in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht: Reidel), 356–79. —— (1974). ‘Speaking of Nothing’, Philosophical Review, 83: 3–32. —— (1978). ‘Speaker Reference, Descriptions and Anaphora’, in P. Cole (ed.), Syntax and Semantics, ix. Pragmatics (New York: Academic Press), 47–68. —— (1979). ‘The Contingent A Priori and Rigid Designators’, in P. French, T. Uehling, and H. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language (Minneapolis: University of Minnesota Press), 45–60. —— (1981). ‘Intuitions and Presuppositions’, in P. Cole (ed.), Radical Pragmatics (New York: Academic Press), 129–42. Dowty, D., and Jacobson, P. (1989). ‘Agreement as a Semantic Phenomenon’, Proceedings of the Eastern States Conference on Linguistics (ESCOL’88) (Columbus: The Ohio State University), 95–108. —— Wall, R., and Peters, S. (1981). Introduction to Montague Semantics (Dordrecht: Reidel). Drier, J. (1990). ‘Internalism and Speaker Relativism’, Ethics, 101: 6–26. Du Bois, J. (1987). ‘The Discourse Basis of Ergativity’, Language, 63: 805–55. Dummett, M. (1967). ‘Truth’, in P. Strawson (ed.), Philosophical Logic (Oxford: Oxford University Press), 49–68. —— (1973). Frege: Philosophy of Language (London: Duckworth). Dylan, B. (1965). ‘Just like Tom Thumb’s Blues’, RCA Records. Eaker, E. (2002). ‘Reporting Attitudes’, Ph.D. dissertation, University of California at Los Angeles. Elbourne, P. (2001). ‘E-type Anaphora as NP-deletion’, Natural Language Semantics, 9: 241–88. Enc¸, M. (1986). ‘Towards a Referential Analysis of Temporal Expressions’, Linguistics and Philosophy, 9: 405–26. Engdahl, E. (1986). Constituent Questions (Dordrecht: Reidel). Epstein, R. (2000). ‘Roles and Non-Unique DeWnites’, in S. Chang, L. Liaw, and J. Ruppenhofer (eds.), Proceedings of the Twenty-Wfth Annual Meeting of the Berkeley Linguistics Society (Berkeley, Calif.: Berkeley Linguistics Society), 122–33.

636

References

Erk, K., Koller, A., and Niehren, J. (2002). ‘Processing UnderspeciWed Semantic Representations in the Constraint Language for Lambda Structures’, Journal of Language and Computation, 1: 449–70. Erteschik-Shir, N. (1997). The Dynamics of Focus Structure (Cambridge: Cambridge University Press). Evans, G. (1973). ‘The Causal Theory of Names’, Aristotelian Society Supplementary Volume, 47: 187–208. —— (1976) ‘Semantic Stucture and Logical Form’, reprinted in G. Evans (1985) Collected Papers (Oxford: Clarendon Press), 49–75. —— (1977). ‘Pronouns, QuantiWers and Relative Clauses (I)’, Canadian Journal of Philosophy, 7: 467–536. —— (1979). ‘Reference and Contingency’, The Monist, 62: 161–89. —— (1980). ‘Pronouns’, Linguistic Inquiry, 11: 337–62. —— (1982). The Varieties of Reference, ed. J. McDowell (Oxford: Clarendon Press). —— (1985). Collected Papers (Oxford: Clarendon Press). Farkas, D. (1981). ‘QuantiWer Scope and Syntactic Islands’, in R. Hendrick, C. Masek, and M. Miller (eds.), Chicago Linguistic Society Proceedings, 17: 59–66. —— (1997). ‘Evaluation Indices and Scope’, in A. Szabolcsi (ed.), Ways of Scope Taking (Dordrecht: Reidel), 183–216. —— and Giannakidou, A. (1996). ‘How Clause-Bounded is the Scope of Universals?’, Santa Cruz (MS). Fernando, T. (2000). ‘Between Proofs and Models: Presupposition’, Dublin (MS). Fiengo, R., and May, R. (1994). Indices and Identity (Cambridge, Mass.: MIT Press). von Fintel, Kai (2001a). ‘Evidence for Presuppositional IndeWnites’ (MS). —— (2001b). ‘What is Presupposition Accommodation?’ (MS). Fodor, J. A. (1975). The Language of Thought (New York: Thomas Crowell). —— (1981). Representations (Cambridge, Mass.: MIT Press). —— (1983). Modularity of Mind (Cambridge, Mass.: MIT Press). —— (1987). Psychosemantics (Cambridge, Mass.: MIT Press). —— (1990). A Theory of Content and Other Essays (Cambridge, Mass.: MIT Press). —— (2001). ‘Language, Thought, and Compositionality’, Mind and Language, 16: 1–15. —— and Lepore, E. (1996). ‘What Cannot be Valued, Cannot Be Valued, and it Cannot be Supervalued Either’, Journal of Philosophy, 93: 516–35. —— and—— (2001). The Compositionality Papers (Oxford: Clarendon Press). Fodor, J. D. (1979). ‘In Defense of the Truth Value Gap’, Syntax and Semantics, 11: 199–224. —— and Sag, I. (1982). ‘Referential and QuantiWcational IndeWnites’, Linguistics and Philosophy, 5: 355–98. Foley, W. (1997). Anthropological Linguistics (Oxford: Blackwell). Forbes, G. (1989). Languages of Possibility (Oxford: Blackwell). Fraurud, K. (1990). ‘DeWniteness and the Processing of NP’s in Natural Discourse’, Journal of Semantics, 7: 395–433. Frege, G. (1892/1970). ‘On Sense and Reference’, in P. Geach and M. Black (eds.), Translations from the Philosophical Writings of Gottlob Frege (Oxford: Blackwell), 56–78.

References

637

—— (1918). Der Gedanke: Eine logische Untersuchung, Beitra¨ge zur Philosophie des deutschen Idealismus 2, 1: 58–77. Gamut, L. (1991). Logic, Language, and Meaning, Volume 2 Intensional Logic and Logical Grammar (Chicago: University of Chicago Press). Garrod, S., and Sanford, A. (1994). ‘Resolving Sentences in a Discourse Context: How Discourse Representation AVects Language Understanding’, in M. Gernsbacher (ed.), Handbook of Psycholinguistics (San Diego, Calif.: Academic Press), 675–97. Gazdar, G. (1979). Pragmatics: Implicature, Presupposition, and Logical Form (New York: Academic Press). Geach, P. (1962). Reference and Generality: An Examination of Some Medieval and Modern Theories (Ithaca, NY: Cornell University Press). Geurts, B. (1995). Presupposing, Ph.D. dissertation, University of Stuttgart. —— (1998). ‘Dynamic Semantics vs. DRT’, Zeitschrift fu¨r Sprachwissenschaft, 16: 209–26. —— (1999). Presuppositions and Pronouns (Amsterdam: Elsevier). —— (2002). ‘Donkey Business’, Linguistics and Philosophy, 25: 129–56. Givon, T. (1984). Syntax: A Functional Typological Introduction (Amsterdam: John Benjamins). Go¨del, K. (1944). ‘Russell’s Mathematical Logic’, in P. A. Schilpp (ed.), The Philosophy of Bertrand Russell (Evanston, Ill.: Northwestern University Press), 125–53. Goodman, N. (1961). ‘About’, Mind, 70: 1–24. GraV, D. (2001). ‘Descriptions as Predicates’, Philosophical Studies, 102: 1–42. Grayling, A. C. (1995) ‘Perfect Speaker Theory’, in J. Hill and P. Kota´tko (eds.), Studies in Reference and Meaning (Prague: Filosoﬁa). Green, M. (1995). ‘Quantity, Volubility, and Some Varieties of Discourse’, Linguistics and Philosophy, 18: 83–112. Grice, H. P. (1957). ‘Meaning’, Philosophical Review, 66: 377–88. —— (1961). ‘The Causal Theory of Perception’, Aristotelian Society Supplementary Volume, 35: 121–52. —— (1969). ‘Vacuous Names’, in D. Davidson and J. Hintikka (eds.), Words and Objections: Essays on the Work of W. V. Quine (Dordrecht: Reidel), 118–45. —— (1975). ‘Logic and Conversation’, in J. Cole and J. Morgan (eds.), Syntax and Semantics, 3 Speech Acts (New York: Academic Press). —— (1978). ‘Further Notes on Logic and Conversation’, in P. Cole (ed.), Syntax and Semantics, 9 Pragmatics (New York: Academic Press), 113–27. —— (1981). ‘Presupposition and Conversational Implicature’, in P. Cole (ed.), Radical Pragmatics (New York: Academic Press), 183–98. —— (1989), Studies in the Way of Words (Cambridge, Mass.: Harvard University Press). Groenendijk, J., and Stokhof, M. (1990). ‘Dynamic Montague Grammar’, in L. Ka´lma´n and L. Po´los (eds.), Papers from the Second Symposium on Logic and Language (Budapest: Akade´miai Kiado´), 3–48. —— and—— (1991). ‘Dynamic Predicate Logic’, Linguistics and Philosophy, 14: 39–100. —— —— and Veltman, F. (1996). ‘Coreference and Modality’, in S. Lappin (ed.), The Handbook of Contemporary Semantic Theory (Oxford: Blackwell), 179–213. Grosz, B., and Sidner, C. (1986). ‘Attention, Intentions and the Structure of Discourse’, Computational Linguistics, 12: 175–204.

638

References

Gundel, J. (1977). Role of Topic and Comment in Linguistic Theory (Bloomington, Ind.: Indiana University Linguistics Club). —— Hedberg, N., and Zacharski R. (1993). ‘Cognitive Status and the Form of Referring Expressions in Discourse’, Language, 69: 274–307. Haack, S. (1996). Deviant Logic, Fuzzy Logic: Beyond the Formalism (Chicago: Chicago University Press). Hahnemann, C. and Neale, S. (forthcoming). ‘Filling the gap’. Hajicˇova´, E. (1984). ‘Presupposition and Allegation Revisited’, Journal of Pragmatics, 8: 155–67. Hardt, D. (1993). ‘Verb Phrase Ellipsis: Form, Meaning and Processing’, Ph.D. dissertation, University of Pennsylvania. Heim, I. (1982). ‘The Semantics of DeWnite and IndeWnite Noun Phrases’, Ph.D. dissertation, University of Massachusetts at Amherst. —— (1983a). ‘File Change Semantics and the Familiarity Theory of DeWnites’, in R. Ba¨uerle, R. Schwarze, and A. von Stechow (eds.), Meaning, Use, and Interpretation of Language (Berlin: de Gruyter). —— (1983b). ‘On the Projection Problem for Presuppositions’, Proceedings of West Coast Conference in Formal Linguistics. 2 (Stanford University), 114–25. —— (1985). ‘Direct Reference Explained Away’ (MS). —— (1988). The Semantics of DeWnite and IndeWnite Noun Phrases (New York: Garland). —— (1990a). ‘E-Type pronouns and Donkey Anaphora’, Linguistics and Philosophy, 13: 137–78. —— (1990b). ‘Presupposition Projection’, in R. van der Sandt (ed.), Presupposition, Lexical Meaning and Discourse Processes: Workshop Reader (Nijmegen: University of Nijmegen Press). —— (1991). ‘Artikel und DeWnitheit’, in A. von Stechow and D. Wunderlich (eds.), Semantics: An Interdisciplinary Handbook of Contemporary Research (Berlin: de Gruyter). —— (1992). ‘Presupposition Projection and the Semantics of Attitude Verbs’, Journal of Semantics, 9: 183–221. —— and Kratzer, A. (1998). Semantics in Generative Grammar (Oxford: Blackwell). Hepple, M. (1990). ‘The Grammar and Processing of Order and Dependency: A Categorial Approach’, Ph.D. dissertation, University of Edinburgh. Higginbotham, J. (1983a). ‘Logical Form, Binding, and Nominals’, Linguistic Inquiry, 14: 395–420. —— (1983b). ‘The Logic of Perceptual Reports: An Extensional Alternative to Situation Semantics’, Journal of Philosophy, 80: 100–27. —— (1985). ‘On Semantics’, Linguistic Inquiry, 16: 547–93. —— (1987). ‘IndeWniteness and Predication’, in E. Reuland and A. ter Meulen (eds.), The Representation of (In)deWniteness (Cambridge, Mass.: MIT Press), 43–70. —— (1988). ‘Contexts, Models, and Meanings: A Note on the Data of Semantics’, in R. Kempson (ed.), Mental Representation: The Interface between Language and Reality (Cambridge: Cambridge University Press), 29–48. —— and May, R. (1981). ‘Questions, QuantiWers, and Crossing’, Linguistic Review, 1: 41–80. Hintikka, J. (1986). ‘Logic of Conversation as a Logic of Dialogue’, in R. Grandy and R. Warner (eds.), Philosophical Grounds of Rationality (Oxford: Clarendon Press, 259–76).

References

639

—— and Sandu, G. (1997). ‘Game-theoretical Semantics’, in J. Benthem and A. ter Meulen (eds.), The Handbook of Logic and Language (Dordecht: Elsevier), 361–410. Hoji, H. (1986). ‘Scope Interpretation in Japanese and its Theoretical Implications’, Proceedings of the West Coast Conference in Formal Linguistics, 5: 85–101. Horn, L. (1986). ‘Presupposition, Theme and Variations’, Chicago Linguistic Society, 22: 168–92. —— (1989). A Natural History of Negation (Chicago: University of Chicago Press). —— (1996). ‘Presupposition and Implicature’, in S. Lappin (ed.), The Handbook of Contemporary Semantic Theory (Oxford: Blackwell), 299–319. Hornsby, J. (1977). ‘Singular Terms in Contexts of Propositional Attitude’, Mind, 86: 31–48. Hossack, K. (2000). ‘Plurals and Complexes’, British Journal for Philosophy of Science, 51/3: 411–43. Huang, C.-T. (1995). ‘Logical Form’, in G. Webelhuth (ed.), Government and Binding Theory and the Minimalist Program (Oxford: Blackwell), 125–75. Husserl, E. (1900/01). Logical Investigations, tr. J. N. Findley (London: Routledge & Kegan Paul). Israel, D., and Perry, J. (1990). ‘What is Information?’, in P. Hanson (ed.), Information, Language and Cognition (Vancouver: University of British Columbia Press), 1–19. JackendoV, R. (1972). Semantic Interpretation in Generative Grammar (Cambridge, Mass.: MIT Press). Jackson, E. G. (1994). ‘Negative Polarity, DeWnites under QuantiWcation and General Statements’, Ph.D. dissertation, Stanford University. Jacobson, P. (1999). ‘Towards a Variable-Free Semantics’, Linguistics and Philosophy, 22: 117–84. Jeshion, R. (2001a). ‘Donnellan on Neptune’, Philosophy and Phenomenological Research, 63: 111–35. —— (2001b). ‘Acquaintanceless De Re Belief ’, in J. Campbell, M. O’Rourke and D. Shier (eds.), Topics in Contemporary Philosophy: Truth and Meaning (New York: Seven Bridges Press), 53–78. —— (2003). ‘Farewell Acquaintance’ (MS). Kadmon, N. (1987). ‘On Unique and Non–Unique Reference and Asymmetric QuantiWcation’, Ph.D. dissertation, University of Massachusetts, Amherst. —— (1990). ‘Uniqueness’, Linguistics and Philosophy, 13: 273–324. —— (2001). Formal Pragmatics: Semantics, Pragmatics, Presupposition, and Focus (Oxford: Blackwell). Kallmeyer, L., and Joshi, A. (2002). ‘Factoring Predicate Argument and Scope Semantics: UnderspeciWed Semantics in LTAG’, Journal of Language and Computation, 1: 395–448. Kamp, H. (1981). ‘A Theory of Truth and Semantic Representation’, in J. Groenendijk, T. Janssen, and M. Stokhof (eds.), Truth, Interpretation and Information (Dordrecht: Foris), 277–322. —— (1990). ‘Prolegomena to a Structural Account of Belief and Other Attitudes’, in C. Anderson and J. Owens (eds.), Propositional Attitudes: The Role of Content in Logic, Language and Mind (Stanford, Calif.: CSLI Publications), 27–90. —— and van Eijck, J. 1997. ‘Representing Discourse in Context’, in J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language (Amsterdam: Elsevier), 179–239.

640

References

Kamp, H., and Reyle, U. (1993). From Discourse to Logic: An Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic, and Discourse Representation Theory (Dordrecht: Kluwer). Kanazawa, M. (1994). ‘Weak vs. Strong Readings of Donkey Sentences and Monotonicity Inference in a Dynamic Setting’, Linguistics and Philosophy, 17: 109–58. —— (2001). ‘Singular Donkey Pronouns are Semantically Singular’, Linguistics and Philosophy, 24: 383–403. Kang, Y. E. Y. (1994). ‘Weak and Strong Interpretations of QuantiWers and DeWnite NPs in English and Korean’, Ph.D. dissertation, University of Texas, Austin. Kaplan, D. (1969). ‘Quantifying in’, in D. Davidson and J. Hintikka (eds.), Words and Objections (Dordrecht: Reidel), 206–42. —— (1979). ‘Dthat’, in P. French, T. Uehling, and H. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language (Minneapolis: University of Minnesota Press), 383–400. —— (1989a). ‘Demonstratives’, in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan (Oxford: Oxford University Press), 481–563. —— (1989b). ‘Afterthoughts’, in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan (New York: Oxford University Press), 565–614. —— (forthcoming). ‘What is Meaning?’ Karttunen, L. (1969). ‘Pronouns and Variables’, Proceedings of the Chicago Linguistics Society, 5: 108–15. —— (1973). ‘Presuppositions of Compound Sentences’, Linguistic Inquiry, 4: 169–93. —— (1974). ‘Presupposition and Linguistic Context’, Theoretical Linguistics, 1: 181–93. —— (1976). ‘Discourse Referents’, in J. McCawley (ed.), Syntax and Semantics, 7 Notes from the Linguistic Underground (New York: Academic Press), 363–86. —— and Peters, S. (1979). ‘Conventional Implicature’, in C. Oh and D. Dinneen (eds.), Syntax and Semantics 11 Presupposition (New York: Academic Press), 1–56. Katz, J. J. (1987). ‘Common Sense in Semantics’, in E. LePore (ed.), New Directions in Semantics (New York: Academic Press), 157–233. —— and Postal, P. (1964). An Integrated Theory of Linguistic Descriptions (Cambridge, Mass.: MIT Press). Keenan, E., and Stavi, Y. (1986). ‘A Semantic Characterization of Natural Language Determiners’, Linguistics and Philosophy, 9: 253–326. Kempson, R. (1975). Presupposition and the Delimitation of Semantics (Cambridge: Cambridge University Press). —— (1986). ‘DeWnite NPs and Context–Dependence: A UniWed Theory of Anaphora’, in C. Travis (ed.), Meaning and Interpretation (Oxford: Blackwell), 269–39. —— (1988). ‘Logical Form: The Grammar Cognition Interface’, Journal of Linguistics, 24: 393–432. —— (1996). ‘Semantics, Pragmatics and Natural Language Interpretation’, in S. Lappin (ed.), Handbook of Contemporary Semantic Theory (Oxford: Blackwell), 561–98. —— and Cormack, A. (1981). ‘Ambiguity and QuantiWcation’, Linguistics and Philosophy, 4: 259–310. —— Meyer-Viol, W., and Gabbay, D. (1999). ‘Towards a Dynamic Structural Account of VP Ellipsis’, in S. Lappin and E. Benmamoun (eds.), Fragments: Studies in Ellipsis and Gapping (Oxford: Oxford University Press), 227–90.

References

641

—— —— and —— (2001). Dynamic Syntax: The Flow of Language Understanding (Oxford: Blackwell). Kim, J. (1977). ‘Perception and Reference without Causality’, Journal of Philosophy, 74: 606–20. King, J. (1988). ‘Are IndeWnite Descriptions Ambiguous?’, Philosophical Studies, 53: 417–40. —— (2001). Complex Demonstratives: A QuantiWcational Account (Cambridge, Mass.: MIT Press). Kohlhase, M., Kuschert, S., and Pinkal, M. (2000). ‘A Type-Theoretic Semantics for Lambda DRT’, in P. Dekker and M. Stokhof (eds.), 10th Amsterdam Colloquium (Amsterdam: ILLC, University of Amsterdam), part II, 479–98. Ko¨nig, J.-P., and Mauner, G. (2000). ‘A-deWnites and the Discourse Status of Implicit Arguments’, Journal of Semantics, 16: 207–36. Koster, J., and Reuland, E. (1991). Long-Distance Anaphora (Cambridge: Cambridge University Press). Kratzer, A. (1989). ‘An Investigation of the Lumps of Thought’, Linguistics and Philosophy, 12: 607–53. —— (1995). ‘Stage-Level and Individual-Level Predicates’, in G. Carlson and J. Pelletier (eds.), The Generic Book (Chicago: Chicago University Press). Krifka, M. (1998). ‘Pragmatic Strengthening in Plural Predications and Donkey Sentences’, Proceedings of the Eighth Conference on Semantics and Linguistic Theory (Ithaca, NY: Cornell University Press). Kripke, S. (1974). ‘First General Discussion Section’, Synthese, 27: 471–508. The discussion in which Kripke took part was published as part of a special issue of this journal on ‘Intentionality, Language and Translation’, ed. J. Troyer and S. Wheeler. —— (1977). ‘Speaker Reference and Semantic Reference’, in P. French, T. Uehling, and H. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language (Minneapolis: University of Minnesota Press), 6–27. —— (1979). ‘A Puzzle about Belief ’, in A. Margalit (ed.), Meaning and Use (Dordrecht: Reidel), 239–83. —— (1980). Naming and Necessity (Cambridge, Mass.: Harvard University Press). —— (1988). ‘A Puzzle about Belief ’, in N. Salmon and S. Soames (eds.), Propositions and Attitudes (Oxford: Oxford University Press), 102–48. Kuroda, S. Y. (1982). ‘Indexed Predicate Calculus’, Journal of Semantics, 1: 43–59. Ladusan, W. (1980). ‘Affective or Factive Verbs and Negative-Polarity Items’, Papers from the Regional Meetings of the Chicago Linguistic Society, 16: 170–84. Lambert, K. (1983). Meinong and the Principle of Independence (Cambridge: Cambridge University Press). —— (2001). ‘Free Logics’, in L. Goble (ed.), The Blackwell Guide to Philosophical Logic (Oxford: Blackwell), 258–79. Lambrecht, K. (1994). Information Structure and Sentence Form: Topic, Focus and the Mental Representations of Discourse Referents (Cambridge: Cambridge University Press). Landman, F. (1986). Towards a Theory of Information: The Status of Partial Objects in Semantics (Dordrecht: Foris). Lappin, S. (1989). ‘Donkey Pronouns Unbound’, Theoretical Linguistics, 15: 263–86. —— and Francez, N. (1994). ‘E-Type Pronouns, I-Sums, and Donkey Anaphora’, Linguistics and Philosophy, 17: 391–428.

642

References

Larson, R., and Segal, G. (1995). Knowledge of Meaning: An Introduction to Semantic Theory (Cambridge, Mass.: MIT Press). Lasersohn, P. (1993). ‘Existence Presuppositions and Background Knowledge’, Journal of Semantics, 10: 112–22. Lasnik, H. (1976) ‘Remarks on Coreference’, Linguistic Analysis, 2: 1–22. Lennon, J. and McCartney, P. (1966) Norwegian Wood., London: EMI. Lepore, E., and Ludwig, K. (2000). ‘The Semantics and Pragmatics of Complex Demonstratives’, Mind, 109: 199–240. Levinson, S. (2000). Presumptive Meanings: The Theory of Generalized Conversational Implicatures (Cambridge, Mass.: MIT Press). Lewis, D. (1970). ‘General Semantics’, Synthese, 22: 18–67. —— (1973). Counter factuals (Cambridge, Mass.: Harvard University Press). —— (1975). ‘Adverbs of QuantiWcation’, in E. Keenan (ed.), Formal Semantics of Natural Language (Cambridge: Cambridge University Press), 3–15. —— (1983a). ‘Individuation by Acquaintance and by Stipulation’, Philosophical Review, 92: 3–32. —— (1983b). ‘Scorekeeping in a Language Game’, in Philosophical Papers: volume 1 (Oxford: Oxford University Press), 233–49. —— (1988). ‘Statements Partly about Observation’, Philosophical Papers, 17: 1–31. —— (1996). ‘Elusive Knowledge’, Australasian Journal of Philosophy, 74: 549–67. Link, G. (1983). ‘The Logical Analysis of Plurals in Mass Terms: A Lattice-Theoretical Approach’, in R. Ba¨uerle, C. Schwarze, and A. von Stechow (eds.), Meaning, Use, and Interpretation of Language (Berlin: de Gruyter). Lo¨bner, S. (1987), ‘DeWnites’. Journal of Semantics, 4: 279–326. Ludlow, P. (1991). ‘Review of ter Meulen and Reuland (eds.), The Representation of (In)deWniteness’, Journal of Semantics, 8: 277–86. —— (1994). ‘Conditionals, Events, and Unbound Pronouns’, Lingua e stile, 29: 165–83. —— (1999). Semantics, Tense, and Time: An Essay in the Metaphysics of Natural Language (Cambridge, Mass.: MIT Press). —— (2002). ‘LF and Natural Logic’, in G. Preyer and G. Peter (eds.), Logical Form and Language (Oxford: Oxford University Press), 132–68. —— and Neale, S. (1991). ‘IndeWnite Descriptions: In Defense of Russell’, Linguistics and Philosophy, 14: 171–202. —— and —— (forthcoming). ‘Descriptions’, in M. Devitt and R. Hanley (eds.) Handbook in the Philosophy of Language (Oxford: Blackwell). Lumsden, M. (1988). Existential Sentences: Their Structure and Meaning (London: Croom Helm). Lycan, W. (1984a). ‘A Syntactically Motivated Theory of Conditionals’, Midwest Studies in Philosophy, 9: 437–56. —— (1984b). Logical Form in Natural Language (Cambridge, Mass: MIT Press). Lyons, J. (1977). Semantics (2 vols. Cambridge: Cambridge University Press). McCawley, J. (1981). What Linguists Always Wanted to Know about Logic (But were Afraid to Ask) (Chicago: University of Chicago Press). McGinn, C. (1981). ‘The Mechanism of Reference’, Synthese, 49: 157–86.

References

643

McNally, L. (1992). ‘An Interpretation for the English Existential Construction’. Ph.D. dissertation, University of California, Santa Cruz, Dept. of Linguistics. —— (1998). ‘Existential Sentences without Existential QuantiWcation’, Linguistics and Philosophy, 21: 353–92. Maling, J. (1984). ‘Non-Clause Bounded ReXexives in Modern Icelandic’, Linguistics and Philosophy, 7: 211–41. May, R. (1985). Logical Form (Cambridge, Mass.: MIT Press). —— (2002). ‘Ellipsis’, in L. Nadel (ed.) Encyclopedia of Cognitive Science (Basingstoke: Palgrave Macmillan), 1094–102. Meinong, A. (1904/1960). ‘On the Theory of Objects’, tr. I. Levi, D. Terrell, and R. Chisholm, in R. Chisholm (ed.), Realism and the Background of Phenomenology (New York: Free Press) 76–117. Meulen, A. ter (1999), ‘Binding by Implicit Arguments’, in P. Dekker (ed.), Twelfth Amsterdam Colloquium (Amsterdam: ILLC, University of Amsterdam), 37–42. Mill, J. S. (1843). A System of Logic (London: Longmans). —— (1872). A System of Logic (deWnitive 8th edn. 1949 reprint London: Longmans, Green & Co.). Millican, P. (1990). ‘Content, Thought, DeWnite Descriptions’, Aristotelian Society Proceedings, supplementary volume, 64: 167–203. Milsark, G. (1974). ‘Existential Sentences in English’, Ph.D. dissertation, MIT. Milward, D. (1995). ‘Integrating Situations into a Theory of Discourse Anaphora’, in P. Dekker and M. Stokhof (eds.), Proceedings of the 10th Amsterdam Colloquium (Amsterdam: ILLC, University of Amsterdam), 519–38. Moltmann, F. (1996). ‘E-Type and Dynamic Approaches to Unbound Anaphoric Pronouns’, CUNY Graduate Center (MS). Montague, R. (1974a). ‘Pragmatics’, in R. Thomason (ed.), Formal Philosophy (New Haven: Yale University Press), 95–118. —— (1974b). ‘The Proper Treatment of QuantiWcation in Ordinary English’, in R. Thomason (ed.), Formal Philosophy (New Haven: Yale University Press), 247–70. Morrill, G. (1994). Type-Logical Grammar (Dordrecht: Kluwer). Neale, S. (1988). ‘Events and ‘‘Logical Form’’ ’, Linguistics and Philosophy, 11: 303–21. —— (1990). Descriptions (Cambridge: MIT Press). —— (1992). ‘Paul Grice and the Philosophy of Language’, Linguistics and Philosophy 15: 509–59. —— (1993). ‘Term Limits’, Philosophical Perspectives, 7: 89–124. —— (1994). ‘Logical Form and LF’, in C. Otero (ed.), Noam Chomsky: Critical Assessments (London: Routledge), 788–838. —— (1995). ‘The Philosophical SigniWcance of Go¨del’s Slingshot’, Mind, 104: 761–825. —— (1997). ‘Speaker’s Reference and Anaphora’, in D. Jutronic (ed.), The Maribor Papers in Naturalized Semantics (Maribor: Pedagoska Fakulteta), 202–14. —— (1999). ‘Coloring and Composition’, in K. Murasugi and R. Stainton (eds.), Philosophy and Linguistics (Boulder, Colo.: Westview Press), 35–84. —— (2000). ‘On Being Explicit: Comments on Stanley and Szabo´, and on Bach’, Mind and Language, 15: 284–94. —— (2001) Facing Facts (Oxford: Clarendon Press).

644

References

Neale, S. (2002) ‘Radical Contextualism Means Radical Indeterminacy Abstract Submitted to the 4th European Congress of Analytic Philosophy, Lund, June 2002. —— (forthcoming a). ‘Interpreting Pronouns’, in Z. Szabo´ (ed.), The Semantics–Pragmatics Distinction (Oxford: Oxford University Press). —— (forthcoming b). Linguistic Pragmatism (Oxford: Clarendon Press). —— (forthcoming c). Descriptions, exp. edn. (Oxford: Clarendon Press). —— (forthcoming d). Pronouns and Variables (Oxford: Clarendon Press). Nelson, M. (1999). ‘Wettstein’s Incompleteness, Salmon’s Intuitions’, Nouˆs, 33: 573–89. Newmeyer, F. (1980). Linguistic Theory in America (New York: Academic Press). Nunberg, G. (1979). ‘The Non-Uniqueness of Semantic Solutions: Polysemy’, Linguistics and Philosophy, 3: 143–84. —— (1993). ‘Indexicality and Deixis’, Linguistics and Philosophy, 16: 1–43. Ockham, W. (1998). Summa Logicae. English tr. in A. Freddoso and H. Schuurman Ockham’s Theory of Propositions (South Bend, Ind.: St Augustine’s Press). Ojeda, A. (1993). Linguistic Individuals (Stanford, Calif.: CSLI Publications). Ostertag, G. (1998). ‘Introduction’, in G. Ostertag (ed.), DeWnite Descriptions: A Reader (Cambridge, Mass.: MIT Press), 1–33. —— (1999). ‘A Scorekeeping Error’, Philosophical Studies, 96: 123–46. Papafragou, A. (1998). ‘Modality and Semantic Indeterminacy’, in V. Rouchota and A. Jucker (eds.), Current Issues in Relevance Theory (Amsterdam: John Benjamins), 237–70. Parsons, T. (1978). ‘Pronouns as Paraphrases’. University of Massachusetts, Amherst (MS). —— (1990). Events in the Semantics of English (Cambridge, Mass.: MIT Press). Partee, B. (1973). ‘Some Structural Analogies between Tenses and Pronouns in English’, Journal of Philosophy, 70: 601–9. —— ter Meulen, A., and Wall, R. (1993). Mathematical Methods in Linguistics (Dordrecht, Kluwer). Peacocke, C. (1975). ‘Proper Names, Reference and Rigid Designation’, in S. Blackburn (ed.), Meaning, Reference and Necessity (Cambridge: Cambridge University Press), 109–32. Pelletier, J., and Schubert, L. (1989). ‘Generically Speaking’, in G. Chierchia, B. Partee, and R. Turner (eds.), Properties, Types and Meaning (Dordrecht: Kluwer). Percus, O. (1997). ‘Prying Open the Cleft’, Proceedings of Northeast Linguistic Society, 27, ed. K. Kusumoto (Amherst, Mass.: GLSA, University of Massachusetts), 337–51. —— (1998). ‘Presupposition Denial and the Origin of Antecedents’, in U. Sauerland and O. Percus (eds.), The Interpretive Tract: MIT Working Papers in Linguistics, 25: 45–62. Perry, J. (1977). ‘Frege on Demonstratives’, Philosophical Review, 86: 474–97. —— (1986). ‘Thought without Representation’, Proceedings of the Aristotelian Society, supplementary volume, 60: 137–51. —— (1993). The Essential Indexical and Other Essays (Oxford: Clarendon Press). —— (1998). ‘Indexicals, Contexts, and Unarticulated Constituents’, in A. Aliseda-Llera, R. van Glabbeek, and D. Westersta˚hl (eds.), Computing Natural Language: Context, Structure, and Processes (Cambridge: Cambridge University Press) —— (2001). Reference and ReXexivity (Stanford, Calif.: CSLI Publications). Poesio, M., and Vieira, R. (1997). ‘A Corpus-Based Investigation of DeWnite Description Use’, In cmp-lg e-print archive at http://xxx.lanl.gov, document cmp-lg/9710007. —— and Zucchi, S. (1992). ‘On Telescoping’, in C. Barker and D. Dowty (eds.), Proceedings of the Second Conference on Semantics and Linguistic Theory (Columbus, Ohio: The Ohio State University Press).

References

645

Portner, P., and Yabushita, K. (1998). ‘The Semantics and Pragmatics of Topic Phrases’, Linguistics and Philosophy, 21: 117–57. Postal, P. (1969). ‘On So-Called Pronouns in English’, in D. A. Reibel and S. A. Schane (eds.), Modern Studies in English: Readings in Transformational Grammar (Englewood CliVs, NJ: Prentice-Hall), 201–24. Putnam, H. (1973). ‘Meaning and Reference’, Journal of Philosophy, 70: 699–711. —— (1975). Mind, Language and Reality: Philosophical Papers ii, volume 2, (Cambridge: Cambridge University Press). Quine, W. V. O. (1940/1951). Mathematical Logic (Cambridge, Mass.: Harvard University Press). —— (1953). From a Logical Point of View (Cambridge, Mass.: Harvard University Press). —— (1960). Word and Object (Cambridge, Mass.: MIT Press). —— (1980). ‘On What There is’, in From a Logical Point of View (2nd edn. Cambridge, MA: Harvard University Press), 1–19. —— (1981). ‘Intensions Revisited’, in P. French, T. Uehling, and H. Wetstein (eds.), Contemporary Perspectives in the Philosophy of Language (Minneapolis: University of Minnesota Press), 268–74. Radford, A. (1997). Syntax: A Minimalist Introduction (Cambridge: Cambridge University Press). Ramchandran, M. (1993). ‘A Strawsonian Objection to Russell’s Theory of Descriptions’, Analysis 53: 209–1 —— (1995). ‘Bach on Behalf of Russell’, Analysis 55: 283–7. Ranta, A. (1994). Type-Theoretic Grammar (Oxford: Clarendon Press). Recanati, F. (1986/87). ‘Contextual Dependence and DeWnite Descriptions’, Proceedings of the Aristotelian Society, 87: 57–73. —— (1989a). ‘Referential/Attributive: A Contextualist Proposal’, Philosophical Studies, 56: 217–49. —— (1989b). ‘The Pragmatics of What is Said’, Mind and Language, 4: 295–329. —— (1993). Direct Reference: From Language to Thought (Oxford: Blackwell). —— (1994). ‘Contextualism and Anti-Contextualism in the Philosophy of Language’, in S. Tsohatzidis (ed.), Foundation of Speech Act Theory (London: Routledge), 156–66. —— (1996). ‘Domains of Discourse’, Linguistics and Philosophy, 19: 445–75. —— (1999). ‘Situations and the Structure of Content’, in K. Murasugi and R. Stainton (eds.), Philosophy and Linguistics (Boulder, Colo.: Westview), 113–65. —— (2000). Oratio Obliqua, Oratio Recta: An Essay on Metarepresentation (Cambridge, Mass.: MIT Press). Reichenbach, H. (1947). Elements of Symbolic Logic (London: Macmillan). Reimer, M. (1992). ‘Incomplete Descriptions’, Erkenntnis, 37: 347–63. —— (1998a). ‘QuantiWcation and Context’, Linguistics and Philosophy, 21: 95–115. —— (1998b). ‘The Wettstein/Salmon Debate: Critique and Resolution’, PaciWc Philosophical Quarterly, 79: 130–51. —— (1998c). ‘Donnellan’s Distinction/Kripke’s Test’, Analysis, 58: 89–100. Reinhart, T. (1978). ‘Syntactic Domains for Semantic Rules’, in F. Guenthner and S. J. Schmidt (eds.), Formal Semantics and Pragmatics for Natural Languages (Dordrecht: Reidel), 107–30.

646

References

Reinhart, T. (1981). ‘Pragmatics and Linguistics: An Analysis of Sentence Topics’, Philosophica, 27: 53–94. —— (1983). Anaphora and Semantic Interpretation (London: Croom Helm). —— (1995). ‘Interface Strategies’, OTS, Utrecht University (MS). —— (1997). ‘QuantiWer Scope: How Labor is Divided between QR and Choice Functions’, Linguistics and Philosophy, 20: 335–97. Reuland, E. (2001). ‘Anaphors, Logophors and Binding’, in P. Cole, G. Herman, and C.-T. J. Huang (eds). Syntax and Semantics, 33. Long Distance ReXexives (San Diego, CA: Academic Press), 343–70. —— and Meulen, A. ter (eds.) (1987). The Representation of (In)deWniteness (Cambridge, Mass.: MIT Press). Reyle, U. (1993). ‘Dealing with Ambiguities by UnderspeciWcation: Construction, Representation and Deduction’, Journal of Semantics, 10: 123–79. Richard, M. (1990). Propositional Attitudes: An Essay on Thoughts and How we Ascribe them (Cambridge: Cambridge University Press). —— (1993). ‘Articulated Terms’, Philosophical Perspectives, 7: 207–30. —— (1989). ‘Modal Subordination and Pronominal Anaphora in Discourse’, Linguistics and Philosophy, 12: 683–721. Roberts, C. (1989). ‘Modal Subordination and Pronomial Anaphora in Discourse’, Linguistics and Philosophy, 12: 683–721. —— (1990). Modal Subordination, Anaphora, and Distributivity (New York: Garland). —— (1995). ‘Domain Restriction in Dynamic Interpretation’, in E. Bach, E. Jelinek, A. Kratzer, and B. Partee (eds.), QuantiWcation in Natural Languages (Dordrecht: Kluwer), 661–700. —— (1996a). ‘Anaphora in Intensional Contexts’, in S. Lappin (ed.), The Handbook of Contemporary Semantic Theory (Oxford: Blackwell), 215–46. —— (1996b). ‘Information Structure: Towards an Integrated Theory of Formal Pragmatics’, in J. Yoon, and A. Kathol (eds.), OSU Working Papers in Linguistics, 49, Papers in Semantics, 91–136. —— (2002). ‘Demonstratives as DeWnites’, in K. Deemter and R. Kibble (eds.), Information Sharing (Stanford, Calif.: CSLI Publications), 89–136. —— (2003). ‘Uniqueness in DeWnite Noun Phrases’, Linguistics and Philosophy, 26: 287–350. —— (forthcoming). Information Structure in Discourse: Developing an Integrated Theory of Formal Pragmatics. Monograph in preparation. Rooth, M. (1987). ‘Noun Phrase Interpretation in Montague Grammar, File Change Semantics, and Situation Semantics’, in P. Ga¨rdenfors (ed.), Generalized QuantiWers (Dordrecht: Reidel), 237–68. —— (1992). ‘A Theory of Focus Interpretation’, Natural Language Semantics, 1: 75–116. Rooy, R. van (1997). ‘Attitudes and Changing Contexts’, Ph.D. dissertation, University of Stuttgart. —— (1999). ‘Anaphoric Relations Across Belief Contexts’, in K. von Heusinger and U. Egli (eds.), Reference and Anaphoric Relations (Dordrecht: Kluwer). —— (2001). ‘Exhaustivity in Dynamic Semantics: Referential and Descriptive Pronouns’, Linguistics and Philosophy, 24: 621–57. Rouchota, V. (1992). ‘On the Referential/Attributive Distinction’, Lingua, 87, 137–67. Russell, B. (1903). Principles of Mathematics (Cambridge: Cambridge University Press).

References

647

—— (1904). ‘Meinong’s Theory of Complexes and Assumptions’, Mind, 13: 204–19, 336–54, 509–24. —— (1905). ‘On Denoting’, Mind, 14: 479–93. —— (1917). ‘Knowledge by Acquaintance and Knowledge by Description’, in Mysticism and Logic (Garden City, NY: Doubleday Anchor), 202–24. —— (1918/1919). ‘The philosophy of logical atomism’, The Monist, 28: 495–527; 29: 32–63, 190–222, 345–80. —— (1919). Introduction to Mathematical Philosophy (London: George Allen & Unwin). —— (1956). ‘The Philosophy of Logical Atomism’, in R. C. Marsh (ed.), Logic and Knowledge: Essays 1901–1950 (London: George Allen & Unwin), 175–281. —— (1959). ‘Mr. Strawson on Referring’, in My Philosophical Development (London: Allen & Unwin), 238–45. —— (1988). ‘Knowledge by Acquaintance and Knowledge by Description’, in N. Salmon and S. Soames (eds.), Propositions and Attitudes (Oxford: Oxford University Press), 16–32. —— and Whitehead, A. (1910–13). Principia Mathematica (Cambridge: Cambridge University Press). Ruys, E. (1992). ‘The Scope of IndeWnites’, Ph.D. dissertation, University of Utrecht. Sag, I. (1976). ‘Deletion and Logical Form’, Ph.D. dissertation, MIT. Sainsbury, R. M. (1979). Russell (London: Routledge & Kegan Paul). —— (2002). Departing from Frege: Essays in Philosophical Logic (London: Routledge). Salmon, N. (1982). ‘Assertion and Incomplete DeWnite Descriptions’, Philosophical Studies, 42: 37–45. —— (1986). Frege’s Puzzle (Cambridge, Mass.: MIT Press). —— (1987/88). ‘How to Measure the Standard Meter’, Proceedings of the Aristotelian Society, 88: 193–217. —— (1989). ‘How to Become a Millian Heir’, Nouˆs, 23: 211–20. —— (1990). ‘A Millian Heir Rejects the Wages of Sinn’, in C. A. Anderson and J. Owens (eds.), Propositional Attitudes: The Role of Content in Logic, Language, and Mind (Stanford, Calif.: CSLI Publications), 215–47. —— (1991). ‘The Pragmatic Fallacy’, Philosophical Studies, 63: 83–97. —— (1993a). ‘Analyticity and Apriority’, in J. Tomberlin (ed.), Philosophical Perspectives, vii. Language and Logic (Atascadero, Calif.: Ridgeview), 125–34. —— (1993b). ‘Relative and Absolute Apriority’, Philosophical Studies, 69: 83–100. —— (1994). ‘Introduction to Part II: Sense and Reference’, in R. M. Harnish (ed.), Basic Topics in the Philosophy of Language (Englewood CliVs, NJ: Prentice-Hall), 99–129. —— (1995). ‘Being of Two Minds: Belief with Doubt’, Nouˆs, 29: 1–20. —— (1998). ‘Nonexistence’, Nouˆs, 32: 277–319. SchiVer, S. (1978). ‘The Basis of Reference’, Erkenntnis, 13: 171–206. —— (1995). ‘Descriptions, Indexicals, and Belief Reports: Some Dilemmas (But Not the Ones you Expect)’, Mind, 104: 107–31. —— (1997). ‘Descriptions, Indexicals, and Belief Reports: Some Dilemmas’, in W. Ku¨nne, A. Newen, and M. Anduschus (eds.), Direct Reference, Indexicality, and Propositional Attitudes (Stanford, Calif.: CSLI Publications), 247–75. Schwarzschild, R. (2002). ‘Singleton IndeWnites’, Journal of Semantics, 19: 289–314. Searle, J. (1969). Speech Acts (Cambridge: Cambridge University Press).

648

References

Searle, J. (1979a). Expression and Meaning (Cambridge: Cambridge University Press). —— (1979b). ‘Referential and Attributive’, The Monist, 62: 190–208. Sellars, W. (1954). ‘Presupposing’, Philosophical Review, 63: 197–215. Shanon, B. (1976). ‘On the Two Kinds of Presuppositions in Natural Language’, Foundations of Language, 14: 247–9. Sharvy, R. (1980). ‘A More General Theory of DeWnite Descriptions’, Philosophical Review, 87: 607–24. Simons, M. (1996). ‘Pronouns and DeWnite Descriptions: A Critique of Wilson’, Journal of Philosophy, 93: 408–20. —— (2003). ‘Presupposition and Accommodation: Understanding the Stalnakerian Picture’. Philosophical Studies, 112: 251–78. Sleigh, R. C. (1967). ‘On Quantifying into Epistemic Contexts’, Nouˆs, 1: 23–31. Smiley, T. J. (1981). ‘The Theory of Descriptions’, Proceedings of the British Academy, 67: 331–7. Soames, S. (1976). ‘A Critical Examination of Frege’s Theory of Presupposition and Contemporary Alternatives’, Ph.D. dissertation, MIT. —— (1986). ‘Incomplete DeWnite Descriptions’, Notre Dame Journal of Formal Logic, 27: 349–75. —— (1989a). ‘Direct Reference and Propositional Attitudes’, in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan (Oxford: Oxford University Press), 393–419. —— (1989b). ‘Presupposition’, in D. Gabbay and F. Guenthner (eds.), Handbook of Philosophical Logic, volume 4 (Dordrecht: Reidel), 553–616. —— (1990). ‘Pronouns and Propositional Attitudes’. Proceedings of the Aristotelian Society, 90: 191–212. —— (1994). ‘Donnellan’s Referential/Attributive Distinction’, Philosophical Studies, 73: 149–68. —— (1995). ‘Beyond Singular Propositions’, Canadian Journal of Philosophy, 25: 515–50. —— (1998a). ‘The Modal Argument: Wide Scope and RigidiWed Descriptions’, Nouˆs, 32: 1–22. —— (1998b). Understanding Truth (Oxford: Oxford University Press). Sommers, F. (1982). The Logic of Natural Language (Oxford: Clarendon Press). Sosa, E. (1970). ‘Propositional Attitudes De Dicto and De Re’, Journal of Philosophy, 67: 883–96. Sperber, D. (1994). ‘Understanding Verbal Understanding’, in J. Khalfa (ed.), What is Intelligence? (Cambridge: Cambridge University Press), 181–200. Sperber, D., and Wilson, D. (1986/1995). Relevance: Communication and Cognition (Oxford: Blackwell). Stalnaker, R. (1970). ‘Pragmatics’, Synthese, 22: 272–89. —— (1973). ‘Presuppositions’, Journal of Philosophical Logic, 2: 447–57. —— (1974). ‘Pragmatic Presuppositions’, in M. Munitz and P. Unger (eds.), Semantics and Philosophy (New York: New York University Press), 197–213. —— (1978). ‘Assertion’, in P. Cole (ed.), Syntax and Semantics, 9 Pragmatics (New York: Academic Press), 315–32. —— (1998). ‘On the Representation of Context’, Journal of Logic, Language and Information, 7: 3–19. —— (1999). Context and Content: Essays on Intentionality in Speech and Thought (New York: Oxford University Press).

References

649

—— (2001). ‘Considering a Possible World as Actual’ (MS). —— (2002). ‘Common Ground’, Linguistics and Philosophy, 25: 701–21. Stanley, J. (1997). ‘Names and Rigid Designation’, in B. Hale and C. Wright (eds.) A Companion to the Philosophy of Language (Oxford: Blackwell), 555–85. —— (2000). ‘Context and Logical Form’, Linguistics and Philosophy, 23: 391–434. —— (2002a). ‘Making it Articulated’ Mind and Language, 17: 149–68. —— (2002b). ‘Nominal Restriction’, in G. Preyer and G. Peter (eds.), Logical Form and Language (Oxford: Oxford University Press), 365–88. —— (2002c), ‘Modality and What is Said’, Philosophical Perspectives, 16: 321–44. —— and Szabo´, Z. (2000a). ‘On QuantiWer Domain Restriction’, Mind and Language, 15: 219–61. —— and—— (2000b). ‘Reply to Bach and Neale’, Mind and Language, 15: 295–8. —— and Williamson, T. (1995). ‘QuantiWers and Context–Dependence’, Analysis, 55: 291–5. Stechow, A. von (1991). ‘Focusing and Background Operators’, in W. Abraham (ed.), Discourse Particles (Amsterdam: John Benjamins). Strawson, P. (1950). ‘On Referring’, Mind, 59: 320–44. —— (1952). Introduction to Logical Theory (London: Methuen). —— (1954). ‘A Reply to Mr Sellars’, Philosophical Review, 63: 216–31. —— (1961). ‘Singular Terms and Predication’, Journal of Philosophy, 58: 393–412. —— (1964). ‘Identifying Reference and Truth-Values’, Theoria, 30: 96–118. —— (1969). ‘Meaning and Truth’, Inaugural lecture, University of Oxford. Reprinted in Logico-Linguistic Papers (London: Methuen), 170–89. —— (1971). ‘Identifying Reference and Truth-Values’, in Logico-Linguistic Papers (London: Methuen), 75–95. Sutton, J. (2001). ‘The Contingent A Priori and Implicit Knowledge’, Philosophy and Phenomenological Research, 63: 251–77. Szabo´, Z. (2000). ‘Descriptions and Uniqueness’, Philosophical Studies, 101: 29–58. Szabolcsi, A. (1994). ‘The Noun Phrase’, in F. Kiefer and K. Kiss (eds.), Syntax and Semantics, 27. The Syntactic Structure of Hungarian (New York: Academic Press), 179–275. —— (1997). ‘Strategies for Scope Taking’, in A. Szabolcsi (ed.), Ways of Scope Taking (Kluwer: Dordrecht), 109–54. Tarski, A. (1956). ‘The Concept of Truth in Formalized Languages’, in Logic, Semantics, and Metamathematics (Oxford: Clarendon Press). Thomason, R. (1990). ‘Accommodation, Meaning, and Implicature: Interdisciplinary Foundations for Pragmatics’, in P. Cohen, J. Morgan, and M. Pollack (eds.), Intentions in Communication (Cambridge, Mass.: MIT Press), 325–63. Travis, C. (1985). ‘On What is Strictly Speaking True’. Canadian Journal of Philosophy, 15: 187–229. —— (1997). ‘Pragmatics’, in B. Hale and C. Wright (eds.), A Companion to the Philosophy of Language (Oxford: Blackwell), 97–107. van der Does, J. (1993). ‘The Dynamics of Sophisticated Laziness’, in J. Groenendijk (ed.), Dyana Deliverable (R22A; Amsterdam: ILLC). —— (1996). ‘An E-Type Logic’, in J. Seligman and D. Westersta˚hl (eds.), Logic, Language and Computation, i (Stanford, Calif.: CSLI Publications), 555–70.

650

References

van der Sandt, Rob (1992). ‘Presupposition Projection as Anaphora Resolution’. Journal of Semantics, 9: 333–77. Vanderveken, D. (1990). Meaning and Speech Acts (Cambridge: Cambridge University Press). Veltman, F. (1983). ‘Data Semantics’, in J. Groenendijk, T. Janssen, and M. Stokhof (eds.), Truth, Interpretation, Information (Dordrecht: Foris), 43–63. Vendler, Z. (1967). Linguistics and Philosophy (Ithaca NY: Cornell University Press). Ward, G., and Birner, B. (1995). ‘DeWniteness and the English Existential’, Language, 71: 722–42. Webber, B. (1979). A Formal Approach to Discourse Anaphora (New York: Garland). Welker, K. (1994). ‘Plans in the Common Ground: Toward a Generative Account of Implicature’, Ph.D. dissertation, Ohio State University. Westersta˚hl, D. (1985a). ‘Determiners and Context Sets’, in J. van Benthem and A. ter Meulen (eds.), Generalized QuantiWers and Natural Language (Dordrecht: Foris), 45–71. —— (1985b). ‘Logical Constants in QuantiWer Languages’, Linguistics and Philosophy, 8: 387–413. Wettstein, H. (1981). ‘Demonstrative Reference and DeWnite Descriptions’, Philosophical Studies, 40: 241–57. —— (1983). ‘The Semantic SigniWcance of the Referential/Attributive Distinction’, Philosophical Studies, 44: 187–96. —— (1986). ‘Has Semantics Rested on a Mistake?’, Journal of Philosophy, 83: 185–209. —— (1988). ‘Cognitive SigniWcance without Cognitive Content’, Mind 97: 1–28. Wiggins, D. (1975). ‘Identity, Designation, Essentialism and Physicalism’, Philosophia, 5: 1–30. Williams, E. (1977). ‘Discourse and Logical Form’, Linguistic Inquiry, 8: 101–39. Wilson, D., and Sperber, D. (1993). ‘Linguistic Form and Relevance’, Lingua, 90: 1–25. Wilson, G. (1978). ‘On DeWnite and IndeWnite Descriptions’, Philosophical Review, 86: 48–76. —— (1984). ‘Pronouns and Pronominal Descriptions: A New Semantical Category’, Philosophical Studies, 45: 1–30. —— (1991). ‘Reference and Pronominal Descriptions’, Journal of Philosophy, 88: 359–87. Winter, Y. (1997). ‘Choice Functions and the Scopal Semantics of IndeWnites’, Linguistics and Philosophy, 20: 399–467. Yoon, Y. (1996). ‘Total and Partial Predicates and the Weak and Strong Interpretations’, Natural Language Semantics, 4: 217–36. Zimmerman, T. E. (1999). ‘Remarks on the Epistemic Role of Discourse Referents’, in L. Moss, J. Ginzburg, and M. de Rijke (eds.), Logic, Language & Computation, volume 2 (Stanford, Calif.: CSLI Publications), 346–68. Zribi-Hertz, A., and Mbolatianavalona, L. (1999). ‘Towards a Modular Theory of Linguistic DeWciency: Evidence from Malagasy Personal Pronouns’, Natural Language, and Linguistic Theory, 17: 161–218. Zubizaretta, M. (1998). Prosody, Focus, and Word Order (Cambridge, Mass.: MIT Press). Zucchi, A. (1995). ‘The Ingredients of DeWniteness and the DeWniteness EVect’, Natural Language Semantics, 3: 33–78. Zvolensky, Z. (1997). ‘DeWnite Descriptions: What Frege got Right and Russell Didn’t’, Aporia, 7: 1–16. —— (2000). ‘The Explicit and Implicit Approaches to Incomplet Descriptions’ (MS).

Index

Accessibility 196, 214, 459–60, 462–3, 476–80, 510, 541 Accommodation 316 n2, 500 n35, 511, 514, 518, 528, 546, 548 n5, 550, 554 Acquaintance 192, 207, 246, 369, 376, 592–4, 593 n4, 596, 608–11, 620–5 Acquaintance-based uses of names 621, 626–8 Ambiguity account of descriptions 69, 194–5, 226, 232–3, 233 n8, 261, 265, 379–80, 395 posited by hidden indexical theories 50–2 pragmatic 70, 195 n12, 232 problem for incomplete descriptions 50–5 of scope 52, 245, 560–1, 604–5 Anaphoric uses of descriptions 28–9, 35, 38–9, 212–13, 372 of indeWnites 529–30, 546–7 of pronouns 19, 144, 212–15, 456–7, 466–71, 476, 480–1, 487–90, 506, 521–2 Assertion(s) de re 240–1, 240 n15, 247–51 literal vs. indirect 245 multiple 245 vs. presupposition 318 vs. semantic content of sentences 239, 241 Binding with implicit arguments 547–9 wholly binding vs. binding into 175–6 Blocking of referential intentions 496–500 Bound uses of descriptions 19 n3, 36–8, 176–80 of pronouns 19, 525 Bridging inferences 548

Chains of communication blocks in 402, 404 sources of 404–7, 412, 414–17 Cleft sentences 343–4, 346, 353–4, 360 Common ground 316, 504, 510, 544 Competence semantic 84–5 syntactic 84 Conditionals 429–31 Conjunction 485, 493, 495–6 Context granularity of 272–5 and domain selection 47–50 and hidden indexicals 45–7 for the interpretation of indexicals 274–5 its role in completion 44–50 of use 43, 408, 410 and what is said 62–4 Contingent a priori 248, 623–4 Conventions for the use of descriptions 225–8, 283–8, 395–7 Conversational implicatures generalized 227, 286, 354 particularized 225, 268–9, 277–8, 283, 285, 338 Declaration of reference markers 550, 552–5 Default meaning 228 Deferred reference 270–2 DeWniteness eVect 431–3 Demonstratives compared to descriptions 190, 228, 288–9 complex 133, 174–5, 278–9, 291–3, 298 Denotation 16, 23, 218, 237, 239, 383, 395

652

Index

De re assertion vs. belief 247, 250–1 attitude ascription 211, 218 knowledge 248–9, 376 pseudo 241–4, 245–6, 250 thoughts 17, 191–2, 211–12, 218, 593–4, 606–7, 611–12, 625 De re/de dicto distinction 240, 245 Descriptions deictic uses of 40 mass 387–8 plural 385–7 predicational uses of 433–4 Diagonalization strategy 469–71 Direct reference 16, 208–9, 230, 266, 292 Discourse model 274 Discourse referent 212–15, 461, 505, 545 Discourse Representation Theory (DRT) 214, 332, 462–3, 558–9 Domain restriction 21, 31–2, 38, 47–50, 52, 92, 105, 119, 160, 273, 298, 301–3 Donkey anaphora 126, 461, 494, 505–6, 530–3 Downward entailment (antipersistence) 134–5, 532 ‘Dthat’ 208–9, 247, 261, 267 Dynamic theories of interpretation 489, 504 of semantics 214, 460–9, 481, 485–7, 544 of syntax 574–81 Ellipsis -cum-vagueness theory 25–6 conversational 98 sentence 103–5, 137–43, 154–9 utterance 98–103, 109–10, 122, 148–51, 153–6 Elliptical (‘explicit’) approach 21, 22–6, 44–5, 83–4, 97, 221, 298 Empty predicatives 392–4 En rapport with 17, 208, 228–9, 246, 304, 376 Existence presuppositions 315–16, 486 Exportation 228–9, 303–5

Familiarity 196 Felicity conditions for naming 594, 599–602 Fictional reference 215 File Change Semantics (FCS) 332 Focus 196, 434–5, 474–6, 481 Footholds for rejection 331–7 Free logic negative 371, 375–6 and empty names 399–400 Generalized referential /attributive distinction 258–60 Givenness 196–8, 424–5 Gricean maxims 198, 200, 268–70, 277, 425–6 Go¨delian completions 171–3 Homophonic truth-conditions 372, 374, 387 Implicatures Conversational 225, 227, 277–8, 283, 286 Conventional 424 Implicit binding 128, 160 Implicit vs. explicit approaches 116–25, 130, 148, 159–71 IndeWnites ambiguity account of 565–6 and conditionals 564–5 deWnite use of 423 inferred 545–7 in interrogatives and imperatives 501 parallelism with anaphora 566–73 referential use of 286–7, 288, 289, 293–5, 297, 304–5, 423, 465, 476 and scope 562–3 speciWc use of 205–6, 213, 293–5, 423 Indeterminacy 89–90 Index (¼context) 81–2, 93–6, 272 Index versus referent 271 Indexed Predicate Calculus 33–4 Indexicals ambiguity problem for 52–5

Index aphonic 107–8, 159–60 descriptive uses of 265–7 hidden 43, 46–59, 264, 270 linguistic meaning of (‘character’) 275, 409–11 overgeneration problem for 55–9 and translational semantics 409–11 Individual concept 255–6 Information structure 509 Intentions communicative 64, 77, 199 de re 253, 259 general vs. speciWc 252, 257 interpreter-directed 77 predicative 81 referential 80–1, 91, 199, 290, 487–90, 496 saying-intentions 78 semantic 253, 257 standing vs. occurrent 257 Kind terms artifactual 617, 626 natural 622–3, 626 Knowing who 217, 246, 254, 395 Knowledge by description 191, 206

653

Modal subordination 482, 508, 528–30 Names as abbreviations 605–6 attributive uses of 236, 252, 258–60 causal theory of 300 communicative eYciency of 628–9 logically proper 194–5, 231, 369, 376, 401 Millian view of 614, 616–20 modal properties of 378–9, 602–4 non-denoting (empty) 354–6, 370–1, 377–8, 400–3, 620, 627 ordinary 193–5, 235, 256, 260 ostensive 591 of past existents 403 semantic content of 595–6, 615 vivid 246 Narrowing of meaning 223 Negation and blocking of intentions 497–500 Negative existentials 350–1, 353, 359–60, 391–2, 397–8, 401–2, 406–7 Non-denoting terms incomplete descriptions as 65–6 Non-literal meaning 218

Lambda DRT 549–52 Linguistic convention 15, 173 Linguistic pragmatism 73–90 Logical form(s) 112–16, 128, 381–5, 391, 397–400

Occam’s Razor (ModiWed) 233, 284 Omniscient Observer theory 402, 409–11, 417 Over-generation problem 43, 55–7

Meinongian realism 356–7 Metaphor 226, 284–5 Meta-language vs. object-language 370–1, 377–8 Metonymic reference 271 Mind cross-identiWcation problem 394, 413, 416 Misdescription, argument from 262–3, 428–9 Missing content problem 393 Mock referential use 255–6, 260 Modal idioms 377, 408–9

Persistence (upward entailment) of determiners 129–31 Possessives 431 Pragmatic truth and falsity 326–7, 340–1 Predicate Logic with Anaphora (PLA) 490–4 Predicates complex 373 denotation of 32 evaluation of 32 existence-receptive vs. -rejective 404, 405, 412 implicit arguments of 545, 547–9

654

Index

Predicates (cont.): stage- vs. individual-level 332–3, 340, 534–5 Presupposition accommodation of 511, 528, 546 vs. assertion 318 of existence 315–16, 323, 344–5, 486 Wlters 318 plugs 318 projection of 316–17, 323, 345–6 of salience 516–18, 527 semantic 318 speaker 316 tests for 317, 323 of uniqueness 264, 273, 341, 507 of weak familiarity 507–12, 527, 546 Pronominal contradiction problem 459, 467 Pronouns ambiguity of 19 anaphoric uses of 19, 144, 191 n2, 456–7, 466–71, 487–90, 495, 497, 499, 521–2 bound uses of 19, 144–5, 505 D-type 126, 144, 214, 521–23 as descriptive 17–8, 20, 125, 180–1, 479, 518–26 dynamic binding account of 524 E-type 457–60, 464, 467, 479, 519–21, 540–1 generic interpretation of 533–6 as incomplete descriptions 180–1 of laziness 305 n51, 519–20 number neutral 536–40 plural 464–5, 481, 533–4 Properties object dependent 24–5 ‘sticking’ 32 Propositions gappy 409 partial 317–18 singular vs. general 19, 111, 191 n2, 198, 206, 219, 232, 240, 245, 396 QuantiWers generalized 274, 477, 482, 531, 551 restricted 38

Reference determining relations 281 Referentialism quasi 20–1, 26, 30 situational 22, 27–31 Referring expressions/terms descriptive condition for 15 pragmatic condition for 15 Russellian constraints on 370, 374–9 Relevance theory 471–2, 474 Rigidity of descriptions 372–3, 378, 408–9, 416 de facto vs. de jure 372–3, 380 of names 372, 614 weak 295–7 Scope distinctions 377–8, 509–10, 527–30, 561–6, 578–80 Semantic reference/meaning/content 16, 21, 79, 233, 241 Semantics from above vs. from below 391, 403, 410–11, 415 Davidsonian 370, 381–2 expression-centered conception of 238–9 game theoretical 498 speech act centered conception of 238–9 translational 391–7, 409, 413–14 Semantic signiWcance thesis 146, 231–7, 260, 281, 414–17 Semantics/pragmatics distinction 85 Sentence/utterance distinction 147–8, 152 Sentential operator reduction 392–3, 412–14 Singling out versus referring 210–12 Singular proposition/thought 17, 191, 195, 200, 204, 262, 280, 290 Singular terms 42, 374 Situational variable 33–4 Situations exercised 32 focus 26, 27–9 generation-of-use 408–9, 416 that highlight an object 39–40 maximal 120 minimal 29, 35–6, 478, 527

Index partial 22 quantiWcation over 29–31, 38 Situation theory 472, 277, 558–9 Slingshot argument 171, 378 Speaker’s reference/meaning 15, 43, 59–60, 189, 199–200, 218, 282, 395, 422–3, 458, 481 Speech act parentheticals 473–4 Standardization 227–8, 286 Static binding approach to anaphora 456–7 Subject-predicate sentences 373–4, 393, 412–14 Subject-subordination 392, 398 Telescoping 538–9 Temporal operators 32 Time of use versus evaluation 407–9 Token reXexives 16

655

Topic-Focus (Topic-Comment) 322, 324–5, 345–6, 347–8, 351–3, 354–6, 498–500 Underdetermination (under-speciWcation) 87–9, 91, 101, 107, 372, 381, 558 Uniqueness implication of 202–3, 220, 224, 421–2, 424–5, 458–9, 466, 482 informational vs. semantic 512–15, 526–7 presupposition of 264, 273–4, 341, 507 Upward entailment (persistence) 131–4, 135–7, 532 What is said 44, 60–6, 78–9, 85–7, 91, 217, 396, 417

Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Reference and Definite Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Descriptions in Nonextensional Contexts

Read more

Paganini For Two - Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Writing Legal Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Searle on Descriptions

Read more

Recommend Documents

Descriptions

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...

Reference and Definite Descriptions

...