No title

VOLUME 25 NUMBER 3 AUGUST 2008 CONTENTS 221 YURIE HARA Evidentiality of Discourse Items and Because-Clauses 229 JAME...

Author: Oxford University Press

29 downloads 242 Views 1MB 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

VOLUME 25 NUMBER 3 AUGUST 2008

CONTENTS 221

YURIE HARA Evidentiality of Discourse Items and Because-Clauses

229

JAMES ISAACS AND KYLE RAWLINS Conditional Questions

269

TIM FERNANDO Branching from Inertia Worlds

321

FORTH COMING ARTICLES PAULA RUBIO-FERNÁNDEZ: Concept Narrowing:The Role of Context-independent Information ARIEL COHEN: No Alternative to Alternatives CHUNG-HYE HAN AND NANCY HEDBERG: Syntax and Semantics of It-Clefts: A Tree Adjoining Grammar Analysis

VOLUME 25 NUMBER 3 AUGUST 2008

TAKAO GUNJI, STEFAN KAUFMANN AND YUKINORI TAKUBO Modality and Evidentiality

JOURNAL OF SEMANTICS

JOURNAL OF SEMANTICS

issn 0167-5133

VOLUME 25 NUMBER 3 AUGUST 2008

Journal of

SEMANTICS www.jos.oxfordjournals.org Special issue on MODALITY AND EVIDENTIALITY Guest editors: TAKAO GUNJI STEFAN KAUFMANN YUKINORI TAKUBO

oxford

JOURNAL OF SEMANTICS A N I NTERNATIONAL J OURNAL FOR THE I NTERDISCIPLINARY S TUDY THE S EMANTICS OF N ATURAL L ANGUAGE

MANAGING EDITOR: ASSOCIATE EDITORS:

OF

BART G EURTS (University of Nijmegen) DAVID B EAVER (Stanford University) R EGINE E CKARDT (Universität Göttingen) I RA N OVECK (Institut des Sciences Cognitives, Lyon) PAUL P ORTNER (Georgetown University,Washington) P HILIPPE S CHLENKER (Institut Jean-Nicod, Paris) YAEL S HARVIT (University of Connecticut, Storrs) A NNA S ZABOLCSI (New York University) EDITORIAL BOARD:

N ICHOLAS A SHER (University of Texas, Austin) C HRIS B ARKER (University of California at San Diego) J OHAN B OS (University of Edinburgh) P ETER B OSCH (University of Osnabrück) R ICHARD B REHENY (University College London) M IRIAM B UTT (University of Konstanz) G REG C ARLSON (University of Rochester) A NN C OPESTAKE (Stanford University) H ENRIËTTE DE S WART (Utrecht University) PAUL D EKKER (University of Amsterdam) K URT E BERLE (Lingenio Heidelberg) M ARKUS E GG (Universität des Saarlandes) U LRIKE H AAS -S POHN (University of Konstanz) L AURENCE R. H ORN (Yale University) H ANS K AMP (University of Stuttgart) G RAHAM K ATZ (University of Osnabrück) T IBOR K ISS (Ruhr University, Bochum) J ONAS K UHN (University of Texas, Austin)

CLAUDIA MAIENBORN (Humboldt University, Berlin) JULIEN MUSOLINO (Rutgers University) FRANCIS JEFFRY PELLETIER (University of Alberta) CHRISTOPHER POTTS (University of Massachusetts, Amherst) MARK STEEDMAN (University of Edinburgh) ZOLTAN GENDLER SZABO (Cornell University) KEES VAN DEEMTER (University of Aberdeen) ROB VAN DER SANDT (University of Nijmegen) ROBERT VAN ROOIJ (University of Amsterdam) KAI VON FINTEL (Massachusetts Institute of Technology) ARNIM VON STECHOW (University of Tübingen) BONNIE WEBBER (University of Edinburgh) HENK ZEEVAT (University of Amsterdam) THOMAS EDE ZIMMERMANN (University of Frankfurt)

EDITORIAL CONTACT: [email protected] © Oxford University Press 2008 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, electronic, mechanical, photocopying, recording, or otherwise without prior written permission of the Publishers, or a licence permitting restricted copying issued in the UK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1P 9HE, or in the USA by the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. Typeset by TnQ Books and Journals Pvt. Ltd., Chennai, India. Printed by Bell and Bain Ltd, Glasgow, UK For subscription information please see back of journal.

Scope of this Journal The Journal of Semantics publishes articles, notes, discussions, and book reviews in the area of academic research into the semantics of natural language. It is explicitly interdisciplinary, in that it aims at an integration of philosophical, psychological, and linguistic semantics as well as semantic work done in logic, artificial intelligence, and anthropology. Contributions must be of good quality (to be judged by at least two referees) and must report original research relating to questions of comprehension and interpretation of sentences, texts, or discourse in natural language. The editors welcome not only papers that cross traditional discipline boundaries, but also more specialized contributions, provided they are accessible to and interesting for a general readership in the field of natural language semantics. Empirical relevance, sound theoretic foundation, and formal as well as methodological correctness by currently accepted academic standards are the central criteria of acceptance for publication. It is also required of contributions published in the Journal that they link up with currently relevant discussions in the field of natural language semantics. Information for Authors Papers for publication should be submitted to the Managing Editor by email as a PDF file or PS file attachment. If this is not feasible please contact the Managing Editor.The receipt of submissions is confirmed by email (when there is more than one author to the first author, whom we assume to deal with all correspondence, unless we are instructed differently), and the paper is reviewed by two members of the editorial board or external experts chosen by the editors. The reviewers remain anonymous. An editorial decision is normally reached within 2-3 months after submission. Papers are accepted for review only on the condition that they have neither as a whole nor in part been published elsewhere, are elsewhere under review or have been accepted for publication. In case of any doubt authors must notify the editor of the relevant circumstances at the time of submission. It is understood that authors accept the copyright conditions stated in the journal if the paper is accepted for publication. The style requirements of the Journal of Semantics can be found at www.jos.oxfordjournals. org, under “Instructions to Authors”, and are binding for the final version to be prepared by the author when the paper is accepted for publication. LATEX submission Please use the Journal class file (http://www3.oup.co.uk/semant/instauth/semant.cls). A tex file (http://www3.oup.co.uk/semant/instauth/guide.tex) is available on how to use the .cls file. Authors who are planning to send source files by email should also include a postscript or PDF version of their paper. Please follow all the instructions to authors that are detailed above and note the text width should be set to 28pc and the text height to 41\baselineskip. Electronic figures can only be used in ps or eps format.

JOURNAL OF SEMANTICS A N I NTERNATIONAL J OURNAL FOR THE I NTERDISCIPLINARY S TUDY THE S EMANTICS OF N ATURAL L ANGUAGE

MANAGING EDITOR: ASSOCIATE EDITORS:

OF

BART G EURTS (University of Nijmegen) DAVID B EAVER (Stanford University) R EGINE E CKARDT (Universität Göttingen) I RA N OVECK (Institut des Sciences Cognitives, Lyon) PAUL P ORTNER (Georgetown University,Washington) P HILIPPE S CHLENKER (Institut Jean-Nicod, Paris) YAEL S HARVIT (University of Connecticut, Storrs) A NNA S ZABOLCSI (New York University) EDITORIAL BOARD:

N ICHOLAS A SHER (University of Texas, Austin) C HRIS B ARKER (University of California at San Diego) J OHAN B OS (University of Edinburgh) P ETER B OSCH (University of Osnabrück) R ICHARD B REHENY (University College London) M IRIAM B UTT (University of Konstanz) G REG C ARLSON (University of Rochester) A NN C OPESTAKE (Stanford University) H ENRIËTTE DE S WART (Utrecht University) PAUL D EKKER (University of Amsterdam) K URT E BERLE (Lingenio Heidelberg) M ARKUS E GG (Universität des Saarlandes) U LRIKE H AAS -S POHN (University of Konstanz) L AURENCE R. H ORN (Yale University) H ANS K AMP (University of Stuttgart) G RAHAM K ATZ (University of Osnabrück) T IBOR K ISS (Ruhr University, Bochum) J ONAS K UHN (University of Texas, Austin)

CLAUDIA MAIENBORN (Humboldt University, Berlin) JULIEN MUSOLINO (Rutgers University) FRANCIS JEFFRY PELLETIER (University of Alberta) CHRISTOPHER POTTS (University of Massachusetts, Amherst) MARK STEEDMAN (University of Edinburgh) ZOLTAN GENDLER SZABO (Cornell University) KEES VAN DEEMTER (University of Aberdeen) ROB VAN DER SANDT (University of Nijmegen) ROBERT VAN ROOIJ (University of Amsterdam) KAI VON FINTEL (Massachusetts Institute of Technology) ARNIM VON STECHOW (University of Tübingen) BONNIE WEBBER (University of Edinburgh) HENK ZEEVAT (University of Amsterdam) THOMAS EDE ZIMMERMANN (University of Frankfurt)

EDITORIAL CONTACT: [email protected] © Oxford University Press 2008 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, electronic, mechanical, photocopying, recording, or otherwise without prior written permission of the Publishers, or a licence permitting restricted copying issued in the UK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1P 9HE, or in the USA by the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. Typeset by TnQ Books and Journals Pvt. Ltd., Chennai, India. Printed by Bell and Bain Ltd, Glasgow, UK For subscription information please see back of journal.

Scope of this Journal The Journal of Semantics publishes articles, notes, discussions, and book reviews in the area of academic research into the semantics of natural language. It is explicitly interdisciplinary, in that it aims at an integration of philosophical, psychological, and linguistic semantics as well as semantic work done in logic, artificial intelligence, and anthropology. Contributions must be of good quality (to be judged by at least two referees) and must report original research relating to questions of comprehension and interpretation of sentences, texts, or discourse in natural language. The editors welcome not only papers that cross traditional discipline boundaries, but also more specialized contributions, provided they are accessible to and interesting for a general readership in the field of natural language semantics. Empirical relevance, sound theoretic foundation, and formal as well as methodological correctness by currently accepted academic standards are the central criteria of acceptance for publication. It is also required of contributions published in the Journal that they link up with currently relevant discussions in the field of natural language semantics. Information for Authors Papers for publication should be submitted to the Managing Editor by email as a PDF file or PS file attachment. If this is not feasible please contact the Managing Editor.The receipt of submissions is confirmed by email (when there is more than one author to the first author, whom we assume to deal with all correspondence, unless we are instructed differently), and the paper is reviewed by two members of the editorial board or external experts chosen by the editors. The reviewers remain anonymous. An editorial decision is normally reached within 2-3 months after submission. Papers are accepted for review only on the condition that they have neither as a whole nor in part been published elsewhere, are elsewhere under review or have been accepted for publication. In case of any doubt authors must notify the editor of the relevant circumstances at the time of submission. It is understood that authors accept the copyright conditions stated in the journal if the paper is accepted for publication. The style requirements of the Journal of Semantics can be found at www.jos.oxfordjournals. org, under “Instructions to Authors”, and are binding for the final version to be prepared by the author when the paper is accepted for publication. LATEX submission Please use the Journal class file (http://www3.oup.co.uk/semant/instauth/semant.cls). A tex file (http://www3.oup.co.uk/semant/instauth/guide.tex) is available on how to use the .cls file. Authors who are planning to send source files by email should also include a postscript or PDF version of their paper. Please follow all the instructions to authors that are detailed above and note the text width should be set to 28pc and the text height to 41\baselineskip. Electronic figures can only be used in ps or eps format.

SUBSCRIPTIONS

A subscription to Journal of Semantics comprises 4 issues. All prices include postage, and for subscribers outside the UK delivery is by Standard Air. Journal of Semantics Advance Access contains papers that have been finalised, but have not yet been included within the issue. Advance Access is updated monthly. Annual Subscription Rate (Volume 25, 4 issues, 2008) Institutional Print edition and site-wide online access: £175/$350/=263 C Print edition only: £166/$332/=249 C Site-wide online access only: £166/$332/=249 C Personal Print edition and individual online access: £65/$130/=98 C Please note: £ Sterling rates apply in Europe, US$ elsewhere There may be other subscription rates available, for a complete listing please visit www.jos.oxfordjournals.org/subscriptions. Full prepayment, in the correct currency, is required for all orders. Orders are regarded as firm and payments are not refundable. Subscriptions are accepted and entered on a complete volume basis. Claims cannot be considered more than FOUR months after publication or date of order, whichever is later. All subscriptions in Canada are subject to GST. Subscriptions in the EU may be subject to European VAT. If registered, please supply details to avoid unnecessary charges. For subscriptions that include online versions, a proportion of the subscription price may be subject to UK VAT. Personal rate subscriptions are only available if payment is made by personal cheque or credit card and delivery is to a private address. The current year and two previous years’ issues are available from Oxford University Press. Previous volumes can be obtained from the Periodicals Service Company, 11 Main Street, Germantown, NY 12526, USA. Email: [email protected]. Tel: +1 (518) 537 4700. Fax: +1 (518) 537 5899. For further information, please contact: Journals Customer Service Department, Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, UK. Email: [email protected]. Tel (and answerphone outside normal working hours): +44 (0)1865 353907. Fax: + 44 (0)1865 353485. In the US, please contact: Journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513, USA. Email: [email protected]. Tel (and answerphone outside normal working hours): 800 852 7323 (toll-free in USA/Canada). Fax: 919 677 1714. In Japan, please contact: Journals Customer Services, Oxford University Press, 1-1-17-5F, Mukogaoka, Bunkyo-ku, Tokyo, 113-0023, Japan. Email: [email protected]. Tel: (03) 3813 1461. Fax: (03) 3818 1522. Methods of payment. Payment should be made: by cheque (to Oxford University Press, Cashiers Office, Great Clarendon Street, Oxford, OX2 6DP, UK); by bank transfer [to Barclays Bank Plc, Oxford Office, Oxford (bank sort code 20-65-18) (UK);

overseas only Swift code BARC GB22 (GB£ Sterling Account no. 70299332, IBAN GB89BARC20651870299332; US$ Dollars Account no. 66014600, IBAN GB27BARC20651866014600; EU= C EURO Account no. 78923655, IBAN GB16BARC20651878923655]; or by credit card (Mastercard, Visa, Switch or American Express). Journal of Semantics (ISSN 0167 5133) is published quarterly (in February, May, August and November) by Oxford University Press, Oxford, UK. Annual subscription price is £175/$350/=263. C Journal of Semantics is distributed by Mercury International, 365 Blair Road, Avenel, NJ 07001, USA. Periodicals postage paid at Rahway, NJ and at additional entry points. US Postmaster: send address changes to Journal of Semantics (ISSN 0167-5133), c/o Mercury International, 365 Blair Road, Avenel, NJ 07001, USA. Abstracting and Indexing Annual Bibliography English Language Literature (ABEL), INSPEC, International Bibliography Sociology, Linguistics Abstracts, Linguistics and Language Behaviour Abstracts (LLBA), MLA: International Bibliography Books, Articles and Modern Language Literature, periodicals Contents Index, Philosopher’s Index, Social Planning Policy and Development Abstracts, Bibliographie Linguistique/Linguistic Bibliography and BLonline. Permissions For information on how to request permissions to reproduce articles/information from this journal, please visit www.oxfordjournals.org/jnls/permissions. Advertising Inquiries about advertising should be sent to Helen Pearson, Oxford Journals Advertising, PO Box 347, Abingdon OX14 1GJ, UK. Email: [email protected]. Tel: +44 (0)1235 201904. Fax: +44 (0)8704 296864. Disclaimer Statements of fact and opinion in the articles in Journal of Semantics are those of the respective authors and contributors and not of Journal of Semantics or Oxford University Press. Neither Oxford University Press nor Journal of Semantics make any representation, express or implied, in respect of the accuracy of the material in this journal and cannot accept any legal responsibility or liability for any errors or omissions that may be made. The reader should make his/her own evaluation as to the appropriateness or otherwise of any experimental technique described.

JOURNAL OF SEMANTICS Volume 25 Number 3

Special issue on MODALITY AND EVIDENTIALITY

Guest editors: TAKAO GUNJI STEFAN KAUFMANN YUKINORI TAKUBO

CONTENTS TAKAO GUNJI, STEFAN KAUFMANN AND YUKINORI TAKUBO Modality and Evidentiality 221 YURIE HARA Evidentiality of Discourse Items and Because-Clauses

229

JAMES ISAACS AND KYLE RAWLINS Conditional Questions

269

TIM FERNANDO Branching from Inertia Worlds

321

Please visit the journal’s web site at www.jos.oxfordjournals.org

Journal of Semantics 25: 221–227 doi:10.1093/jos/ffn006

Modality and Evidentiality TAKAO GUNJI Kobe Shoin Women’s University STEFAN KAUFMANN Northwestern University

Linguistic expressions of modality and evidentiality have been the object of much active and exciting research for many years, but continue to pose challenges in all areas of semantic theory. Key aspects of their behaviour at the syntax–semantics interface, such as interactions with quantifiers and other operators or the calculation of modal implicatures triggered in embedding contexts, are not yet fully understood. Likewise, both the semantic interpretation and the pragmatic role of nondeclarative sentences in modal constructions have only recently begun to attract significant attention, partly because the standard formal apparatus requires substantial augmentations in order to be applicable in that domain. But even one of the most fundamental and least controversial aspects of that formal apparatus, the use of possible worlds in the model theory, is subject to criticism and controversy. The articles in this collection address open issues in each of these three areas. The first paper, Yurie Hara’s ‘Evidentiality of Discourse Items and Because-clauses’, addresses the problem of implicature computation involved in Japanese, focusing on contrastive wa-marking and sooda, a marker of ‘hearsay’ evidentiality. The use of contrastive wa-marking presupposes a contextually given stronger alternative, e.g. Mary and John came in (1b). (All glosses in this section are Hara’s. ‘Con’ glosses contrastive wa.) When this presupposition is satisfied, the sentence conventionally implicates that the speaker does not know that the stronger alternative is true, or knows that it is false. (1)

a. Among John and Mary, who came to the party? b. JOHN-wa kita John-Con came JOHN came. [Mary didn’t come, or I don’t know about Mary.] Ó The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected].

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

YUKINORI TAKUBO Kyoto University

222 Modality and Evidentiality Examining the interpretation of wa-marking in embedded clauses, Hara shows that the implicatures induced by wa can be relativized to an attitude holder other than the speaker. As can be seen in (2), both a ‘local’ and a ‘global’ association of the implicature are possible. Hara suggests that the contrastive operator Con is represented in the syntax and that its location determines the attitude bearer of the induced implicature, i.e. the speaker or the subject of the attitude predicate. (2) MARY-wa kita-to Mary-Con come-Comp

John-ga John-Nom

shinjite-iru believe-Prog Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

John believes that (at least) Mary came. Implicature: a. Local: John doesn’t know [whether Peter came] b. Global: The speaker doesn’t know [whether John knows that Peter came] Against this background, Hara discusses a parallelism between contrastive wa-marking and the evidential marker sooda: both can appear in clauses headed by node ‘because’, but not in temporal clauses headed by toki ‘when’ or ato ‘after’. The crucial difference between the two contexts, Hara argues, is that node ‘because’ takes proposition-denoting complements, whereas temporal connectives (as well as conditional ones) require complements denoting properties of events. The proposition required by the ‘because’ operator is obtained from an event property via existential closure of its event variable. Following Johnston (1994), Hara assumes that this operation is also available to other operators requiring propositional arguments—specifically, the denotations of contrastive wa and evidential sooda. Both take proposition-denoting complements and, moreover, produce proposition-denoting phrases, suitable for embedding under node ‘because’ but not under temporal or conditional connectives. Thus, the distributional similarities between contrastive wa and evidential sooda are caused by the same type-driven mechanism. Semantically, Hara suggests that the implicature induced by wa-marking concerns an agent’s attitude towards a proposition. Returning to (2), Hara observes that the use of contrastive wa and its association with an attitude holder are subject to blocking by certain syntactic configurations that are also islands for wh-movement. To capture this similarity, she proposes an LF movement account for the positioning of the Con operator. Con moves on a par with wh-phrases, which explains its island sensitivity. Hara’s paper has interesting ramifications regarding the connection between implicatures and evidentiality, both of which have so far been

Takao Gunji, Stefan Kaufmann and Yukinori Takubo 223

(3)

If Alfonso comes to the party, will Joanna leave?

In order to capture the intuition that (3) is more limited in scope than a simple question like (4) below, they propose a two-step dynamic interpretation. In the first step, a temporary context is created where the propositional content of the antecedent holds. In the second step, the question in the consequent raises an issue relative only to the temporary context. (4)

Will Joanna leave?

This approach provides for an interesting treatment of the denial of the antecedent, as in the following: (5)

a. If Alfonso comes to the party, will Joanna leave? b. Alfonso isn’t coming.

In such a case, there is no need to answer the question expressed by the consequent. Unlike previous accounts (Hulstijn 1997; Velissaratou 2000) that treat the denial of the antecedent as one alternative answer to the question—what they call semanticization of a pragmatic effect—the analysis presented here treats it as a presupposition failure. The same line of analysis also works for counterfactuals, where the presupposition is different: (6)

a. b.

If Jo could have fixed the car, would you have kept on using it? Jo could have fixed the car.

In the case of counterfactuals, the presupposition is that the antecedent is false, so the affirmation of the antecedent is issue dispelling. In this

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

treated within semantics–pragmatics. Recent studies have broadened the perspective by introducing syntax into the picture. Chierchia (2004) argues that syntactic structures must be accessible for implicature computation. Tenny (2006) and Speas (2004) hypothesize that there is a syntactic representation of point of view. The observation that because-clauses can embed an attitude-bearing element while temporal clauses cannot has been around for some time in Japanese linguistics, but no formal explanation to the distributional asymmetry has been given. Hara’s paper thus offers novel and insightful approach to implicature computation in embedding contexts. James Isaacs and Kyle Rawlins, in their paper ‘Conditional Question’, provide an analysis of conditional questions (CQ) which utilizes both a dynamic semantics (for conditionals) and a partition semantics (for questions). The CQs they treat are of the following type:

224 Modality and Evidentiality way, Isaacs and Rawlins give a uniform account for both noncounterfactual and counterfactual conditionals as to the denial/ affirmation of the antecedent. The technical formalization of their analysis is an extension of modal subordination, as in the example below (Roberts 1989): (7) a. A thief might break into the house. b. He would take the silver.

(8) a. Pat was eating. b. Pat ate. (9) a. Pat was eating a hamburger. b. Pat ate a hamburger. The crucial difference between (8) and (9) lies in the aspectual class of the radicals embedded under the progressive (activity and accomplishment, respectively). Thus, any successful analysis of the phenomenon

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In order to capture the effect of the temporary context formally, they utilize the stack-based model of Kaufmann (2000) and define the update of the context at the top of the stack in terms of partition semantics (Groenendijk 1999). The contrast between assertions and questions is that, while the former provide information that persists, the latter only raise issues, which do not percolate down the stack and therefore do not persist. In addition to refining and extending Kaufmann’s stack-based model, the paper is strong contribution to two lively strands of theoretical work: on the formal analysis of non-declarative sentences (see Portner 2004; Schwager 2006 and others on imperatives; Zanuttini & Portner 2003 on exclamatives) as well as the representation of discourse context in terms of dynamically changing hierarchies of questions (Roberts 1996; Groenendijk 2008, among others). In ‘Branching from Inertial Worlds,’ Tim Fernando addresses deeper questions concerning the ontological underpinnings of the formal analysis of modality. Hara as well as Isaacs and Rawlins spell out their respective semantic analyses in terms of possible worlds. This ontological construct is familiar and widely used by semanticists working in the area of modality and evidentiality, even though there are some important notions for whose formalization it is less than perfectly suited. One such notion is that of inertia, first proposed by Dowty (1977) in his analysis of the ‘imperfective paradox’. The problem is illustrated in the contrast between (8) and (9): whereas the past progressive sentence in (8a) entails its simple past counterpart (8b), the corresponding entailment from (9a) to (9b) does not hold.

Takao Gunji, Stefan Kaufmann and Yukinori Takubo 225

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

must take aspectual behaviour into account. But the main reason why these data are of interest in the context of modality is that while (9a) may be true when (9b) is false, the falsehood of the latter is not sufficient for the truth of the former. Short of entailing its simple past counterpart, (9a) does imply that Pat’s eating at the past time in question was in some sense ‘bound’ to run its course and culminate in the complete consumption of the hamburger. Dowty (1977) proposed to capture this intuition by stipulating that the eating culminated in all ‘inertia worlds’—a set of alternative continuations of history at the time in question which may or may not include the actual one. A satisfactory formalization of this idea in terms of possible worlds has proven elusive. Roughly speaking, inertia worlds are neither those that were most likely at the time nor those that are most similar to the actual one (see also Landman 1992, among others), but rather those in which a particular eventuality—that referred to by the sentence radical—continues without interruption. The idea of eventualities running independently or interrupting each other does not sit well with the usual view of possible worlds as indivisible totalities of facts. While it may not be impossible to integrate such a notion by adding internal structure to worlds and ‘taking them apart’, in a sense this would be the wrong end from which to approach the problem. For what ultimately matters are the pieces—states and processes—and once we can refer to those, the analysis can be spelled out without reference to the worlds that contain them. Fernando forcefully defends this view by producing an account in which simple event descriptions, called ﬂuents with reference to the AI literature, form the basic building blocks of a formal language for describing complex processes—partial objects of which possible worlds are but a limiting case—and their interactions. Under this account, inertia is a property of fluents, more precisely the tendency of some fluents to continue indefinitely until they are acted upon by a force. In the formal language, such interruptions give rise to branching expressions which describe both the undisturbed course of events and the one resulting from the interruption, thus in effect describing counterfactual possibilities. Fernando demonstrates the utility of this device in the analysis of modal expressions referring to alternative courses of events, such as the progressive and temporal before (Beaver & Condoravdi 2003). One intriguing aspect of the analysis whose broader implications seem worth exploring is a distinction between single-branching strings on the one hand and sets of alternative strings on the other, which Fernando suggests captures the difference between (objective)

226 Modality and Evidentiality indeterminacy and (subjective) ignorance. It is well known that this difference has linguistic consequences (Condoravdi 2002; Kaufmann 2005), thus Fernando’s proposal may provide a useful analytical tool for a wide range of data. Acknowledgements

TAKAO GUNJI Shoin Institute for Linguistic Sciences Kobe Shoin Women’s University 1-2-1 Shinohara, Obanoyama-cho, Nada Kobe 657-0015 Japan e-mail: [email protected] STEFAN KAUFMANN Department of Linguistics Northwestern University 2016 Sheridan Road Evanston IL 60208 USA e-mail: [email protected] YUKINORI TAKUBO Department of Linguistics Graduate School of Letters Kyoto University Yoshida-Honmachi, Sakyo-ku Kyoto 606-8501 Japan e-mail: [email protected]

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This collection grew out of a conference on ‘Language under Uncertainty’, organized by the editors at Kyoto University, Japan, on 21–23 January 2005, and sponsored by Group 36 in the research project ‘Towards a Center of Excellence for the Study of Humanities in the Age of Globalization’ (Group Leader: Prof. Yukinori Takubo). The program is available at http://www.hmn.bun.kyoto-u.ac.jp/langlogic/symposium-a/ (June 2008). We thank all those who made the conference a success: the speakers and the audience, as well as J-R Hayashihita, Rieko Okada, Wumpradit Apasara, Sanae Tamura, Tamami Shimada, Yuka Hayashi and other graduate students at the Department of Linguistics, Kyoto University, for their practical assistance. Nine of the 14 speakers submitted manuscripts upon our invitation. We are grateful to 18 external reviewers for detailed and helpful comments, and especially to Bart Geurts, the managing editor of this journal, for his encouragement and patience.

Takao Gunji, Stefan Kaufmann and Yukinori Takubo 227

REFERENCES Formalizing the Dynamics of Information. CSLI. Stanford, CA. Kaufmann, S. (2005), ‘Conditional truth and future reference’. Journal of Semantics 22:231–80. Landman, F. (1992), ‘The progressive’. Natural Language Semantics 1:1–32. Portner, P. (2004), ‘The semantics of imperatives within a theory of clause types’. In K. Watanabe and R. Young (eds.), Proceedings of Semantics and Linguistic Theory 14. CLC Publications. Ithaca, NY. Roberts, C. (1989), ‘Modal subordination and pronominal anphora in discourse’. Linguistics and Philosophy 12:683–721. Roberts, C. (1996), ‘Information structure: towards an integrated formal theory of pragmatics’. In J. H. Yoon and A. Kathol (eds.), Papers in Semantics, volume 49 of OSUWPL. The Ohio State University Department of Linguistics. Columbus, Ohio. 91–136. Schwager, M. (2006), Interpreting Imperatives. Ph.D. thesis, Johann Wolfgang Go¨ethe Universita¨t Frankfurt am Main. Speas, M. (2004), ‘Evidentiality, logophoricity and the syntactic representation of pragmatic features’. Lingua 114:255–76. Tenny, C. (2006), ‘Evidentiality, experiencers, and the syntax of sentience in Japanese’. Journal of East Asian Linguistics 15:245–88. Velissaratou, S. (2000), ‘Conditional questions and which-interrogatives’. Master of logic thesis, University of Amsterdam, ILLC Publications. Zanuttini, R. & P. Portner. (2003), ‘Exclamative clauses: at the syntaxsemantics interface’. Language 79: 39–81.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Beaver, D. & C. Condoravdi (2003), ‘A uniform analysis of before and after’. In R. Yound and Y. Zhou (eds.), Proceedings of SALT XIII. 37– 54. Chierchia, G. (2004), ‘Scalar implicatures, polarity phenomena, and the syntax/pragmatics interface’. In A. Belletti (ed.), Structures and Beyond, volume 3 of The Cartography of Syntactic Structures, chapter 2. Oxford University Press. New York. 39–103. Condoravdi, C. (2002), ‘Temporal interpretation of modals: modals for the present and for the past’. In D. I. Beaver, L. Casillas, B. Clark and S. Kaufmann (eds.), The Construction of Meaning. CSLI Publications. Stanford. 59–88. Dowty, D. (1977), ‘Towards a semantic analysis of verb aspect and the English ‘‘imperfective’’ progressive’. Linguistics and Philosophy 1:45–77. Groenendijk, J. (1999), ‘The logic of interrogation’. In T. Matthews and D. L. Strolovitch (eds.), Proceedings of SALT IX. CLC Publications. New York. 109–126. Groenendijk, J. (2008), ‘Inquisitive semantics and dialogue pragmatics’. Unpublished MS, University of Amsterdam. Hulstijn, J. (1997), ‘Structured information states. Raising and resolving issues’. In A. Benz and G. Ja¨ger (eds.), Proceedings of MunDial97. University of Munich. Munich, Germany. 99–117. Johnston, M. (1994), The Syntax and Semantics of Adverbial Adjuncts. Ph.D. thesis, University of California, Santa Cruz, CA. Kaufmann, S. (2000), ‘Dynamic context management’. In M. Faller, S. Kaufmann and M. Pauly (eds.),

Journal of Semantics 25: 229–268 doi:10.1093/jos/ffn001 Advance Access publication June 20, 2008

Evidentiality of Discourse Items and Because-Clauses YURIE HARA Kyoto University

There is a parallelism between contrastive marking and evidential marking with respect to their distribution among adjunct clauses. I take this fact to show that both contrastive marking and evidential marking express some attitude towards a closed proposition, following Johnston’s (1994) analysis that the semantics of temporal and if-clauses involve event quantification while that of because-clauses is a relation between two particular events. Furthermore, this association between the implicature and the attitude-holder cannot be established in certain constructions, namely adjunct clauses and relative clauses. Hence, I argue that the computation of contrastive marking involves an island-sensitive movement of an operator.

1 INTRODUCTION The Japanese contrastive marker wa can appear within a because-clause as in (1)1 while it cannot appear within a temporal clause headed by toki ‘when’ as in (2a) (and other temporal clauses; see (20)), and an if-clause (2b).2,3 (1) Itsumo uchi-ni KODOMO-wa kuru node oyatsu-o always house-Dat children-Con come because, sweets-Acc yooi-su-ru prepare-do-Present ‘Because (at least) children come to our house, I always prepare sweets.’ 1

Later in this paper, it will be shown that the distinction of two types of because, transparent and opaque (Kratzer 1998), is crucial, since only opaque because allows embedding of wa. See section 6.3 for details. 2 The same asymmetry is also found in Swada & Larson (2004). 3 Kuroda (2005) uses the following example and the contrastive wa within if is judged grammatical. According to Kuroda (2005), it implies that ‘if at least Nomo had been well, Dodgers would have won, even if others had not been’ (p. 17). (i)

mosi Nomo wa genki dattara, Dodgers ga katta daroo if well were won would ‘if Nomo had been well, Dodgers would have won.’ (Kuroda 2005: 17)

His judgment and reading cannot be replicated by Japanese speakers I have consulted, and I will put this issue aside. The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected].

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Abstract

230 Evidentiality of Discourse Items and Because-Clauses

Interestingly, the same asymmetry is found with the postpropositional evidential morpheme, sooda/soona which follows a finite (inflected) predicate and indicates that the proposition is asserted based on reported evidence.4 (3) a. Kodomo-ga kur-u soona node, oyatsu-o children-Nom come-Pres Evid because, sweets-Acc yooi-shita. preparation-di ‘Because children are coming (I heard), I prepared sweets.’ b. *Kodomo-ga kuru soona toki, oyatsu-o children-Nom come Evid when, sweets-Acc yooi-shita. preparation-did ‘When children are coming (I heard), I prepared sweets.’ c. *Moshi kodomo-ga kuru soonar-a(ba), oyatsu-o if children-Nom come Evid-Comp, sweets-Acc yooi-suru. preparation-do ‘If children are coming (I heard), I will prepare sweets.’ 4 As the editor has pointed out to me, the post-propositional sooda/soona contrasts with the propositional evidential morpheme sooda/soona, which follows a bare predicate and translates as ‘it seems/appears that. . .’. The morpheme sooda/soona can occur under a when-clause and if-clause.

(i) a. Kodomo-ga ki soona node, oyatsu-o yooi-shita. children-Nom come seem because, sweets-Acc prepare-did ‘Because it seemed that children would come, I prepared sweets.’ b. Kodomo-ga ki soona toki, oyatsu-o yooi-shita. children come seem when sweets-Acc prepare-did ‘When it seemed that children would come, I prepared sweets.’ c. Moshi kodomo-ga ki soonar-a(ba), oyatsu-o yooi-suru. if children come seem-Comp sweets-Acc prepare-do ‘If it seems that children would come, I will prepare sweets.’

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(2) a. *Itsumo uchi-ni KODOMO-wa kuru toki, always house-Dat children-Con come when, inu-ga hoe-ru. tea-Acc offer-Present ‘When (at least) children come to our house, the dog always barks.’ b. *Moshi John-ga hon-o 3-SATSU-wa yom-eba, if John-Nom book-Acc 3-Class-Con read-Comp, gookaku-suru. pass-do ‘If John reads (at least) 3 books, he will pass.’

Yurie Hara 231

2 CONTRASTIVE MARKING As noted by Kuno (1973), Japanese contrastive marking involves a morphological marker wa and a prosodic peak in the intonation (indicated by capitals). A sentence with contrastive marking is accompanied by a certain implicature as illustrated in (4b). (4) a. Among John and Mary, who came to the party? b. JOHN-wa kita. John-Con came. ‘John came. (Mary didn’t come, or I don’t know about Mary.)’ In order to capture this intuition, I argued in Hara (2005, 2006, ch. 2) that the prosodic peak of contrastive marking creates a partition of the asserted proposition into B (background) and F (focus) just like question abstracts in the Structured Meaning Approach (cf. von Stechow 1991; Krifka 2001). The morphological wa-marking then introduces the CON operator that takes the structured meaning as its argument and yields a certain interpretation as formulated in (5).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This paper investigates this parallelism between node ‘because’ and other adverbial clauses with respect to compatibility with contrastive and evidential marking and accounts for the data in terms of the evidential component of because-clauses and discourse items. The first section gives a brief overview of Hara (2005, 2006, ch. 2) where it is claimed that the Japanese contrastive marker wa presupposes the speaker’s limited knowledge and induces conventional scalar implicatures. In section 3, I investigate wa-marking in embedded clauses and show that wa-induced implicatures can be relativized to an attitudeholder other than the speaker. In particular, I demonstrate that the definition of wa-marking involves shiftable indexicals in the sense of Schlenker (2003). Section 4 presents in detail the asymmetry sketched above between because-clauses on the one hand and if-clauses and temporal clauses on the other, with some cross-linguistic data from Japanese, English and German. Section 5 summarizes Johnston’s (1994) analysis of the syntactic and semantic differences of various adjunct clauses. Given Johnston’s (1994) analysis, in section 6 I explain the asymmetry of wa-marking and evidentials found for various adjunct clauses in terms of a type mismatch. Finally, in section 7, I discuss the global computation of wa-implicatures in adjunct clauses and argue for syntactic representations for implicature computation (cf. Chierchia 2004).

232 Evidentiality of Discourse Items and Because-Clauses (5) Let w be a world variable, sp the speaker, F the focus-marked elements, B the background, R the restriction. CON(w)(sp)(B(F)) a. asserts: B(F)(w) ¼ 1 b. presupposes: dF#[[F# 2 R] & [B(F#) 0 B(F)] & [B(F) L B(F#)]] There exists B(F#) which is stronger than B(F) c. implicates: "F#[[[F# 2 R] & [B(F#) 0 B(F)] & [B(F) L B(F#)]] /dw#½w# 2 Doxsp ðwÞ½BðF#Þðw#Þ ¼ 0 If there exists B(F#) which is stronger than B(F), it is possible that B(F#) is false.

(6) a. B ¼ kx. x came. F ¼ John F# ¼ John and Mary b. assertion: John came. c. implicates: the speaker considers the possibility that ‘John and Mary came’ is false. d. assertion + implicature: the speaker considers the possibility that ‘Mary came’ is false. This treatment of contrastive marking predicts that if a contrastivemarked proposition, i.e. B(F), is the strongest among its alternatives, the sentence causes a presupposition failure. This prediction is borne out by the following example. In (7), assuming that the quantificational 5 Logical entailment might not be the ideal tool to determine whether a proposition is stronger than the other. We might appeal to the notion of ‘Horn Scale’ which is formed by items that are salient and relevant in the context (Horn 1972a; Gamut 1991). See Hara (2006, ch. 2) for details on this matter.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

First, the assertional content of the contrastive-marked sentence is the proposition obtained by applying the Focus-marked element F to the question abstract B (5a). Second, the contrastive-marked sentence presupposes that there exists a stronger alternative to the asserted proposition (5b), i.e. there is a scalar alternative that entails but is not entailed by the original assertion.5 Finally, if the presupposition is met, the sentence with CON conventionally implicates that the speaker considers the possibility that the stronger alternative is false (5c). In other words, by contrastive marking, the speaker indicates his/her limitation of knowledge with respect to the question under discussion (e.g. ‘Who came to the party?’), i.e. the asserted proposition is his/her maximal knowledge, and as for alternative propositions, the speaker either considers them false or unknown (see Spector [2003] and van Rooij and Schulz [2004] for the notion of order of knowledge). The interpretation of (4b) is depicted in (6).

Yurie Hara 233

domain is not empty, the asserted proposition, ‘Everyone came’, is the strongest among its alternatives. Namely, it entails all of its scalar alternatives, ‘Someone came’, ‘Most people came’ etc., and none of the alternatives entail it. As a consequence, contrastive marking is not compatible with the asserted proposition. (7) #ZEN’IN-wa kita. Everyone-Con came (no implicatures)

(8) ZEN’IN-wa ko-nakat-ta. Everyone-Con come-Neg-Past a. It is not the case that all the people came. (:") b. *All the people are such that they didn’t come. (*":) 3 EMBEDDED CONTRASTIVE In this section, I present data that suggest wa-induced implicatures can be associated with an attitude-holder other than the speaker. I utilize Schlenker’s (2003) notion of ‘shiftable indexicals’ in order to identify the agent of the implicatures in different contexts.

3.1 Relativized implicatures Let us first see what happens to wa in an embedded context. Implicature computation by wa-marking interacts with attitude predicates. In (9), wa can be associated to an attitude-bearer other than the speaker (i.e. John) since wa is embedded within an attitude predicate. Hence, assuming that we only consider Mary and Peter, (9) is ambiguous between a local implicature (the implicature relativized to John) (9a) and the global implicature (the implicature relativized to the speaker) (9b). (9) MARY-wa kita-to John-ga shinjite-iru Mary-Con come-Comp John-Nom believe-Prog ‘John believes at least Mary came.’ (ambiguous)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Similarly, this analysis of contrastive marking makes the correct prediction about the scope inversion fact observed by Bu¨ring (1997). If a sentence contains a universal quantifier and negation as in (8), it can only have :" reading. The ": reading causes a presupposition failure since it is the strongest proposition among alternatives; hence, this reading is not available for (8) (see Hara [2005, 2006, ch. 2] for a comparison with Bu¨ring [1997]).

234 Evidentiality of Discourse Items and Because-Clauses a. Local: The speaker knows [John believes Mary came] Implicature: John doesn’t know whether Peter came] b. Global: The speaker knows [John believes Mary came] Implicature: The speaker doesn’t know [whether John knows that Peter came]

3.1.1 Schlenker (2003) Kaplan (1989) claims that the referent of an indexical is always determined by the context of the actual utterance, which is summarized in the following thesis. (10) Fixity Thesis (a corollary of Direct Reference): The semantic value of an indexical is fixed solely by the context of the actual speech act and cannot be affected by any logical operators. (Kaplan 1989; restatement by Schlenker 2003) For example, in English, the indexical I always refers to the actual speaker of the sentence. Consequently, in order to describe the situation in (11), the subject of the reported speech act has to be referred to with the third-person pronoun he. Example (11b) is not an accurate description of the situation in (11), since English I can only refer to the actual speaker. (11) Situation to be reported: John says: ‘I am a hero.’ a. English: Johni says that hei is a hero. b. English: *Johni says that Ii am a hero. (Schlenker 2003) Observing this fact, Kaplan (1989) claims that no natural language has operators that shift the context that determines the value of indexicals. He says that if such operators existed, they would be monsters. 6 van Rooij & Schulz (2004) also modify their framework in order to generate a desired ‘local’ conversational implicature as observed by Chierchia (2004) and Landman (2000).

(i)

John believes that his colleague makes $100 an hour. a. Local: John believes that his colleague makes not more than $100 an hour. b. Global: It is not the case that John believes that his colleague makes more than $100 an hour.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

As seen in the previous section, the use of wa introduces the operator CON (5). The previous section only looked at cases where the implicature is associated with the speaker. I now claim that if the operator is embedded in an attitude report, the induced implicature can be relativized to the agent of the reported attitude.6 To accommodate this intuition, I modify the denotation of CON so that it contains shiftable indexicals in Schlenker’s (2003) sense.

Yurie Hara 235

As a reply to Kaplan’s observation, Schlenker (2003) argues that ‘every attitude verb is a Kaplanian monster’ (p. 37). In Amharic, e.g., the first-person indexical shifts in attitude reports to the agent of the reported attitude as depicted in (12) (the actual example in Amharic is given in (13)). (12) Situation to be reported: John says: ‘I am a hero.’ Amharic (lit.): Johni says that Ii am a hero. (Schlenker 2003) ˇjegna n -n˜n˜ yil -all (13) ˇjon John hero be.PRT-1sO 3M.say -AUX.3M ‘John says that he is a hero.’ (D. Petros, personal (lit. Johni says that Ii am a hero.) communication to Schlenker)

(14) SAYÆJohn,

now, actuallyæ ci

be-a-hero (agent(ci), time(ci), world(ci)) (Schlenker 2003)

In (14), the context of the reported speech act, ci is bound by the attitude predicate. As a result, in Amharic, -n˜n˜ is interpreted as agent(ci), which refers to the speaker in the embedded context, John. English I is not shiftable, i.e. it can only pick up the actual context (c@), and therefore, it can only be interpreted as the speaker in the context of the actual utterance ([[I]]g ¼ agent(c@)). 3.1.2 Wa-implicatures and shiftable indexicals Following Schlenker’s (2003) approach to indexicals, I reformulate my CON operator as follows. It takes shiftable indexicals, agent(c) and world(c), as its arguments. (15) CON(w(c))(agent(c))(B(F)) a. asserts: B(F)(w) ¼ 1 b. presupposes: There exists B(F#) which is stronger than B(F) c. implicates: dw#[w# 2 Doxagent(c)(w(c))][B(F#)(w#) ¼ 0] In some doxastic worlds accessible to the agent in context c, the stronger alternative is false. Hence, the induced implicature could be associated with an agent other than the speaker.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Schlenker (2003) proposes the following logical structure for the Amharic sentence, in which he treats the semantics of attitude predicates as quantification over contexts. In addition, the embedded clause contains shiftable indexicals, agent(ci), time(ci), world(ci), which are functions from contexts to individuals/times/worlds.

236 Evidentiality of Discourse Items and Because-Clauses Let us go back to the ambiguity of (9) repeated here as (16). Recall that the implicature induced by contrastive marking can be relativized to either the agent of the actual utterance (the speaker) or the agent of the reported attitude (John). (16) MARY-wa kita-to John-ga shinjite-iru Mary-Con come-Comp John-nom believe-Prog ‘John believes at least Mary came.’

(17) a. Local: c@[CP[IP ci[CP CON[XP Mary-wa] came Comp] John-ga believe]] b. Global: c@[CP CON[IP ci[CP[XP Mary-wa] came Comp] John-ga believe]] Let us illustrate how these LF (Logical Form) structures generate different implicatures. I assume that background B is a question predicate in the Structured Meaning Approach (von Stechow 1991) obtained by lambda abstraction using a designated variable (Kratzer 1991b). The operator CON in (17a) takes the embedded IP (Inflectional Phrase). The context of the embedded speech act picks out ‘John’ as the agent of knowledge of the proposition (18b) and generates a local implicature (18c). (18) Computation of the local implicature 1=y a. Bl ¼ ky 2 De.[[Mary1 came]]g;h ¼ ky.came(h1/y(1)) ¼ ky.came(y) b. agent(ci) ¼ j c. CON(w(ci))(j)(came(m)) implicates: In some of the doxastic worlds compatible with John’s belief, it is not the case that Mary and Peter came. On the other hand, in (17b), CON operates over the matrix sentence. As a result, the context of the actual speech act picks out the speaker as the agent/seat/locus of knowledge (19b) and generates a global implicature (19c). (19) Computation of the global implicature 1=y a. Bg ¼ ky 2 De.[[John believes Mary1 came]]g;h ¼ ky.think(j)(came(h1/y(1))) ¼ ky.think(j)(came(y))

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I propose that the operator CON has a syntactic representation and the syntactic location of the operator determines which implicatures are induced. Namely, the syntactic position of the operator determines the attitude-bearer of the induced implicature (the speaker or the subject of the attitude predicate) and the contrasted proposition (matrix or the embedded clause; the size of B).

Yurie Hara 237

b. agent(c@) ¼ sp c. CON(w(c@))(sp)(think(j)(came(m))) implicates: In some of the doxastic worlds compatible with the speaker’s belief, it is not the case that John believes that Mary and Peter came. In summary, the CON operator sitting at a clause-initial position (either embedded or matrix) determines the agent and locus of wa-implicatures.

3.2 Interim summary

4 DATA: PARALLELISM OF ASYMMETRY Japanese contrastive wa cannot appear within adjunct clauses, i.e. temporal clauses and if-clauses. Examples are repeated here in (20) with additional examples of other temporal clauses (before-clauses and afterclauses). (20) a. *Itsumo uchi-ni KODOMO-wa kuru toki, inu-ga always house-Dat children-Con come when, tea-Acc hoe-ru. offer-Present ‘When (at least) children come to our house, the dog always barks.’ * b. Moshi John-ga hon-o 3-SATSU-wa yom-eba, if John-Nom book-Acc 3-Class-Con read-Comp, goukaku-suru. pass-do ‘If John reads (at least) 3 books, he will pass.’ c. *Kinou uchi-ni KODOMO-wa kuru mae, yesterday house-Dat children-Con come before, daremo i-nakat-ta. anyone exist-Neg-Past ‘Before (at least) children came to our house yesterday, no one was home.’

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I have shown that the implicature triggered by wa can be relativized to different agents in an embedded context following Schlenker’s (2003) analysis of attitude predicates as operators that change the context of utterance. I have reformulated my definition of CON so that it contains shiftable indexicals. In the next section, I will present some parallels between contrastive marking and evidentiality and explore the semantic structure of Japanese evidentials.

238 Evidentiality of Discourse Items and Because-Clauses d. *Kinou uchi-ni KODOMO-wa kita ato, yesterday house-Dat children-Con came after minna-de shokuji-o shita. everyone-with meal-Acc did ‘After (at least) children came to our house, we had meal together.’ However, there seems to be an exception to this observation. Wa-marking is available in a because-clause, which is also an adjunct clause.7

A parallel asymmetry is observed for evidential morphemes: the Japanese evidential morpheme soona/sooda indicates that the proposition is uttered based on reported evidence (hearsay evidence). Like wa, the morpheme soona/sooda cannot be embedded under temporal clauses or if-clauses, as we have seen in (3) repeated here as (22) with additional examples of before and after temporal clauses. (22) a. *Kodomo-ga kuru soona toki, oyatsu-o yooi-shita. children-Nom come Evid when, sweets-Acc preparation-did ‘When children are coming (I heard), I prepared sweets.’ b. *Moshi kodomo-ga kuru soonar-a(ba), oyatsu-o if children-Nom come Evid-Comp, sweets-Acc yooi-suru. preparation-do ‘If children are coming (I heard), I will prepare sweets.’ c. *Kodomo-ga kaeru soona mae, watashi-wa children-Nom go.home Evid before, I-Add kaet-ta go.home-Past Intended: ‘Before children went home (I heard), I went home.’ 7 Japanese has two forms for because, node and kara. In this paper, I use node since the literature I cite in this section uses node. As far as the semantic data in this section are concerned, these two forms are interchangeable, although there is an inflectional change with an evidential sooda/soona: soona kara is ungrammatical in standard Japanese (Tokyo dialect), and sooda kara is used instead.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(21) Itsumo uchi-ni KODOMO-wa kuru node oyatsu-o always house-Dat children-Con come because, sweets-Acc yooi-su-ru. prepare-do-Present ‘Because (at least) children come to our house, I always prepare sweets.’

Yurie Hara 239

d. *Kodomo-ga kaetta soona ato, watashi-mo children-Nom went.home Evid after, I-Add kaet-ta go.home-Past Intended: ‘After children went home (I heard), I went home, too.’ In contrast, soona/sooda can be embedded under because-clauses as we have seen in (3a) repeated here as (23).

This asymmetry between temporal clauses and if-clauses on the one hand and because-clauses on the other is not limited to Japanese discourse items but is found cross-linguistically. In the following sections, I highlight some data from English and German.

4.1 English The English adverb obviously indicates an expressive attitude towards a proposition, while the adjective obvious can be analysed as either expressive or propositional. Tredinnick (2004) points out that sentence (24a) is ambiguous. One meaning is that Mary is upset because of the fact that John does not love her, and the speaker comments that it is obvious that John does not love her (some native speakers comment that this reading is harder to get than the other). The other meaning is that Mary is upset over the obviousness of John’s lack of love for her (she might not care whether John actually loves her or not). If we switch the adjective with the adverb obviously as in (24b), only the former reading, namely the speaker’s comment, is available. (24) a. Mary is upset because it is obvious that John doesn’t love her. b. Mary is upset because obviously John doesn’t love her. The adverb obviously is unambiguously expressive. Namely, it indicates that the speaker has some attitude (and perhaps some indirect evidence) towards the embedded proposition. Remarkably, the adverb obviously cannot appear under when (25a) or if (25b), while it can under because as illustrated in (24b).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(23) Kodomo-ga kur-u soona node, oyatsu-o children-Nom come-Pres Evid because, sweets-Acc yooi-shita. preparation-di ‘Because children are coming (I heard), I prepared sweets.’

240 Evidentiality of Discourse Items and Because-Clauses (25) a. *Mary got upset when obviously she failed the exam. b. *Mary will be upset if obviously she fails the exam.

4.2 German Similarly to Japanese evidentials and English obviously, the German discourse particle ja, which indicates the speaker’s assumption that the expressed content might be known to the addressee, can occur within a because-clause but not in temporal clauses or if-clauses.8 nicht not

nicht not

(27) Maria ist a¨rgerlich, weil John sie ja nicht liebt. Maria is angry, because John her JA not love ‘Maria is angry, because John JA doesn’t love her.’ In sum, there exists a cross-linguistic asymmetry regarding embeddability of expressive/discourse morphemes. They can be embedded under because, while they cannot be embedded under temporal clauses and if-clauses. In order to provide an explanation for this asymmetry, I examine the difference between temporal clauses and because-clauses in the subsequent sections. 5 DIFFERENT TYPES OF ADJUNCTS In this section, I review two previous studies on the semantics of adjunct clauses, Johnston (1994) and Kratzer (1991a). Johnston (1994) 8

(i)

The following is Kratzer’s (1999) definition of ja. Ja a is appropriate in a context c if the proposition expressed by a in c is a fact of wc which—for all the speaker knows—might already be known to the addressee. (Kratzer 1999) Kratzer (1999) also shows that it can be relativized to an attitude-bearer other than the speaker if it is embedded within an attitude predicate:

(ii)

Webster sagte, dass er ja nienmanden Webster said that he JA nobody ‘Webster said he hadn’t know anybody.

gekannt habe know had (Kratzer 1999)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(26) a. *Maria wurde a¨rgerlich, als sie die Pru¨fung ja Maria was angry, when she the exam JA bestanden hatte. passed have ‘Maria was angry, when she JA didn’t pass the exam.’ b. *Maria wird a¨rgerlich sein, wenn sie die Pru¨fung ja Maria will angry be, if she the exam JA besteht. pass ‘Maria will be angry, if she JA doesn’t pass the exam.’

Yurie Hara 241

treats the semantics of temporal adjuncts as quantification over event predicates and the semantics of because adjuncts as a relation between propositions. Kratzer (1991a) offers an analysis for if-clauses parallel to Johnston’s analysis for temporal clauses.

5.1 Temporal adjuncts and Because: Johnston (1994)

(28) Marcia always writes a letter when she is at the cafe. Johnston (1994) defines a maximal eventuality function, MAX(/)(e), as follows: ‘MAX(/)(e) is true if and only if e meets the description / and there is no other eventuality meeting that description whose runtime contains the runtime of e.’ (29) MAX(/)(e) ¼ 1 iff[/(e) & ;de#.[/(e#) and (e 6¼ e#) & [f(e) 4 f(e#)]]] (Johnston 1994) Since when needs to apply the maximal event e to the temporal runtime function f, the argument of when must be an event predicate of type Æs, tæ, not a proposition of type t. (30) a. when Marcia was at the cafe b. Marcia was at the cafe 0 ke#.Marcia – was – at – the – cafe#(e#) c. when 0 k/ 2 DÆs, tæki[de.[MAX(/)(e) & i ¼ f(e)]] d. when Marcia was at the cafe 0 ki[de.[MAX(ke#.Marcia-was-at-the-cafe#(e#))(e) & i ¼ f(e)]] (abbreviated as when#e (at#(Marcia, the cafe, e)); f is the temporal runtime function) (Johnston 1994) Following earlier proposals, Johnston (1994) assumes that a temporal clause is always a restriction of an adverb of quantification (AoQ). When the quantification is done by an implicit existential, an episodic

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

According to Johnston (1994), when combines with a property of events and yields a time interval description. That is, the when-clause denotes the time interval which is the temporal runtime of the maximal eventuality denoted by the description in the adjunct clause. This maximality of eventuality is required since atelic eventuality descriptions like ‘she is at the cafe’ can serve as the restrictions on adverbs of quantification as in (28). That is, (28) is interpreted as ‘for each maximal interval for which Marcia is at the cafe, there must be an eventuality of Marcia writing a letter whose runtime is contained within that interval’.

242 Evidentiality of Discourse Items and Because-Clauses reading is derived as in (31a). On the other hand, (31b) is an instance of an overt AoQ. (31) a. Marcia wrote a letter when she was at the cafe. (Episodic) d[when#e (at#(Marcia, the cafe, e1))][write#(Marcia, a letter, e2)] b. Marcia always writes a letter when she is at the cafe. (AoQ) "[when#e (at#(Marcia, the cafe, e1))][write#(Marcia, a letter, e2)] (Johnston 1994)

(32) a. Marty sold his bike because the gears broke. b. because#(de1[sold#(Marty, his bike, e1)], de2[break#(Marty, his bike, e2)]) Johnston (1994) further argues that a because-clause cannot be a restriction of an AoQ since the adverb always cannot quantify over propositions, i.e. the sentence in (33a) does not receive the interpretation in (33b). The clause under because is an existentially closed proposition; hence the because-clause itself does not contain a variable. (33) a. Jane always fixes the car because John wrecks it. b. #"[because#de1[wrecks#(John, the car, e1)]][ﬁx#(Jane, the car, e2)] #All (relevant) events caused by John’s wrecking the car are ones of Jane’s fixing it. In sum, because takes a proposition of type t, whereas when necessarily takes an event predicate of type Æs, tæ.10

5.2 If-clauses: Kratzer (1991a) If-clauses have a structure similar to temporal clauses. According to Kratzer (1991a), an if-clause restricts the domain of adverbial quantification. The sentences in (34) have logical representations in (35). 9

For the purpose of exposition, I delay an explanation of the semantics of because until section 6.3. In (31), the content of the adverbial adjunct (the when-clause) is presupposed, while in (32), the content of the because-clause is not presupposed but asserted. Sawada & Larson (2004) claim that Johnston’s (1994) analysis of the difference in quantificational structures accounts for the difference regarding the presupposition between temporal clauses and because-clauses. This is predicted by the common assumption that the restriction of the quantification is presupposed to be non-empty. 10

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

On the other hand, Johnston (1994) claims that because takes a proposition and expresses a binary relation between two particular events. In other words, because itself does not provide existential quantification; hence, the complement of because needs to be already existentially closed before because applies to it as depicted in (32b).9

Yurie Hara 243

(34) a. Sometimes, if a man buys a horse, he pays cash for it. b. Always, if a man buys a horse, he pays cash for it. c. Most of the time, if a man buys a horse, he pays cash for it. (Kratzer 1991a) (35) a. There is an event e [if e is an event that involves a man buying a horse, then e is part of an event in which this man pays cash for it] b. For all events e [if . . . (e) . . ., then . . . (e) . . .] c. For most events e [if . . . (e) . . ., then . . . (e) . . .] (Kratzer 1991a)

6 ATTITUDES AND EVENT QUANTIFICATION

6.1 The case of wa: the semantic type of B(F) In the previous section, I reviewed Johnston (1994) who claims that when-clauses are restrictions of event quantification, while becauseclauses express a relation between two particular events. Now, the question is how does this semantic difference account for the distribution of wa-marking? As shown in section 2, contrastive marking indicates the limit of some attitude-holder’s knowledge regarding a certain question. That means, the attitude-holder knows of some propositions that they are true. It is not possible to have knowledge of a property of events, i.e. it is not possible to assign a truth-value to a property of events. Therefore, the argument of CON, i.e. B(F), cannot be an event predicate of type Æs, tæ, but must be a proposition of type t. Now, let us go back to the adjunct asymmetry of contrastive marking. Relevant examples are repeated here as (36) and (37). (36) *Itsumo uchi-ni KODOMO-wa kuru toki, inu-ga always house-Dat children-Con come when, tea-Acc hoe-ru. offer-Present Intended: ‘When (at least) children come to our house, the dog always barks.’ (¼(20a)) (37) Itsumo uchi-ni KODOMO-wa kuru node oyatsu-o always house-Dat children-Con come because, sweets-Acc yooi-su-ru. prepare-do-Present

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This suggests that the same line of reasoning for temporal clauses applies to if-clauses. Clauses under if are event predicates Æs, tæ like the ones under when.

244 Evidentiality of Discourse Items and Because-Clauses ‘Because (at least) children come to our house, I always prepare sweets.’ (¼(21)) The local computation of CON for (36) is impossible due to a type mismatch. The CON operator requires a closed proposition as its argument. The interpretation of the IP under when is an event predicate of type Æs, tæ. So, it cannot serve as the argument for CON, which must be of type t.11 (38) *[CP[IP[AdjunctP[CON[IPÆs,

tæ

came(j, e)]] when] . . .]]

(39) [CP[IP[AdjunctP[CON[IPt de.[IPÆs, tæ came(j, e)]]] because] . . .]] In conclusion, the adjunct asymmetry for the embeddability of wa is due to the following semantic difference: temporal clauses and if-clauses involve event quantification, and hence these clauses denote properties of events, while a because-clause expresses a relation between two particular events/situations. CON, as we have seen in sections 2 and 3.1.2, indicates some attitude-holder’s limit of knowledge regarding a proposition; hence, the argument of CON needs to be type t. Therefore, CON cannot appear under when since CON blocks binding of an event variable, which causes an intervention effect. CON is compatible with a becauseclause since the clause under because is interpreted as a proposition.

6.2 Semantics of evidentials: the locus of knowledge I propose that evidentials also take propositions as their arguments. For example, the Japanese evidential morpheme soona/sooda indicates that the proposition is uttered based on some reported evidence (hearsay evidence) as in (40a). The sentence in (40a) is interpreted as (40b) (ignoring tense): the speaker has hearsay evidence x for p, and p is such that there exists an event e, and e is John’s home-going. 11 By ‘AdjunctP’ in (38), I do not mean there is a Projection headed by an adjunct, but I use it to indicate the boundary of adjunct clauses, i.e. temporal clauses, if-clauses and because-clauses 12 It will be shown in section 7 that the global computation of CON in (36) is unavailable due to an island violation.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In other words, if CON appeared under when, the IP would have to be of type t by type-shifting or existential closure. This would cause a semantic crash in the higher computation, i.e. it would result in vacuous quantification. The adverbial quantifier fails to bind a variable.12 In contrast, the local computation for (37) is available since the clause under because is interpreted as a proposition. As a result, wa can be embedded under because and it induces a local implicature.

Yurie Hara 245

(40) a. John-ga kaet-ta-sooda. John-Nom go.home-Past-Evid ‘John went home (I heard).’ b. Evid(x, p) & hearsay(sp, x) & p ¼ de.go.home#(j, e)

(41) John-ga kaetta soona node, watashi-mo kaeru John-Nom went.home Evid because, 1sg-Add go.home koto-ni shita. NMNL-Dat decided ‘Because John went home (I heard), I decided to go home, too.’ (42) because#(de.[go.home#(sp, e)), (Evid(x, p) & hearsay(sp, x) & p ¼ de.go.home#(j, e)) On the other hand, when and if always take an event predicate. Therefore, it is predicted that embedding soona/sooda under when and if is not possible because of a type mismatch. This prediction is borne out, as seen in (43) (see (22) for other examples).13 13 An anonymous reviewer pointed out that embedding English obviously under when and if is sometimes possible as in (i).

(i)

a. If you’ve obviously put so much effort into getting into law school, they should take it into account and admit you. b. Mary got upset when obviously she shouldn’t have.

I take obviously in (ia) as a VP (Verb Phrase) modifying adverb rather than a sentential (evidential) adverb, and hence, it does not cause a type mismatch. Example (ib) is more problematic to my analysis, and it might be necessary to shift the type of when. Indeed, Chris Davis (personal communication) has noted that when in (ib) seems different than a purely temporal when and seems to mean something like ‘even though’.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In what follows, I show how the semantic structure above explains the asymmetry regarding the availability of evidential morphemes under adjunct clauses. As discussed above, there is a clear semantic distinction between because-clauses on the one hand and temporal clauses and if-clauses on the other. Given this distinction, the event quantification analysis of soona/sooda explains why soona/sooda can be embedded under because but not under when. As formalized in (40), soona/sooda existentially closes the event predicate and the sentence expresses that the speaker has hearsay evidence for the particular event denoted by the prejacent clause. Since because takes a proposition, (40a) can be embedded under because as in (41), yielding the formula in (42). The speaker explains the causal relation between two particular events: the event of his/her home-going and the event of John’s home-going, for which he/she has hearsay evidence.

246 Evidentiality of Discourse Items and Because-Clauses (43) *John-ga kaetta soona toki, watashi-mo kaet-ta John-Nom went.home Evid when, 1sg-Add go.home-Past Intended: ‘When John went home (I heard), I went home, too.’ (¼(3b)) Kratzer (1999) also points out the same phenomenon for the German particle ja. As (44) and (45) show, ja cannot be embedded under when.

(45) Als ich (*ja) in Syracuse gewohnt habe, war ich oft When I JA in Syracuse lived have, was I often in Ithaca in Ithaca ‘When I JA lived in Syracuse, I was often in Ithaca.’ (Kratzer 1999) Kratzer (1999) gives an explanation which is parallel to the current proposal: ‘Since the scope of a discourse particle has to express a proposition, the scope of a discourse particle cannot include pronouns that are bound from outside. That is, no discourse particle can intervene between a bound variable pronoun and its binder.’ In short, the interpretations of when-clauses in (44) and (45) contain event variables; hence, they are properties of events. Since ja expresses some attitude towards a particular event, it cannot operate over predicates.14

14

Kratzer (1999) goes on to show the following seeming exception to her analysis. Since ja needs to take a proposition, it blocks binding. However, attitude-holders can bind into a clause which is in the scope of ja as in (i). (i)

Jeder der Zeugen behauptete, er habe ja mit eigenen Augen gesehen, dass . . .. Each witnesses claimed he had JA with own eyes seen that ‘Each of the witnesses claimed he had JA seen with his own eyes that. . ..’ (Kratzer 1999)

In (i), the expressive meaning induced by ja is attributed to the reported attitude situation, rather than the actual utterance situation. More specifically, the assumption that the content of the embedded clause ‘he had seen with his own eyes that. . .’ might be already known to the addressee is anchored to the agent of the reported utterance ‘each of the witnesses’. von Fintel & Iatridou (2002) explain this interpretation by analysing the pronoun in the complement clause in (i) as a logophor or a shiftable indexical. Namely, the pronoun er ‘he’ is not bound by the quantifier of the matrix subject, but it refers to the agent of the reported utterance. See Hara (2006, ch. 4) for further discussion on this topic.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(44) *Maria wurde a¨rgerlich, als sie die Pru¨fung ja nicht Maria was angry, when she the exam JA not bestanden hatte. passed have ‘Maria is angry, when she JA didn’t pass the exam.’ (¼(26a))

Yurie Hara 247

6.3 Extension 1: opaque and transparent Because

(46) a. I fell because the principal did. (transparent) b. I went to the pageant because the principal did. (opaque) (Kratzer 1998) All the examples with because in the sections above use an opaque because; they are the speaker’s or some attitude-holder’s explanation of the relation between two propositions. Unlike an opaque because, the two conjuncts of a transparent because are not connected by the speaker’s reasoning but are simply in a causal relation. To illustrate with a Japanese example, node in (47a) is a transparent because; the two conjuncts of a transparent because are not particular events connected by someone’s reasoning but are event predicates simply combined by an event quantification and a causal relation as depicted in (47b). (47) a. ame-ga futta node, kion-ga sagat-ta. rain-Nom fell because because, temperature-Nom down-Past ‘Because it rained, the temperature went down.’ (transparent because) b. de.[down#(temp, e) & de#.[cause(e#, e) & rain#(e#)]] Now, if a wa-marked element is embedded under a clearly transparent because as in (48b), the sentence turns out to be ungrammatical. The ungrammaticality of (48b) can be explained along the same line as the case of when: in the complement of transparent because, the event predicate is not existentially closed, so locally the argument of the CON operator is not a closed proposition of type t. (48) a. KOOCHOO-SENSEI-wa geki-ni it-ta node, principal-Con pageant-to go-Past because,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Johnston’s proposal regarding the semantic difference between temporal clauses and because-clauses can be extended to the two interpretations of because observed by Davidson (1967) and Kratzer (1998): a singular causal statement (transparent because) and a causal explanation (opaque because). A singular causal statement expresses a relation between events. For example, the sentence in (46a) expresses a scenario where the principal fell and also knocked down the speaker. On the other hand, a causal explanation expresses a relation between propositions, which is set up by an inference made by some attitude-holder. The sentence in (46b) gives the reason for the speaker’s action of going to the pageant in virtue of certain properties of the expressed events.

248 Evidentiality of Discourse Items and Because-Clauses boku-mo it-ta. 1sg-Add go-Past ‘Because the principal went to the pageant, I also went to it.’ (opaque because) * b. sakki AME-ga/ wa futta node, kion-ga while-ago rain-Nom/Con fell because, temperature-Nom sagat-ta. down-Past ‘Because it rained a while ago, the temperature went down.’ (transparent because)

(49)

*It got cooler because obviously it rained.

(50) a. *ame-ga futta soona node, kion-ga sagat-ta. rain-Nom fell Evid because, temperature-Nom down-Past Intended: ‘Because it rained (I heard), the temperature went down.’ b. Koochoo-sensei-ga geki-ni it-ta soona node, principal-Nom pageant-to go-Past Evid because, boku-mo it-ta. 1sg-Add go-Past ‘Because the principal went to the pageant (I heard), I also went to it.’ This shows that a transparent because patterns just like when. A transparent because merely denotes a causal relation between two events, 15 Unfortunately, this story cannot be straightforwardly extended to German ja, since the German translation of (49) with ja is acceptable.

(i)

Es ist kalt, weil das Fenster ja offen it is cold because the window JA open ‘It is cold because the window JA is open.’

ist. is

Presently, I do not have a convincing explanation for this difference between German ja on one hand and Japanese soona and English obviously on the other.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Similarly, having an evidential under a transparent because is predicted to be ungrammatical. The semantics of the transparent because consists of an event quantification and a causal relation (47b). Having an evidential would existentially close the event predicate and there would be no event variable to be quantified over, resulting in a type-mismatch or vacuous quantification. This is indeed the correct prediction as witnessed in (49) for English and also in (50a), while the opaque counterpart is grammatical.15

Yurie Hara 249

while an opaque because connects two propositions by some attitudebearer’s reasoning. It is the point-of-view-ness that is crucial for hosting wa.

6.4 Extension 2: Because and evidentials as context-shifters

6.4.1 Because is a context-shifter It is the speaker’s or some attitudebearer’s reasoning that connects the two conjuncts of opaque because; therefore, there is some representation of point of view in the complement of because parallel to that of an attitude-operator. Hence, the agent of reasoning introduced by opaque because can be an attitudeholder other than the speaker (from this section on, all instances of because in the examples are opaque because). In (51), for example, it is the president who draws the inference that connects the two conjuncts. Hence, the implicature induced by wa ‘Possibly, John doesn’t speak other languages’ is also associated with the president. (51) Shachoo-wa John-ga NIHONGO-wa dekiru-node, president-Top John-Nom Japanese-Con capable-because, saiyoo-shi-ta. hire-do-Past ‘Because John can speak (at least) Japanese, the president hired him.’ Suppose that the company is looking for someone who can speak either Japanese or Korean. The use of wa in (52a) indicates some

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This section argues that contrastive marking, evidentials and the because operator all take a proposition as their argument and express some attitude towards the event or situation that the proposition denotes. Contrastives express the fact that the information denoted by the embedded proposition is the most informative answer that the speaker (or some other locus of the knowledge) has. Evidentials indicate the source of the speaker’s or some other attitude-holder’s belief of the embedded proposition. The because operator expresses the speaker’s or some attitude-bearer’s inference about the connection between two propositions. These expressions interact with the context of the utterance. In the following, I adduce further evidence for the evidentiality of the because operator and the discourse item CON and sooda/soona, by briefly showing that evidentials and the because operator set up a context for the embedded proposition just like attitude predicates do. See Hara (2006, ch. 4) for a detailed discussion on this topic.

250 Evidentiality of Discourse Items and Because-Clauses attitude-holder’s limited knowledge and generates an implicature ‘Possibly, John doesn’t speak Korean’. If the agent of this implicature were the actual speaker, the continuation in (52b) would be infelicitous, since the speaker has complete knowledge.16

The implicature is relativized to the shifted context that assigns the president as the locus of knowledge as depicted in (53). (Here, I slightly modify Johnston’s analysis of because as a relation between events and make it a relation between situations.) (53) a. [CP[IP[AdjunctP OpÆpresidentæ ci CON ds.[IP john-capable-ofJapanese(s) ] because] . . .]] 16 We can find an actual example in which the agent of the reasoning must be the speaker with German causal denn. German has two causal connectives, denn and weil. Unlike weil, denn seems to unambiguously indicate the speaker’s reasoning, since (ia) is understood as a contradiction.

(i) a. #Der Firmenleiter hat sich entschieden Mary einzustellen, denn sie The company.boss has SELF decided Mary hire because she spricht Holla¨ndisch. Aber ich glaube Mary spricht kein Holla¨ndisch. speaks Dutch. But I believe Mary speaks no Dutch ‘The president decided to hire Mary because she speaks Dutch. But I don’t think Mary speaks any Dutch. b. Der Firmenleiter hat sich entschieden Mary einzustellen, weil sie The company.boss has SELF decided Mary hire because she Holla¨ndisch spricht. Aber ich glaube Mary spricht kein Holla¨ndisch. speaks Dutch. But I believe Mary speaks no Dutch ‘The president decided to hire Mary because she speaks Dutch. But I don’t think Mary speaks any Dutch.’ Scheffler (2005) analyses the causal relation expressed by denn as a conventional implicature. (ii) In a sentence ‘A, denn B’, with [[A]] ¼ / and [[B]] ¼ w, denn has the following semantics: Assertion: / ^ w Conventional Implicature: CAUSE(/, w) If Scheffler’s (2005) analysis is right, the assertion of the first sentence in (ia) entails that the speaker believes ‘Mary speaks Dutch’ is true. Hence, the continuation leads to a contradiction. Furthermore, Potts (2003) claims that conventional implicatures are always speaker oriented. It follows that the causal relation in (ia) is attributed to the speaker. On the other hand, the causal relation denoted by weil is part of the assertion. Therefore, the content of the weil-clause does not need to be believed by the speaker, and the reasoning can be attributed to the president.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(52) a. Shachoo-wa John-ga NIHONGO-wa dekiru-node saiyoo-shi-ta. ‘Because John can speak (at least) Japanese, the president hired him.’ b. Demo, John-wa jitsuwa kankokugo-mo dekiru. But, John-Top actually Korean-Add capable ‘But, actually, John can speak Korean, too.’

Yurie Hara 251

b. agent(ci) ¼ the president c. CON (w(ci))(agent(ci))(ds.john-capable-of-Japanese(s)) implicates: In some of the doxastic worlds compatible with the president’s belief, it is not the case that John speaks other languages. In sum, the use of because introduces a new context that binds the context variable of the shiftable indexicals in CON.17

(54) John-ga kaet-ta sooda. John-Nom go.home-Past Evid ‘John went home (I heard).’

(¼(40))

Evidentials can provide an appropriate context for direct experience predicates. The Japanese adjectives of direct experience such as samui ‘to feel cold’, sabishii ‘to feel sad’, etc., restrict their subjects to the first person (Kuno 1973; Kuroda 1973; Aoki 1986; Tenny 2006). 17

In Hara (2006, ch. 4), I provide support for the claim that because introduces an attitudeoperator that can shift contexts by examining some data pointed out by Tenny (2006) on direct experience and long-distance reflexives.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

6.4.2 Evidentials shift contexts The semantics of evidential morphemes has not received any formal treatment until recently and the semantic contribution that evidential morphemes make is still controversial. The current analyses suggest that Evidentials introduce a local context. For example, Izvorski (1997) claims that indirect evidentiality presupposes that the speaker has indirect evidence. Following Kratzer’s (1987) standard analysis of modality, Izvorski (1997) treats the semantics of the indirect evidential as quantification over possible worlds, where the presupposition restricts the modal base to the propositions that can be inferred by the indirect evidence. On the other hand, Faller (2002) analyses evidentials as speech act modifiers. In both analyses, an evidential morpheme seems to express a relation between the speaker and the proposition to which the evidential attaches. More specifically, an evidential sets up a context where the truth of the embedded proposition holds. I propose that Japanese evidentials function like attitude reports and the because operator (see also McCready 2006). Specifically, evidentials are attitude-operators that bind context variables of shiftable indexicals. For example, the use of the hearsay evidential sooda in (40), repeated here as (54), introduces an attitude-operator that indicates that the truth of the embedded proposition is based on hearsay evidence.

252 Evidentiality of Discourse Items and Because-Clauses (55) a. Watashi/*anata/*kare-wa samui I/you/he-Top cold ‘I am/*you are /*he is cold.’ b. Watashi/*anata/*kare-wa sabishii I/you/he-Top sad ‘I am/*you are /*he is sad.’

desu. Cop desu. Cop (Tenny 2006)

In (56b), the evidential morpheme sooda lifts the person constraint on direct experience, suggesting that the evidential provides a local context which changes the agent/locus of knowledge from the speaker to ‘John’, who provided hearsay evidence to the speaker.

Now, let us examine how the evidential sooda interacts with contrastive marking. Consider (57), identical to (55) except that the subject is marked with the contrastive morpheme wa. (57) JOHN-wa kaet-ta sooda. John-Con go.home-Past Evid ‘(At least) John went home (I heard).’ Remember from sections 2 and 3 that contrastive marking indicates that the locus of knowledge does not have the maximal knowledge with respect to the property in question, and the agent can be shifted if contrastive marking is embedded under an attitude-operator. If sooda is an attitude-operator that specifies the agent of the knowledge as someone other than the speaker, the implicature induced by wa should be attributed to the agent specified by the shift in the context induced by sooda. Indeed, in (57), the implicature can be associated to the evidence the speaker has rather than the speaker, as in (58c). (58) a. c@[ci CON[IP de.[went.home(j, e)]] sooda] b. agent(ci) ¼ hearsay evidence c. CON(w(ci))(agent(ci))(de.went.home(j,e)) implicates: In some of the doxastic worlds compatible with the hearsay evidence the speaker has, it is not the case that other people went home.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(56) a. *John-wa samui. John-Top cold. ‘John is cold.’ b. John-wa samui sooda. John-Top cold Evid. ‘John is cold (I heard).’

Yurie Hara 253

(61) a. Did Mary and Peter pass the exam? b. John niyoruto MARY-wa ukat-ta-soode, Bill niyoruto John according.to Mary-Con pass-Past-Evid, Bill according.to PETER-mo ukat-ta-sooda. Peter-Add pass-Past-Evid ‘According to John, Mary passed (I heard), and according to Bill, Peter passed, too (I heard).’

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This intuition is further supported by the following examples. Recall from section 2 that contrastive marking cannot be used when the speaker’s knowledge is the strongest among alternatives (when all the individuals are in the extension of the property) as in (59b). The same explanation applies to the infelicity of (59c). The shifted locus of the knowledge (the hearsay evidence) cannot have the maximal knowledge with respect to the property in question. (59) a. Mary-to Peter-wa shiken-ni ukat-ta-no? Mary-and Peter-Top exam-Dat pass-Past-Q ‘Did Mary and Peter pass the exam?’ b. ??MARY-wa ukat-te, PETER-mo ukat-ta. Mary-Con pass-and, Peter-Add pass-Past ‘Mary passed and Peter passed, too.’ c. ??MARY-wa ukat-ta-soode, PETER-mo ukat-ta-sooda. Mary-Con pass-Past-Evid, Peter-Add pass-Past-Evid ‘Mary passed (I heard) and Peter passed, too (I heard).’ Interestingly, (59c) is improved if one of the evidential markers is removed as follows. (60) a. Did Mary and Peter pass the exam? b. MARY-wa ukat-ta-soode, PETER-mo ukat-ta. ‘Mary passed (I heard), and Peter passed, too.’ c. MARY-wa ukat-te, PETER-mo ukat-ta-sooda. ‘Mary passed, and Peter passed, too (I heard).’ This contrast is not surprising, since unlike (59c) the propositional content of each conjunct of (60b) relies on different agents of knowledge. For example, the first conjunct of (60b) implicates that according to the hearsay evidence, it is possible that Peter did not pass. The second conjunct of (60b) entails that the speaker believes that Peter passed. These interpretations do not contradict each other. Similarly, (59c) can also be improved by specifying a different source of evidence overtly as in (61b). Again, because each conjunct has a different attitude-holder for the asserted content and the implicature, their interpretations do not cause a contradiction.

254 Evidentiality of Discourse Items and Because-Clauses In this section, I argued that the because operator and the evidential morpheme sooda are attitude-operators that shift the context of utterance just like attitude predicates do.18 Both because and sooda shift the agent of the wa-implicature.

6.5 Interim summary

7 GLOBAL IMPLICATURE COMPUTATION BLOCKED BY SYNTAX The previous section revealed that the local computation of CON under temporal and if-clauses is not permissible due to a type mismatch. Now, remember from section 3.1 that the sentence in (62) has two possible LFs, shown in (63); the CON operator can be placed at matrix or embedded clause-initial position. (62) MARY-wa kita-to John-ga shinjite-iru Mary-Con come-Comp John-Nom believe-Prog ‘John believes (at least) Mary came.’ (ambiguous) (63) a. Local: c@[CP[IP ci[CP CON[XP Mary-wa] came Comp] John-ga believe]] b. Global: c@[CP CON[IP ci[CP[XP Mary-wa] came Comp] John-ga believe]] What prevents the LF of (2a), repeated here as (64), from having the structure in (65) and inducing a global implicature outside the whenclause? 18 19

Indeed, Tenny (2006) argues that Japanese node ‘because’ is an evidential marker. See Hara (2006, ch. 6; forthcoming) for the discussion of relative clauses.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Both contrastive marking and evidentials express some attitude towards a particular event or situation. Hence, their arguments cannot be predicates but must be propositions. Temporal clauses and if-clauses are properties of events that contain an event variable bound from outside. The data above show that embedding of attitude expressions such as wa and evidential morphemes blocks binding of an event variable, which is required for the semantics of these clauses, i.e. event quantification.19 On the other hand, the because operator takes a proposition just like CON and evidentials do. Hence, those discourse morphemes such as Japanese wa, English obviously and German ja are compatible with (opaque) because-clauses.

Yurie Hara 255

(64) *Itsumo uchi-ni KODOMO-wa kuru toki, inu-ga always house-Dat children-Con come when, tea-Acc hoe-ru. offer-Present ‘When (at least) children come to our house, the dog always barks.’ (65) *c@ CON[CP[IP[AdjunctP ci[XP John-wa] came when] . . .]]

(66) Itsumo uchi-ni KODOMO-wa kuru node oyatsu-o always house-Dat children-Con come because, sweets-Acc yooi-su-ru. prepare-do-Present a. *‘(at least) Children are such that because they comes to our house, I always prepare sweets.’ b. ‘Because (at least) children come to our house, I always prepare sweets.’ (67) a. *c@ CON[CP[IP[AdjunctP[XP John-wa] came because] . . .]] b. c@[CP[IP[AdjunctP ci CON[XP John-wa] came because] . . .]] In addition to adjunct clauses, contrastive marking is not available within a relative clause: (68) *Itsumo amerika-de CHOMSKY-wa kai-ta hon-ga always America-at Chomsky-Con write-Past book-Nom shuppan-sa-re-ru. publish-do-Pass-Present ‘The book which (at least) Chomsky wrote in America is always published.’ In the syntactic literature, these constructions are known to be islands for wh-movement. Hence, it seems that contrastive marking is constrained by the syntax. More specifically, it seems that the association between CON and the Focus-marked element cannot be established across a syntactic island. Observing these facts, this section

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

A closer look at the interpretation of (21) repeated here as (66) reveals that the CON operator introduced by wa cannot be positioned outside of the because-clause, since the global computation of the waimplicature is not available for (66). The LF structure in (67a) is not available; hence, the only legitimate reading for (66) comes from the local computation of the wa-implicature from the LF in (67b).

256 Evidentiality of Discourse Items and Because-Clauses argues for movement of the CON operator by showing that contrastive marking is sensitive to island effects. To better understand this phenomenon, I first give an overview of islands for wh-movement in Japanese, and then I compare the structure of Japanese wh-questions proposed by Nishigauchi (1990) with the distribution of Japanese contrastive marking.

7.1 Wh-islands in Japanese

(69) *[Kare-ga naze kai-ta hon]-ga omosiroi-desu-ka? he-Nom why write-Past book-Nom interesting-is-Q ‘Why are books that he wrote t interesting?’ In this LF-movement approach, it is difficult to understand why some Japanese wh-words can appear within adjunct islands (70) and complex NP Islands (71). (70) Mary-wa [John-ga nani-o yomu mae-ni] dekaketa no? Mary-TOP John-NOM what-ACC read before left Q ‘Mary left before John read what?’ (Pesetsky 1987: 110) (71) kimi-wa [dare-ga kai-ta hon-o] yomi-masi-ta ka? you-TOP who-NOM wrote book-ACC read.POL-PAST Q ‘You read books that who wrote?’ (Nishigauchi 1990: 40) To save the LF-movement approach, Nishigauchi (1990) argues for LF pied-piping (see also Choe [1987] and Pesetsky [1987]; see Hagstrom [1998] for a different approach). In Nishigauchi’s (1990) approach, what actually moves covertly is not the wh-word, but the island that contains the wh-word. For example, (72a) has the LF structure in (72a) instead of (72b). 20 Also, Kikuchi (1987) has shown that Japanese Comparative Deletion involves operator movement, which is sensitive to island constructions. See Kikuchi (1987) for detailed discussions.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Japanese is a wh-in situ language in view of Huang’s (1982a, 1982b) theory of wh-movement. That is, wh-words move covertly to clauseperipheral positions at LF.20 For example, naze ‘why’ cannot appear within a complex NP as in (69). In the LF-movement approach, this is understood as follows. Even though naze is in the base-generated position in overt syntax, it moves to the clause-initial position in covert syntax, which violates the complex NP island constraint Ross (1967).

Yurie Hara 257

(72) a. ½CP ½ who-Nom wrote book -Acci ½IP ½VP ti read Q [ b. *½CP who-Nom ½IP ½VP ½ ti wrote book -Acci read Q [

(73) John-wa [Mary-ga nani-o katta-ka] imademo John-Top Mary-Nom what-Acc bought-Q still shiri-tagat-teiru-no? know-want-Prog-Q a. ‘Does John still want to know what Mary bought?’ b. ?‘What1 is such that John still wants to know [whether Mary bought it1]?’ (Deguchi & Kitagawa 2002) According to Deguchi & Kitagawa (2002) and Ishihara (2002), this seeming wh-island effect in Japanese reported in earlier literature is a misinterpretation of the preference towards a non-monotonic prosody.21 (74) John-wa [Mary-ga NAni-o katta-ka] imademo John-Top Mary-Nom what-Acc bought-whether still shiri-tagat-teiru-no? know-want-Prog-Q ‘Does John still want to know what Mary bought?’ Deguchi & Kitagawa (2002) and Ishihara (2002) propose a prosodicsensitive association of the wh-word and the Q-morpheme and show that the global association in (73b) becomes much more readily available if the post-focal reduction continues to the sentence-final Qmorpheme (Global Emphatic Prosody [Global EPD] in the 21 I use italics to indicate the post-focal reduction. See Deguchi & Kitagawa (2002) or Ishihara (2002) for more a precise representation of the prosody patterns.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Hence, although it appears that Japanese wh-questions do not obey a general constraint that prohibits movement across adjunct and complex NP islands, the acceptability of the construction is due to amelioration by LF pied-piping. On the other hand, it has been claimed that a wh-word inside a whisland is not acceptable (Nishigauchi 1990; Watanabe 1992; Shimoyama 2006), since there seems to be a preference towards the local association of the wh-word with the embedded Q-morpheme -ka over the global association with the matrix -ka.

258 Evidentiality of Discourse Items and Because-Clauses terminology of Deguchi & Kitagawa and Focus Intonation [FI] in Ishihara’s terminology) as in (75). (75) John-wa [Mary-ga NAni-o katta-ka] imademo John-Top Mary-Nom what-Acc bought-whether still shiri-tagat-teiru-no? know-want-Prog-Q ‘What1 is such that John still wants to know [whether Mary bought it1]?’ (Deguchi & Kitagawa 2002)

7.2 Wa-marking and islands In this section, I demonstrate that wa-marking has a parallel distribution to Japanese wh-questions in terms of embedding under islands. First, consider the case of adjunct clauses. When wa-marking is under temporal and if-clauses, the global computation of waimplicatures (e.g. (65)) is somehow blocked as we have seen in (2). However, in temporal clauses, contrastive marking can be rescued by changing the construction so that it has a structure parallel to the pied-piped wh-question. More specifically, if wa is attached to the whole island leaving the Focus-marking on the contrasted argument, the sentence becomes acceptable with the desired global implicature (‘For other people, I don’t know whether the dog barks when they come (76a)).22 22

(i)

Unfortunately, a contrastive within an if-clause cannot be saved by pied-piping. *[Moshi John-ga hon-o 3-SATSU yomu-nara]-wa, if John-Nom book-Acc 3-Class read-Comp-Con, ‘If John reads (at least) 3 books, he will pass (the exam).’

I do not have an explanation for this difference at moment.

goukaku-suru. pass-do

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Deguchi & Kitagawa (2002) and Ishihara (2002) attribute the preference for local wh-scope observed in (73b) to the shorter postfocal reduction (local EPD or FI) as depicted in (74), which is preferred due to a tendency to avoid monotonic prosody. Therefore, a Japanese embedded wh-question does not constitute an island for a matrix whquestion if the correct prosody is assigned to the question. In summary, in Japanese, a wh-word moves at LF, and as a consequence it obeys island constraints. The seeming exception of an adjunct island or a complex NP island is shown to be the result of LF pied-piping of the whole island. On the other hand, an embedded whquestion does not constitute an island in Japanese.

Yurie Hara 259

(76) a. Itsumo [uchi-ni JOHN-ga kuru toki]-wa, inu-ga always house-Dat John-Nom come when-Con, tea-Acc hoe-ru. offer-Present ‘At least when John comes to our house, the dog always barks.’ b. Kinou [JOHN-ga uchi-ni kuru mae]-wa, daremo i-nakat-ta. ‘At least before John came to our house, no one was home.’ c. Kinou [JOHN-ga uchi-ni kita ato]-wa, minna-de shokuji-o shita. ‘At least after John came to our house, we had meal together.’

(77) Itsumo [amerika-de CHOMSKY-ga kai-ta hon]-wa always Amerika-at Chomsky-Nom write-Past book-Con shuppan-sa-re-ru. publish-do-Pass-Present ‘At least the book which Chomsky wrote in America is always published.’ Now, let us turn to wh-islands. Wa-marking seems to be available in wh-islands: (78) boku-wa ano-mise-de JOHN-wa nani-o kat-ta ka kii-ta. I-Top that-shop-at John-Con what-Acc buy-Past Q ask-Past ‘I asked what (at least) John bought at that shop.’ As discussed by Deguchi & Kitagawa (2002) and Ishihara (2002), a Japanese embedded wh-question does not constitute an island for a matrix wh-question if the correct prosody is assigned to the question. For this reason, I do not take (78) as a counter-example to my generalization. In short, wa-marking is not available within adjunct and complex NP islands. However, most of the constructions (except for if-clauses) can be improved by overt pied-piping-like structures. In other words, it is possible to contrastive-mark an element within islands and obtain global implicatures if the wa morpheme is realized at the edge of those domains that can undergo pied-piping in wh-questions. In addition,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Contrastives also observe complex NP islands. Wa-marking cannot be used for NPs within relative clauses as seen in (68). Similar to adjunct islands, (68) can be improved if the contrastive morpheme wa is realized at the edge of the complex NP island as in (77).

260 Evidentiality of Discourse Items and Because-Clauses wa-marking is possible within a wh-island. Overall, the distribution of contrastive wa-marking is parallel to the distribution of Japanese whquestions.

7.3 Movement of Con

(79) John-ga MARY-wa kuru-to omot-ta toki, kanojo-ga John-Nom Mary-Con come-Comp think-Past when 3sg-Nom heya-ni haitte kita. room-Dat in come-Past ‘When John thought that at least Mary comes, she came into the room.’ Hence, observing the data shown in section 7.2, I propose a syntactic movement account for this problem. The CON operator is originally generated locally as in (80) and moves to yield the LF structures which determine which attitude-bearer, the speaker or the subject of the attitude predicate, is responsible for the induced implicature.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The data shown above suggest that the placement of the CON operator is constrained by syntactic islands. One might speculate that there could be some principle which simply restricts wa from being embedded within adjunct clauses or complex NPs. Such a speculation might come from the fact that contrastive wa is homophonous with thematic wa, which is claimed to mark the topic in information structure (Vallduvı´ 1992; Heycock 1993) and appears at left-peripheral position. Indeed, contrastive wa is called ‘contrastive topic’ by some authors. I agree that contrastive wa and thematic wa are homophonous not by accident and that contrastive wa also involves the notion of information structure and left periphery (e.g. I propose that the CON operator is placed at a clause-initial position.). However, a more sophisticated explanation is needed because the simple stipulation that bans wa from appearing in adjunct clauses and complex NPs makes the wrong empirical prediction when wa is embedded under an attitude predicate. For example, in (79), even though the wa-marked element is within an island construction, the sentence is judged grammatical. This is unexpected if we assume that wa can only appear in the matrix clause. A better explanation for (79), one that correctly explains the grammatical judgment, is that a wamarked element needs to be local to an attitude-bearer (the speaker or the subject of the attitude predicate).

Yurie Hara 261

(80)

Let us go back to the ambiguity of (9) repeated here as (81). (81) MARY-wa kita-to John-ga shinjite-iru Mary-Con come-Comp John-nom believe-Prog ‘John believes at least Mary came.’ (¼(9))

(82)

global Y c@ CON ½CP ½IP ci ½CP CON ½XP t Mary-wa came Comp John-ga believe [ local

A movement analysis straightforwardly explains the ungrammaticality of (2a) repeated here as (83). The context variables of the CON operator generated under when need to be saturated. However, there is 23 Note that the contrastive operator CON does not form a constituent with the contrastivemarked NP Mary at LF, where scope is computed. The configuration like (80) is necessary for the following reason. Consider sentence (i), in which the wa-marked quantifier zen’in ‘everyone’ is embedded in the complement clause.

(i)

Zen’in-wa kur-u-to omow-anakat-ta. Everyone-Con come-Present-Comp think-Neg-Past ‘At least, I didn’t think that everyone would come.’ (Implicature: I thought someone would come.)

The local implicature is impossible since ‘Everyone comes’ does not satisfy the presupposition of CON as we have seen in section 2. On the other hand, if the operator formed a constituent with the quantifier and moved to the matrix clause along with it, the syntax would yield an LF-structure, "x:think([person(x)][come(x)]), which again fails to satisfy the presupposition, since the assertion exhausts all the individuals in the domain. Hence, it fails to induce the implicatures required by wa. Nonetheless, (i) is acceptable; therefore, we have to allow the global computation of a waimplicature without moving the quantifier zen’in ‘everyone’. If CON alone is placed in the sentenceinitial position (but leaving the quantifier in situ), :think("x[person(x)][come(x)]) indeed has an implicature, ::think(dx[person(x)][come(x)]) ‘I thought some people would come’. Therefore, the contrastive operator CON is detachable from the contrastive-marked element as depicted in (ii), which is a structural representation of (i). (ii) ½CP CON ½NegP ½CP ½XP t ½XP everyone -wa came Comp think Neg Past [

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

If the operator moves to the edge of the embedded clause, it induces the local implicature ‘John considers the possibility that Peter didn’t come’. If it moves to the edge of the matrix IP, it induces the global implicature ‘The speaker considers the possibility that John doesn’t believe Peter came’.23

262 Evidentiality of Discourse Items and Because-Clauses no attitude predicate under when; hence, it targets the matrix clause. This movement crosses an adjunct island. (83) *Itsumo uchi-ni JOHN-wa kuru toki, inu-ga hoe-ru. always house-Dat John-Con come when, tea-Acc offer-Present ‘When (at least) John comes to our house, the dog always barks.’ (¼(2a))

(85) John-ga MARY-wa kuru-to omot-ta toki, kanojo-ga John-Nom Mary-Con come-Comp think-Past when 3sg-Nom heya-ni haitte kita. room-Dat in come-Past ‘When John thought that (at least) Mary comes, she came into the room.’ (¼(79)) The CON operator does not need to cross an island since it can find a local attitude-operator that binds its context variable. (86) ½IP . . . ½AdjunctP ½VP ci CON ½CP ½IP ½XP t ½XP Mary -wa come Comp thought when . . . [

7.4 Arguments for a movement approach The introduction of a syntactic movement operation to account for a semantics–pragmatics phenomenon like implicature computation may seem unconventional. In general, semantic associations such as focus associations (Rooth 1985, 1992) and choice function binding (Reinhart 1997) are immune to islands. Moreover, contrastive marking an argument within an island per se should be acceptable on semantic grounds alone, since there are other ways to express the intended meaning. There are two ways to ameliorate the constructions in (2), in which wa-marking occurs within adjunct clauses: one is pied-piping the contrastive marking to outside of the island as shown in section 7.2, and the other is base-generating the contrastive-marked element at the clause-initial position. In the following section, I demonstrate

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(84) *½SpeechActP ½speaker ½ Con ½IP . . . ½AdjunctP ½IP t John-wa come when [ When wa is further embedded under an attitude predicate, the sentence is acceptable even within an island ((79), repeated here as (85)).

Yurie Hara 263

specifically how the sentences in (2) are ameliorated and how they are interpreted.24 7.4.1 Pied-piping First, if wa is attached to a when-clause and the when-clause contains an argument that bears a focus marked by a sentential stress, it is possible to compute a global implicature. In (76a), e.g. the when-clause ‘JOHN-ga kita toki’ contains the argument NP John, which has a focus-marking dissociated from wa. The wa is attached to this when-clause and implicates ‘I don’t know if it’s true that when other people come to our house, the dog always barks’.

(87) JOHNi-wa itsumo uchi-ni proi kuru toki, inu-ga John-Con always house-Dat pro come when, dog-Nom hoe-ru. bark-Present ‘At least Johni is such that when proi comes to our house, the dog always barks.’ (Implicature: As for Mary, it might not bark.) Since CON is generated outside the island, it does not cross the island in order to be bound by the actual context. (88) c@ CON ½XP t ½ John -wa always ½AdjunctP pro came toki dog-ga (¼(87)) barks[ Together with the pied-piping facts presented in the previous section, this possibility of amelioration by co-indexation with

24

For space reasons, I only go over examples with adjunct islands, but the same argument goes through with complex NP islands. See Hara (2006, ch. 3; forthcoming) for examples with complex NP islands. 25 Some informants report that (87) is not completely acceptable. The sentence (87) improves if it is read with post-focal reduction after the contrastive-marked element John-wa (cf. Ishihara 2002, 2003) and with a pause after John-wa.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7.4.2 Co-indexation with pro In addition to ‘pied-piping’-like constructions, (2a) can be ameliorated by generating a wa-marked NP overtly outside the island construction and co-indexing it with pro. For example, in (87), the wa-marked NP JOHNi-wa is co-indexed with pro, within a temporal clause, and it induces the intended implicature.25

264 Evidentiality of Discourse Items and Because-Clauses pro demonstrates that the ungrammaticality of (2) is not due to semantic constraints but due to syntactic ones.26 In short, the unavailability of the global computation of CON in (2) is not due to semantic constraints but due to syntactic ones, since the intended interpretations are successfully derived by changing the syntactic structures.

7.5 Section summary

8 CONCLUDING REMARKS This paper has dealt with several different semantics–pragmatics concepts: the implicature triggered by the contrastive marker wa, evidentials, and reasoning expressed by because, which all share a common property: these discourse items operate over a proposition, and hence, they block binding of an event variable. I have also proposed that the computation of CON involves syntactic movement which determines the size of the proposition it takes and the context which binds the indexicals. This movement is blocked if wa is embedded within a relative clause or an adjunct clause which are islands for movement. The facts presented in this paper also have interesting ramifications regarding the connection between implicatures and evidentiality. Both 26 The sentence in (87) is not an instance of overt movement of John-wa. Hoji (1985) provides the following anaphor-binding test to show that the sentence-initial wa-marked phrase is not an instance of movement. In (ia), if the sentence-initial wa-marked phrase ‘sono zibun nituite-no hon’ were originally generated under VP and preposed by movement, zibun could be bound by John, since it could be reconstructed into the argument position as in (ib). This interpretation is, however, not possible; hence, the sentence-initial wa-marked phrase is not an instance of overt movement. It is base generated in the initial position and co-indexed with an empty pronoun (pro in more recent terminology) at the argument position.

(i) a. *[NP b.

sono zibuni nituite-no hon ]j -wa Johni-ga [VP ej suteta ] that self about book -Top John-nom threw-away ‘As for [that book about himselfi] j, Johni threw itj away.’ [S[NP sono zibuni nituite-no hon]j -o [S Johni ga [VP tj suteta ]]] that self about book -Acc John-nom threw-away ‘That book about himselfi, Johni threw away.’ (Hoji 1985: 129, 133)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The use of wa triggers implicatures which are associated with the speaker or some attitude-bearer. This association is blocked by certain syntactic configurations, namely adjunct islands and complex NP islands. To capture these facts, I have proposed a syntactic movement account for the positioning of the CON operator. CON moves in order to locally identify the context that saturates its shiftable indexicals.

Yurie Hara 265

concepts were previously treated within semantics–pragmatics, while recent studies have started to explore the phenomena in the context of syntax–semantics–pragmatics interfaces. For example, Chierchia (2004) argues that syntactic structures must be visible for implicature computation. Recent work by Tenny (2006) and Speas (2004) (following Cinque 1999) also argues that there is a syntactic representation for point-ofview arguments. Consider again the following example with a hearsay evidential.

Following Tenny’s (2006) formulation, (89) has the structure depicted in (90). The evidential morpheme sooda projects an evidential phrase which contains xj, someone other than the speaker or the hearer, as an invisible argument. (90) [SpeechActP[the speakeri][EvidentialP xj[Evidential [Evidential sooda]]]]

John-ga kaet-ta

Since Japanese evidentials have a very rigid syntax, it is plausible to posit a syntactic projection for this lexical category. Tenny (2006) further treats Japanese node ‘because’ as an evidential and argues that because-clauses are Evidential Phrases. The syntactic movement proposed for wa-implicature computation accords well with Tenny’s proposal. If because-clauses head or embed an Evidential Phrase a` la Tenny, they can host a CON operator, while temporal clauses or ifclauses, which cannot host an Evidential Phrase, cannot host a CON operator either. In other words, the presence or absence of an Evidential Phrase can be regarded as a syntactic reflection of Johnston’s semantic difference between temporal clauses and because-clauses.

Acknowledgements I am grateful to Benjamin Bruening, Chris Davis, Bart Geurts, Jim Huang, Manfred Krifka, Eric McCready, Anna Papafragou, Chris Potts, Maribel Romero, Shoichi Takahashi, Satoshi Tomioka, Youri Zabbal and the audience at the Language Under Uncertainty workshop at Kyoto University for discussion and comments. Special thanks go to two anonymous reviewers and the editors for their helpful extensive comments. All errors are my own.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(89) John-ga kaet-ta sooda. John-Nom go.home-Past Evid ‘John went home (I heard).’

266 Evidentiality of Discourse Items and Because-Clauses YURIE HARA Department of Linguistics Graduate School of Letters Kyoto University Yoshida-honmachi Sakyo-ku Kyoto 606-8501 Japan e-mail: [email protected]

REFERENCES guage and Meaning. Chicago, Chicago University Press. 195–212. Hagstrom, P. (1998), Decomposing Questions. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA. Hara, Y. (2005), Contrastives and Gricean principles. In P. Dekker and M. Franke (eds.), Fifteenth Amsterdam Colloquium. Amsterdam, Universiteit van Amsterdam. 101–6. Hara, Y. (2006), Japanese Discourse Items at Interfaces. Ph.D. thesis, University of Delaware, Newark, DE. Hara, Y. (2007), Movement of a Shifty Operator. In Proceedings of Formal Approaches to Japanese Linguistics. 4: 61–72. MIT Working Papers in Linguistics 55. Cambridge: MITWPL, forthcoming. Heycock, C. (1993), ‘Focus projection in Japanese’. In M. Gonzalez (ed.), Proceedings of NELS, 24. 159–87. Amherst, MA.GLSA. Hoji, H. (1985), Logical Form Constraints and Conﬁgurational Structure in Japanese. Ph.D. thesis, University of Washington. Horn, L. R. (1972), ‘Greek Grice’. In Proceedings of the Ninth Regional Meeting of the Chicago Linguistics Society. Chicago Linguistics Society. Chicago University, Chicago, IL. 205–14. Huang, C.-T. J. (1982a), Logical Relations in Chinese and the Theory of Grammar. Ph.D. thesis, Massachusetts

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Aoki, H. (1986), ‘Evidentials in Japanese’. In W. Chafe and J. Nichols (eds.), Evidentiality: The Linguistic Coding of Epistemology. Ablex Publishing Corporation. Norwood, New Jersey. 223–236. Bu¨ring, D. (1997), ‘The great scope inversion conspiracy’. Linguistics and Philosophy 20:175–94. Chierchia, G. (2004), ‘Scalar implicatures, polarity phenomena, and syntax/pragmatics interface’. In A. Belletti (ed.) Structures and Beyond. Oxford University Press. Oxford. 39–103. Choe, J. W. (1987), ‘LF movement and pied-piping’. Linguistic Inquiry 18:348–53. Cinque, G. (1999), Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford, Oxford University Press. Davidson, D. (1967), ‘Causal relations’. Journal of Philosophy 64:691–703. Deguchi, M., & Y. Kitagawa (2002), ‘Prosody and Wh-questions’. In Masako Hirotani (ed.) Proceedings of the Thirty-second Annual Meeting of the North-Eastern Linguistic Society. 73–92. Amherst, MA: GLSA. Faller, M. (2002), Semantics and Pragmatics of Evidentials in Cuzco Quechua. Ph.D. thesis, Stanford University, Palo Alto, CA. Gamut, L. T. F. (1991), ‘Pragmatics: meaning and usage’. In Logic, Lan-

Yurie Hara 267 Kratzer, A. (1991b), ‘The representation of focus’. In A. von Stechow and D. Wunderlich (eds.), Semantics: An International Handbook of Contemporary Research. Walter de Gruyter. Berlin. 825–34. Kratzer, A. (1998), ‘Scope or pseudoscope? Are there wide-scope indefinites?’ In S. Rothstein (ed.), Events and Grammar. Dordrecht, Kluwer Academic Publishers. 163–96. Kratzer, A. (1999), ‘Beyond ouch and oops. How descriptive and expressive meaning interact’. Handout, Cornell Conference on Theories of Context Dependency, Ithaca. March 26. Krifka, M. (2001), ‘For a structured meaning account of questions and answers’. In C. Fery and W. Sternefeld (eds.) Audiatur Vox Sapientia. A Festschrift for Arnim von Stechow. Akademie Verlag. Berlin. 287–319. Kuno, S. (1973), The Structure of the Japanese Language. MIT Press. Cambridge, MA. Kuroda, S.-Y. (1973), ‘Where epistemology, style, and grammar meet: a case study from Japanese’. In S. Anderson and P. Kiparsky (eds.), A Festschrift for Morris Halle. Holt, Rinehart and Winston. New York. 377–91. Kuroda, S.-Y. (2005), ‘Focusing on the matter of topic: a study of Wa and Ga in Japanese’. Journal of East Asian Linguistics 14:1–58. Landman, F. (2000), Events and Plurality: The Jerusalem Lectures. Kluwer. Dordrecht, Holland. McCready, E. (2006), ‘Shifting contexts? That might be good’. Slides presented at The 11th Sinn und Bedeutung Conference, September 21–23, Barcelona. Nishigauchi, T. (1990), Quantiﬁcation in the Theory of Grammar. Kluwer. Dordrecht.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Institute of Technology, Cambridge, MA. Huang, C.-T. J. (1982b), ‘Move WH in a language without WH movement’. The Linguistic Review 1:369–416. Ishihara, S. (2002), ‘Invisible but audible Wh-scope marking: Wh-constructions and deaccenting in Japanese’. In Line Mikkelsen and Chris Potts (eds.) Proceedings of the Twenty-ﬁrst West Coast Conference on Formal Linguistics. University of California, Santa Cruz, Somerville, MA. Cascadilla Press. 180– 93. Ishihara, S. (2003), Intonation and Interface Condition. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA. Izvorski, R. (1997), The present perfect as an epistemic modal. In A. Lawson and E. Cho (eds.) The Proceedings of SALT, volume 7. 222–39. Cornel: CLC Publications. Johnston, M. (1994), The Syntax and Semantics of Adverbial Adjuncts. Ph.D. thesis, University of California, Santa Cruz, Santa Cruz, CA. Kaplan, D. (1989), ‘Demonstratives: an essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals’. In J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford University Press. Oxford. 481–563. Kikuchi, A. (1987), Comparative Deletion in Japanese. Unpublished MS, Yamagata University, Yamagata, Japan. Kratzer, A. (1987), ‘Modality’. In A. von Stechow and D. Wunderlich (eds.), Handbook of Semantics. Walter de Gruyter. Berlin. 639–50. Kratzer, A. (1991a), ‘Conditionals’. In A. von Stechow and D. Wunderlich (eds.), Semantics: An International Handbook of Contemporary Research. Walter de Gruyter. Berlin. 651–56.

268 Evidentiality of Discourse Items and Because-Clauses Pesetsky, D. (1987), ‘Wh-in-situ: movement and unselective binding’. In E. J. Reuland and A. G. ter Meulen (eds.), The Representation of (in)deﬁniteness. MIT Press. Cambridge, MA. 98–129. Potts, C. (2003), The Logic of Conventional Implicatures. Ph.D. thesis, University of California, Santa Cruz, Santa Cruz, CA.

Rooth, M. (1985), Association with Focus. Ph.D. thesis, University of Massachusetts at Amherst. Rooth, M. (1992), ‘A theory of focus interpretation’. Natural Language Semantics 1:75–116. Ross, J. R. (1967), Constraints on Variables in Syntax. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA. Sawada, M., & R. Larson (2004), ‘Presupposition & root transforms in adjunct clauses’. In M. Wolf and K. Moulton (eds.), Proceedings of NELS 34. Amherst, MA, GLSA. 517–28. Scheffler, T. (2005), ‘Syntax and semantics of causal Denn in German’. In P. Dekker and M. Franke (eds.), Fifteenth Amsterdam Colloquium. Unversiteit van Amsterdam, Amsterdam, The Netherlands. 215–220. Schlenker, P. (2003), ‘A plea for monsters’. Linguistics and Philosophy 26:29–120. Shimoyama, J. (2006), ‘Indeterminate phrase quantification in Japanese’. Natural Language Semantics 14:139–73.

First version received: 08.07.2006 Second version received: 15.05.2007 Accepted: 05.08.2007

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Reinhart, T. (1997), ‘Quantifier scope: how labor is divided between QR and choice functions’. Linguistics and Philosophy 20:335–97.

Speas, M. (2004), ‘Evidentiality, logophoricity and the syntactic representation of pragmatic features’. Lingua 114:255–76. Spector, B. (2003), ‘Scalar implicatures: exhaustivity and Gricean reasoning’. In B. ten Cate (ed.), Proceedings of the ESSLLI’03 Student Session. Vienna. 277–288. Tenny, C. (2006), ‘Evidentiality, experiencers, and the syntax of sentience in Japanese’. Journal of East Asian Linguistics 15:245–88. Tredinnick, V. (2004), Modal Flavor and Quantiﬁcational Force in Free Relatives with -ever. Ph.D. thesis, University of Pennsylvania, Philadelphia, PA. Vallduvı´, E. (1992), The Informational Component. Garland. New York. von Fintel, K., & S. Iatridou (2002), ‘The meanings of epistemic modality’. Presentation at Sinn und Bedeutung 7, Universita¨t Konstanz, October 5. van Rooij, R., & K. Schulz (2004), ‘Exhaustive interpretation of complex sentences’. Journal of Logic, Language and Information 13:491–519. von Stechow, A. (1991), ‘Focusing and backgrounding operators’. In W. Abraham (ed.), Discourse Particles. Benjamins. Amsterdam. 37–84. Watanabe, A. (1992), ‘Subjacency and Sstructure movement of wh-in-situ’. Journal of East Asian Linguistics 1: 255–91.

Journal of Semantics 25: 269–319 doi:10.1093/jos/ffn003

Conditional Questions JAMES ISAACS University of California Santa Cruz KYLE RAWLINS University of California Santa Cruz

This paper provides an analysis of conditional questions (CQs) that combines a dynamic semantics for conditionals with a partition semantics for questions. We propose that CQs are interpreted in two steps. First, a temporary context is created in which the propositional content of the antecedent is obtained. Second, the question in the consequent is asked relative to this temporary context. Subsequent answers are then asserted relative to the temporary context. Our analysis also has a pragmatic component. Previous analyses have augmented the semantics to account for denials of the antecedents of CQs. We show that the effect of denying the antecedent of a CQ is not due to the semantics of the question. Instead, denials of the antecedent deny the presuppositions of the conditional and do not directly address the question at all.

1 INTRODUCTION This paper provides an analysis of conditional questions (CQs) such as (1). CQs are conditional sentences with interrogative consequents. (1)

If Alfonso comes to the party, will Joanna leave?

The puzzling fact about (1) is that, as a question act, it is much more limited in scope than the plain polar question ‘Will Joanna leave?’ While both ask for similar information about Joanna, the CQ is only concerned with a limited set of circumstances: the ones in which Alfonso comes to the party. To capture this, the semantics of the conditional antecedents must interact with the semantics, and especially the dynamics, of questions. This paper addresses that interaction. We argue that CQs are best analysed by combining a dynamic semantics for conditionals (along the lines of Stalnaker 1968; Karttunen 1973, 1974; Heim 1983, 1992, and much other work) with the partition semantics for questions proposed in Groenendijk (1999). We propose that a conditional adjunct restricts the domain of the question The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected].

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Abstract

270 Conditional Questions

(2) Alfonso isn’t coming. Intuitively, following such a response (a denial of the antecedent), no further answer to the CQ is required. This fact has led previous accounts to propose that denials of the antecedent are in fact answers of some sort or at least that they play a role in the alternative structure that the question introduces. We argue that these accounts semanticize a pragmatic effect—that denials of the antecedent are denials of a presupposition of the conditional (that the antecedent is possible). We conclude that responses of this kind do not play a role in the question’s alternative structure, and that their interpretation involves quite different principles—belief revision and conversational backtracking. A major issue in the background throughout the paper is what Groenendijk and Stokhof (1997) call ‘Hamblin’s Picture’. This is a view of questions that takes the alternative structure they introduce to be a partition, i.e. it consists of mutually exclusive alternatives that exhaust the logical space. An attractive alternative analysis of the denial of the antecedent data, as well as the semantics of CQs, involves abandoning the mutual exclusivity component of Hamblin’s Picture (Velissaratou 2000). We argue that this is not the right way to go and show that mutual exclusivity can be retained as a property of questions and of contexts. However, we also show that partially adopting an approach along the lines of Velissaratou (2000) leads to a simpler semantics for embedded CQs. The remainder of the paper is structured as follows. In section 2, we provide an overview of the background assumptions that guide our

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

operator, and we provide a technical mechanism for describing this domain restriction in a dynamic semantics. In our approach, the interpretation of the antecedent and the consequent of a CQ is decomposed into two steps. The first step involves making a temporary copy of the current context and updating the copy with the propositional content of the antecedent. The question in the consequent then raises an issue relative only to the temporary context. Temporary contexts for conditionals are treated on par with the temporary contexts involved in some analyses of modal subordination (Frank 1996; Kaufmann 2000). What this means is that they may persist past the sentence that introduces them. Our formalization of this persistence uses the stack-based account of modal subordination proposed in Kaufmann (2000) to model temporary contexts. The analysis we propose also has a pragmatic component to deal with responses such as (2).

James Isaacs and Kyle Rawlins 271

2 BACKGROUND We assume a dynamic possible worlds semantics, with a Stalnaker (1978)-style context. For our purposes, the important member of the context is the context set. The context set is the set of possible worlds that describe what the actual world could be like as far as the discourse participants are concerned. For example, if the discourse participants all publicly accept that it is raining at the time of utterance, the context set will only include worlds in which it is raining at this time. While many other elements are necessary to present a full representation of an utterance context (such as the time and place of utterance, information about the speaker, hearers), we disregard all but the context set here. The meaning of an utterance, in this kind of system, is given as a context change potential (CCP)—a function from contexts to contexts. Assertions are proposals to change the context by reducing the context set. Prior to the assertion of It is raining, the context set contains worlds in which it is raining and ones in which it is not. Once the discourse participants accept that the assertion is true, the worlds in which it is not raining are removed from the context set. In (3), we give a typical dynamic semantics definition of the CCP of an assertion (focusing exclusively on context sets).1 The static interpretation function [[ ]]w,c is given a standard treatment. d

1

We do not, for the most part, provide a compositional dynamic semantics below the sentence level though this would be possible; see Muskens (1996) and Bittner (2001) for dynamic semantics accounts that do.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

analysis. In section 3, we consider previous analyses of CQs (Hulstijn 1997; Velissaratou 2000) which call some of these assumptions into question and discuss the problems raised by and confronting alternative approaches. In this section, we also review a broad range of data that an analysis must cover. In section 4, we give a presuppositional account of denials of the antecedent. In section 5, we provide our analysis of matrix CQs, which is then exemplified in section 6. Some issues that remain unsettled up to this point, involving the status of issues and the interpretation of counterfactual CQs, are addressed in section 7. We extend our analysis to embedded CQs in section 8. Finally, in section 9, we discuss some larger issues raised by our analysis, and in particular how it relates to the analysis of conditionals proposed by Kratzer (e.g. in Kratzer 1986).

272 Conditional Questions (3) Assertive update (4) on contexts (preliminary) For any context c and clause2 /: c4/ ¼ def fw 2 c j ½½/w;c ¼ 1g Asserting / (as long as the assertion is accepted) amounts to removing worlds from the input context where the proposition denoted by / is not true, or alternatively, leaving behind only worlds where the proposition is true.

We adopt a partition semantics for questions, following most closely Groenendijk (1999) (see also Higginbotham and May 1981; Groenendijk and Stokhof 1984; Higginbotham 1993; Groenendijk and Stokhof 1997, among others). Just as in the line of work stemming from Hamblin (1958, 1973) (e.g. Karttunen 1977; Kratzer and Shimoyama 2002), to know the meaning of a question is to know the possible answers to the question. In a partition semantics, possible answers correspond to cells (or for a more general term, alternatives) in a partition that the question induces on the set of possible worlds. The word ‘partition’ is used in the technical sense—the alternatives represented by a partition are mutually exclusive, and exhaust the range of possibilities. For the sake of simplicity, we restrict our examples to polar questions. Following Groenendijk (1999), we incorporate the possibility of partitioning into a dynamic semantics by treating the context as an equivalence relation on worlds. An equivalence relation on some set is a symmetric, transitive and reflexive relation and is isomorphic to a partition of that set. The equivalence relation is represented by taking the context set to be a set of pairs of worlds. In standard dynamic semantics, a world is a live candidate for the actual world when it is present in the context set. In a Groenendijk-style dynamic semantics, a world is a live candidate for the actual world if it is present as either member of a pair in the context set. Assertive update has the affect of eliminating pairs of worlds in a context set in which one world in the pair makes the content of the assertion false. 2

We use the word ‘clause’ to refer pretheoretically to the syntactic object that is roughly a sentence without force, or a sentence without a speech–act operator. It does not matter for our purposes whether this is a CP, an IP, or something else.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

2.1 Questions and assertions in a dynamic partition semantics

James Isaacs and Kyle Rawlins 273

(4)

Assertive update (4) on contexts For any context (set) c and clause /: c4/ ¼def fÆw1 ; w2 æ 2 c j ½½/w1 ;c ¼ ½½/w2 ;c ¼ 1g

The kind of update associated with questions does not ever remove worlds altogether, but rather disconnects parts of the context. It leaves worlds completely connected only if they correspond to the same complete answer—if they resolve the issue raised by the question in the same way. That is, a questioning update partitions the context set. (5)

c / ¼ def fÆw1 ; w2 æ 2 c j ½½/w1 ;c ¼ ½½/w2 ;c g The remainder of this section contains an example illustrating the basic dynamic semantics for questions. The example models a discussion about whether it is raining. The example will use a simple context set consisting of five worlds (w1, . . ., w5), shown in (6) (6)

Let us assume that in w1, w2 and w3 it is in fact raining, and in the other two worlds, it is not. We can then see what would happen if a speaker asked the question in (7). (7) Is it raining? This question is a polar question and, as such, partitions the context into two cells. The worlds in which it is raining constitute a cell, as do the ones in which it is not. Contexts consisting of more than one cell are called inquisitive contexts (Groenendijk 1999). The world pairs in bold are ones which resolve the question affirmatively.

(8)

In this context, worlds w4 and w5 remain completely connected to each other, as do the worlds w1, w2 and w3. There is no pair connecting

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Inquisitive update ( ) on contexts For any context c and clause /:

274 Conditional Questions worlds such as w1 with w4, which resolve the question of whether it is raining in different ways. Answers like (9) take the form of assertions. (9) Yes, it’s raining. Answers reduce the context set to a single, completely connected cell. When a context set contains only one cell, the context is uninquisitive (Groenendijk 1999). The effect of updating c# above with the assertion in (9) is given in (10). (10)

3 DENIAL OF THE ANTECEDENT CQs allow for a kind of response, which we refer to as a ‘denial of the antecedent’, which is not available to other kinds of questions. An example discourse with a denial of the antecedent response is given in (11). (11) A: If Alfonso comes to the party, will Joanna leave? B: Alfonso isn’t coming to the party. Intuitively, there is some reason to think that denial of the antecedent responses are answers to CQs. Responses like (11B) are pertinent to the discourse and they seem to dispel the issue raised by the question—after the response in (11B), there is no longer any obligation to respond to the CQ in (11A). We will refer to this as the issue-dispelling effect. This effect has previously only been associated with complete answers to questions and requires some explanation with respect to CQs. There is a parallel problem, when starting from a partition theory of questions: the problem of what to do with worlds where the antecedent of the conditional is false. These worlds are obviously in the domain that the whole sentence is interpreted relative to, and so we have to reconcile them with the partition that the question picks out, which does not obviously include these worlds. The fact that denial of the antecedent responses exist suggests that these worlds could form some part of the partition (or of the alternative structure, if we drop some

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In (10), the worlds where it was not raining, w4 and w5, have been removed. This leaves behind the cell of the partition in c# which corresponds to an affirmative answer.

James Isaacs and Kyle Rawlins 275

3.1 The tripartition analysis A tripartition approach to indicative (polar) CQs was first suggested by Groenendijk and Stokhof (1997, fn. 29). Hulstijn (1997) gives a logical and dynamic account of many properties of questions, and in the process provides a logical implementation of the tripartition approach to CQs (though this is by no means the focus of the paper), which deals with denial of the antecedent responses by having polar CQs induce a tripartition on the context set. The tripartition comprises the usual two cells in which the antecedent is true: one in which the consequent of the CQ is also true (corresponding to ‘yes’) and the other in which the consequent is false (‘no’). There is also a third cell that contains world pairs in which the antecedent is false. The tripartition corresponding to a CQ //?w is shown in Figure 1. If other kinds of questions are taken into account beyond polar questions, this might be called an n + 1 partition approach, as it adds an extra cell to whatever sized partition you would get from the

Figure 1 :/ worlds get their cell.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

assumptions of partition semantics). If they do not fit into the alternative structure, an independent explanation of the issue-dispelling property of denials of the antecedent must be given. Previous analyses of CQs (Hulstijn 1997; Velissaratou 2000) take the first option and fit worlds where the antecedent is false into the alternative structure in different ways. Hulstijn (1997) makes denials of the antecedent complete answers, giving a semantics for CQs that induces a tripartition on the context, instead of a bipartition. Velissaratou (2000) argues against this approach, and instead relaxes the mutual exclusivity constraint on partitions, proposing that CQs pick out exhaustive but not necessarily mutually exclusive alternatives. Attention is restricted in both these analyses to indicative CQs. We discuss each of these possibilities in turn.

276 Conditional Questions non-conditionalized form of the question. This extra cell would always contain all worlds where the antecedent is false. For the sake of clarity, we focus on polar questions. This approach makes the denial of the antecedent problem a purely semantic one—denials of the antecedent are complete answers to a CQ.

3.1.2 Additional arguments The first argument is the other side of the coin of Velissaratou’s argument regarding what denial of the antecedent responses are about. Given a CQ of the form p/?q, a tripartition analysis predicts not only that :p will be an answer but also that p will be a partial answer (assuming a suitable definition of partial answerhood, such as Groenendijk’s (1999) licensing). This is because a p

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

3.1.1 Arguments against a tripartition analysis—Velissaratou (2000) Velissaratou (2000) presents three compelling arguments against a tripartition analysis of CQs. The one we judge most important is an argument from intuitions. The key intuition is that denial of the antecedent responses are in some sense not ‘about’ the issue raised by the question (though they may be relevant to some larger topics under discussion). As Velissaratou puts it, ‘if one allows ½½: p to be an answer to ½½p/? q, we have the strange aspect that a proposition which cancels the reason the question was posed in the first place, is an answer.’ We have found this intuition to be reliably reproducible by native speakers presented with denial of the antecedent data, and use it in section 4 to guide the proposal we develop as a response to Velissaratou (2000). The second argument Velissaratou brings forward is that the tripartition analysis incorrectly predicts that full conditionals will not be complete answers to CQs. A conjunction of the antecedent and the consequent stand in this role in Hulstijn (1997), but the conjunctive response feels infelicitous or overinformative. The (felicitous) conditional response is shown in (12B), and the odd conjunctive response is shown in (12B#). (12) A: If Alfonso comes to the party, will Joanna leave? B: If he comes, Joanna will leave. B#: #He will come and Joanna will/won’t leave. Finally, Velissaratou presents a series of facts that the tripartition analysis cannot derive (see Velissaratou 2000, fact 2). We do not discuss these in detail here; the most linguistically interesting point is that a theory of CQs should predict that p/?p and :p/?p are trivial questions, and a tripartition theory does not. Before discussing Velissaratou’s (2000) analysis, we give two further arguments against the tripartition analysis.

James Isaacs and Kyle Rawlins 277

response affirming the antecedent of the CQ removes the :p cell, leaving behind the other two. This conflicts with intuitions regarding what responses of this kind are about. When speakers are presented with discourses such as the following, they judge the p responses to be odd and irrelevant in some way. (13) A: If Alfonso comes to the party, will Joanna leave? B: #Alfonso is coming to the party.

(14) A: Who is teaching Syntax A this quarter? B: Not Jim. The B response is a perfectly felicitous partial answer. There is a second problem that has no obvious explanation under a tripartition analysis, though this is a problem that one could potentially respond to while maintaining a tripartition analysis. ‘Yes’ and ‘no’ are in some sense privileged answers that are only available to polar questions. All polar questions that have previously received significant discussion in the literature are bipartitions. In those cases, it is not surprising that there are two privileged answers, and it is at least intuitively obvious why they correspond to the cells of the partition they correspond to. But these privileged answers are still usable with CQs formed out of an interrogative that has the syntactic characteristics of a polar question. The obvious concern for a tripartition account is why they are usable at all with a tripartition. Beyond that, it is unclear why they map to the cells that they map to. For instance, why does a ‘no’ answer fail to deny the antecedent? This is not an argument against a tripartition account per se, as to the best of our knowledge, no one has tried to answer this question, but it is another issue that any tripartition account would need to deal with. Additionally, several of the arguments that we develop later against Velissaratou’s (2000) account apply equally well against a tripartition account.

3.2 Abandoning mutual exclusivity A defining characteristic of answers to questions, due originally to Hamblin (1958), is that every question has a unique, complete answer.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The judgment in (13) is different from hearers’ responses to denials of the antecedent, which are felicitous but not about the question in particular. Affirmations of the antecedent (for indicative conditionals) are simply infelicitous in some way. This is quite surprising, as partial answers are generally perceived as intuitively relevant. An example that should be parallel to the B response, but is not, is given in (14).

278 Conditional Questions

Figure 2 :/ worlds inside both alternatives.

3 It would also be possible to challenge exhaustivity in a similar way. We do not try to develop an analysis directly along these lines, and to our knowledge it has not been done, but our proposal could in fact be seen as going in this direction. We assume (see section 5) a form of local exhaustivity only—questions partition exhaustively relative to the local context set, which reflects the assumptions that have been made. 4 Velissaratou’s (2000) primary motivation for developing a logic for CQs is to provide an analysis of ‘which’-questions. We do not discuss ‘which’-questions here. However, applying the kind of domain restriction we develop here to the problems of ‘which’-questions, in parallel to Velissaratou’s (2000) analysis, provides a tempting prospect for future work.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It follows from this notion that answers to questions are mutually exclusive—the truth of an answer implies that all other possible answers to the question are false. It also follows that answers are exhaustive. While both these properties can be challenged (see Groenendijk and Stokhof (1997) for discussion), they are fundamental to defining partitions in a partition semantics for questions. One way of accepting denial of the antecedent responses as (partial) answers to CQs is to maintain that answers need not be mutually exclusive.3 This is the position taken in Velissaratou (2000), which provides a (non-dynamic) logic for CQs.4 In this theory, questions pick out alternatives that exhaust the domain, but may not be mutually exclusive. They form maximal covers of the domain, in the sense that the union of the alternatives will include the domain. Since they do not have to be mutually exclusive, these alternatives can overlap, and in the case of CQs, they do. Given a polar CQ p/?q, worlds where the antecedent is false constitute the overlap, as shown in Figure 2. There are two alternatives, corresponding to p/q and p/:q. They overlap, exactly on the :p worlds. Because they overlap on these worlds, denying the antecedent of a CQ does not choose between the alternatives, but rather simply removes the worlds where the covers fail to overlap. In dynamic terms, denying the antecedent would result in a context without distinct alternatives, i.e. a non-inquisitive context. Denial of

James Isaacs and Kyle Rawlins 279

5 The obvious contender is Groenendijk’s (1999) notion of licensing (see definitions 8 and 10 of that paper). Jeroen Groenendijk (personal communication) has pointed out to us that denials of the antecedent would not be licensed, and therefore not answers, as they fail to remove any entire alternative. However, as he also points out, the notion of licensing as-is will not work with overlapping alternatives. This is because removing part of an alternative is never licensed, but if there is overlap, any response which removes all of one alternative (i.e. a complete answer, in the case of a polar CQ) would remove part of the other, and therefore not be licensed. A different notion of licensing might be developed to get around this issue, as the one in Groenendijk (1999) was tailored to a partition semantics, but that is beyond the scope of this paper.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the antecedent can be construed as issue dispelling for this reason, but does not resolve the issue raised by the question. That is, it does not choose between alternatives introduced by the question. Ordinary (i.e. non-conditionalized) polar questions involve no overlap, and the set of maximal covers would be a partition. Therefore, all issue-dispelling responses to a plain polar question also definitively choose one alternative or the other. In fact, all non-conditionalized questions would involve normal partitions. This analysis works well for the core CQ data, and provides an interesting treatment of denials of the antecedent—dissociating issuedispelling responses from responses which actually resolve the issue. It overcomes Velissaratou’s arguments against the tripartition approach (though see below for further discussion of the ‘aboutness’ issue), as well as the two additional arguments we raise. Denials of the antecedent are possible but not complete answers (when answerhood is defined in terms of entailment), and this is because their negation (affirmation of the antecedent) is not an answer to a CQ at all. This accordingly solves the problem we raise, where the affirmation of the antecedent of an indicative conditional seems uninformative and infelicitous—it does not advance the discourse. There is no problem about which meanings to assign ‘yes’ and ‘no’ to, because there are only two alternatives, and whatever mechanism associates the two privileged answers with the correct cells in partition theory (not a trivial matter), could presumably work here. However, there are several problems, both conceptual and empirical, that motivate revision of this analysis. We have pointed out that there is some sense in which denials of the antecedent do not seem to be ‘about’ the question itself, but rather about the ground upon which the question stands, and Velissaratou (2000) expressed a similar intuition. But it is unclear that the non-mutual exclusivity analysis actually captures this intuition, as a denial of the antecedent is still a possible answer—just not a complete one. We will shortly propose an analysis which directly captures this intuition, though this might also be repaired by using a notion of answerhood that is not based on entailment.5 A second conceptual problem is that to accept the analysis

280 Conditional Questions

3.2.1 Counterfactual CQs Velissaratou (2000) treats the behaviour of denial of the antecedent responses as a semantic property of conditionals.7 As such, we would expect it to generalize to all sorts of conditionals. In this section, we consider counterfactual conditionals. It is immediately clear that the expected class of responses are not, in fact, issue dispelling. This is shown by the response in (15B).8 (15) A: If Jo could have fixed the car, would you have kept on using it? B: Jo couldn’t have fixed the car. Intuitively, this response does not dispel the issue, and even seems somewhat infelicitous following the counterfactual CQ, despite being the negation of the antecedent. The non-mutual exclusivity account has something to explain here. Not only, but also ideally, that the explanation will not have anything to do with the questioning nature of the consequent. This seems to be a general fact about counterfactuals—von Fintel (2001) observes that similar responses do not have any effect on counterfactuals with declarative consequents (example (27) there). There is a slightly different response, which does seem to be issue dispelling. This is illustrated in (16). (16) A: If Jo could have(/had) fixed the car, would you have kept on using it? B: Jo could have fixed(/did fix) the car. 6 Except ‘which’-questions, which under Velissaratou (2000) have a logical structure based on CQs. 7 This problem applies as well to Hulstijn (1997), which also semanticizes the problem. 8 Thanks to Geoff Pullum for this example.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

one must abandon the position that mutual exclusivity is a core property of questions. This is certainly a move that must be considered and made if the data warrant it, but we shortly provide arguments that the denial of the antecedent data can be explained in another way, without any need to give up mutual exclusivity. It is also important to note that mutual exclusivity is abandoned in a way tailored to CQs—the analysis reduces to a partition analysis for other nonconditionalized questions.6 The question then arises why a property that seemed otherwise so general has a small pocket of the grammar where it fails. The following sections provide several empirical arguments against the non-mutual exclusivity analysis. We also return to the argument in later sections of the paper where we discuss the prospects for combining aspects of our proposal with that of Velissaratou (2000).

James Isaacs and Kyle Rawlins 281

3.2.2 Modality and responses to CQs Denial of the antecedent responses are incompatible with ‘would’-modals, whereas true answers are consistently compatible with them. This is illustrated by the oddness of the B$ response below. (17) A: If Alfonso comes to the party, would Joanna be mad? B: Yes, she will/would. B#: Alfonso is not coming/won’t come. B$: #Alfonso would not come.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It seems that for a counterfactual, the affirmation of the antecedent is issue dispelling, but not resolving (though this discourse leads naturally to asking the plain factual form of the consequent). This is completely unexpected on the account in Velissaratou (2000). The denial of the antecedent response is infelicitous. At the least, something must be done differently for counterfactuals under this analysis. The situation could, at first glance, be remedied by defining a different counterfactual conditional operator that places a different set of worlds in the overlap between alternatives. The alternatives in the example above would have to overlap on worlds where Jo could have fixed (or did fix) the car. However, there is a paradox—these are the worlds that would need to be split into two by the question in the antecedent. The alternatives cannot simultaneously overlap on these worlds and break them into the ‘yes’ and ‘no’ alternatives. This paradox points to a deeper problem. In the case of counterfactual CQs, the normal answers are interpreted relative to counterfactual worlds. In von Fintel’s (2001) terminology, a counterfactual CQ involves partitioning some worlds on the ‘modal horizon’, and the answers should be interpreted there. But a denial of the antecedent is fundamentally a non-counterfactual statement and is about worlds contained in the regular context set—it is not about worlds on the modal horizon. In the case of a counterfactual CQ, a denial of the antecedent seems to be a fundamentally different kind of response than a regular answer. In section 4, we propose an entirely different account of these data, which does not semanticize the denial of the antecedent behaviour. We propose that these responses deny the presuppositions of conditional sentences, which in the case of counterfactuals are presuppositions that the antecedent is actually false. The denial indicates presupposition failure on the part of the denier at a point prior even to the acceptance of the question. This captures in an entirely different way Velissaratou’s (2000) observation that the denial of the antecedent responses ‘cancel the reason the question was posed in the first place’.

282 Conditional Questions In a tripartition theory, we would not expect any difference in the choice of modals between the three possible answers. They are also surprising for Velissaratou (2000), since we would not expect it to be possible to use a particular modal in a complete answer but not in a partial answer. We need an explanation for why ‘would’-modals are disallowed in denial of the antecedent responses. This point can be taken further—the use of ‘would’-modals patterns with sentences uttered in modal subordination contexts (Roberts 1989; Frank 1996; Kaufmann 2000).

In fact, the possibility of using a ‘would’-modal in the answers patterns with responses to questions asked in a modal subordination context. (20) A: A thief might break in. Do you think he would steal the silver? B: Yes, he would. (21) A: I hope Alfonso will come. Do you think he would be entertaining? B: Yes, he would. In summary, a theory of CQs should ideally explain why true answers pattern with answers to modally subordinated questions, and denial of the antecedent responses do not. 3.2.3 ‘Yes’, ‘no’ and fragment answers Velissaratou (2000) makes certain assumptions about the meaning of ‘yes’ and ‘no’. In response to a CQ with the logical structure p/?q, ‘yes’ means p/q and ‘no’ means p/:q. From a linguistic perspective, this is intuitively odd. The B response in (22) demonstrates the oddness. (22) A: If Alfonso comes to the party, will Joanna be mad? B: Yes, she will. B#: Yes, if he comes, she will be mad. If what ‘yes’ means is the same thing as response B#, then it is not clear why the continuation in B, without the ‘if ’-clause, should be possible. One assumption might be that there is an unpronounced (i.e. elided) ‘if ’-clause, but this assumption seems to require justification rather than disproof. A similar point can be made about fragment answers: (23) A: If Alfonso comes to the party, who will Joanna talk to? B: Bill. B#: She’ll talk to Bill.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(18) A thief might break in. (19) He would steal the silver.

James Isaacs and Kyle Rawlins 283

The analysis also requires fragment answers like (23B) to be treated as including an elided ‘if ’-clause.9 This may not appear to be surprising, at least from the perspective of an analysis of fragments like that of Merchant (2004) where such a response includes much other elided material. What is surprising, however, is that responses like B# also must be treated as including an elided ‘if ’-clause, despite looking like non-fragment answers. 4 ACCOUNTING FOR DENIAL OF THE ANTECEDENT

4.1 Presupposition denial We call conditionals non-counterfactuals when they carry the presupposition that their antecedents are possible (see Stalnaker 1968, and much other work), i.e. the context set contains both worlds in which 9 More strictly, they would have to be interpreted as having such an ‘if ’-clause present at logical form. Ellipsis is the only obvious way we see to achieve this. However, the analysis we propose accomplishes a similar result via domain restriction, without requiring anything of the form or meaning of the answer.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In the previous section, we gave several reasons against treating denial of the antecedent responses as either complete or partial answers to CQs. This leaves us with two important questions that must be answered. The first comes in two parts: what are denials of the antecedent, if they are not answers, and why do they sometimes have an issue-dispelling effect? The second question is how can we reconcile the worlds where the antecedent of a CQ is false with the partition theory of questions, i.e. how do these worlds fit into the partition (if at all)? We return to the second issue in section 5, where we provide our formal analysis of the CCP of CQs. In this section, we deal with the first issue by providing a pragmatic account of responses to CQs which address the content of the antecedent. Responses of this kind are shown to be issue dispelling when they deny the presuppositions of the conditional structure that must be in place prior to the interpretation of the question in the consequent. These responses are identified as indications of presupposition failure. In certain conditionals, denying the antecedent is issue dispelling. In others, affirming the antecedent has an issue-dispelling effect. We treat responses that fail to be issue dispelling as failures to advance the conversation towards resolving the issue raised by the question, i.e. they are infelicitous and uninformative.

284 Conditional Questions the antecedent is true and ones in which the antecedent is false.10 The non-counterfactual conditional in (24) can only be uttered in a context in which Alfonso’s coming to the party is a live option. (24) If Alfonso comes to the party, Joanna will leave. Following von Fintel (1999) (among others), we call conditionals counterfactuals when they carry the presupposition that the proposition expressed by the antecedent is not possible, i.e. not a member of the context set. The counterfactual conditional in (25) can only be uttered in a context in which it has already been established that Jo did not fix her car. Denial of the antecedent responses come in two varieties. Sometimes they have the effect of nullifying the utterance they address. The response in (26B) indicates that the conditional is no longer worth considering. (26) A: If Alfonso comes to the party, Joanna will leave. B: Alfonso isn’t coming to the party. Other times, denying the antecedent, simply fails to advance the conversation in the anticipated way. The response in (27B) does not accept or reject (27A), but rather seems to miss the point of the assertion. The example in (28) from von Fintel (1999) makes a similar point. (27) A: If Jo could have fixed the car, she would have lent it to us. B: Jo couldn’t have fixed the car. (28) A: If John had been at the party, it would have been much more fun. B: But John wasn’t at the party. A: Yes. I said if he had been there, it would have been more fun. This pattern can be given a simple and natural explanation by appealing to the presuppositions carried by different kinds of conditionals. Denial of the antecedent responses are felicitous exactly when they reject the presuppositions of a conditional. Otherwise, they simply fail to advance the conversation in the anticipated way.

10

In other parts of the paper, we refer to indicative conditionals. In this section, since the kind of presupposition the conditional carries cannot be reliably predicted on the basis of morphological mood, we temporarily adopt different terminology.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(25) If Jo could have fixed her car, she would have lent it to us.

James Isaacs and Kyle Rawlins 285

The non-counterfactual conditional in (29) carries the presupposition that it is possible that Alfonso will come to the party. The response in (29B) directly refutes this presupposition, indicating presupposition failure, in saying that it is not possible that Alfonso will come to the party. (29) A: If Alfonso comes to the party, Joanna will leave. B: Alfonso isn’t coming to the party.

(30) A: If Jo could have fixed the car, she would have lent it to us. B: Jo could have fixed the car. In general, presupposition failure is indicated by denial of the antecedent responses with respect to non-counterfactuals and affirmation of the antecedent responses with respect to counterfactuals. We see a similar effect when the consequent of the conditional is a question, and the same explanation holds. A denial of the antecedent is felicitous (and issue dispelling) exactly when it denies the presuppositions of the conditional. In non-counterfactual CQs, presupposition failure occurs when the questioner’s interlocutor claims that the antecedent is not possible, as in (31). This is the standard denial of the antecedent case considered in the previous literature on CQs. (31) A: If Alfonso comes to the party, will Joanna leave? B: Alfonso isn’t coming to the party. Presupposition failure is indicated in relation to counterfactual CQs when the questioner’s interlocutor suggests that the antecedent is true, as in (32). (32) A: If Jo could have fixed the car, would you have kept on using it? B: Jo could have fixed the car. The responses in both (31B) and (32B) are issue dispelling. The non-issue-dispelling cases are affirmations of the antecedent of a non-counterfactual conditional and denials of the antecedent of a counterfactual. These neither deny a presupposition nor address the issue raised in the consequent of the preceding conditional. All they do is fail to advance the conversation in any useful way, regardless of

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The same thing happens when the presuppositions of counterfactual conditionals are denied. The antecedent of the counterfactual conditional in (30) carries the presupposition that Jo could not have fixed that car. The response in (30B) indicates presupposition failure by making the opposite claim.

286 Conditional Questions whether the consequent is an assertion or a question, as shown by (33B) and (34B). (33) A: B: (34) A: B:

If Alfonso comes to the party, Joanna will leave. Alfonso is coming to the party. If Alfonso comes to the party, will Joanna leave? Alfonso is coming to the party.

The same holds true of responses to counterfactual conditionals (35) and counterfactual CQs (36). If Jo could have fixed the car, she would have lent it to us. Jo couldn’t have fixed the car. If Jo could have fixed the car, would she have lent it to us? Jo couldn’t have fixed the car.

The conclusion is that denials of the antecedent are not answers of any kind. They are either indications of presupposition failure (the licit cases) or simply responses which fail to advance the conversation in the anticipated way (the cases that miss the point). This treatment of the licit cases of denials of the antecedent directly captures Velissaratou’s (2000) intuition that denials of the antecedents of non-counterfactual CQs are not about the issue raised by the question. They are about the presuppositions of the structure in which the question is embedded. If they are about any issue, this issue is the issue of whether the presuppositions of the conditional are true (which may be an issue that has been salient in the discourse). They are still pertinent to the discourse (and therefore felicitous) in the sense that they indicate a significant misunderstanding of the common ground/context set. When the hearer indicates presupposition failure, this marks that the hearer did not get to the point of considering the consequent of the conditional. What is indicated is that the context set is not what the asker thought it was, and the asker must undertake some kind of (nonmonotonic) belief revision in order to determine the correct context set. Part of this revision will involve ‘undoing’the asking of the question—the context is rolled back to a state prior to the utterance of the CQ. We do not attempt to detail this process here, except to point out that it is the very same process that should happen in any case of presupposition denial, regardless of the linguistic element that bears the presupposition.

4.2 Prospects for previous analyses The semantic analysis in Velissaratou (2000) is designed to deal exclusively with non-counterfactual CQs. Questions in Velissaratou

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(35) A: B: (36) A: B:

James Isaacs and Kyle Rawlins 287

5 THE FORMAL INTERPRETATION OF CQS We provide a theory of CQs in which their interpretation is decomposed into two steps. The first step involves making a temporary copy of the current context and updating the copy with the propositional content of the antecedent. In the second step, the question in the consequent raises an issue relative to the temporary context. The goal of this analysis is to account for the fact that worlds where the antecedent of the conditional is false do not seem to play any role in the alternative structure introduced by the question. In section 5.1, we discuss previous decompositional approaches to conditionals and the relevance of temporary contexts to theories of conditionals and modal subordination. In section 5.2, we provide an

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(2000) denote alternatives which are taken to overlap when the question is a CQ in order to treat denial of the antecedent responses as answers in a technical sense. The fact that denial of the antecedent responses are issue dispelling provides motivation for this move. However, denial of the antecedent responses to counterfactual CQs is not issue dispelling. As shown in (35), the issue remains raised following the response in (35B). As a result, the system in Velissaratou (2000) fails to generalize, i.e. it would have to provide a different set of alternatives for non-counterfactual and counterfactual CQs. The same is true of the tripartition account. Our pragmatic treatment of denial of the antecedent could be appended onto either of the previous analyses. However, doing so would make the semanticization of denial of the antecedent responses in both these theories redundant—each analysis places the worlds where the antecedent is false in the alternative structure, and those parts of the alternative structure would never be used. In other words, having a pragmatic analysis removes the motivation for the third cell (for Hulstijn 1997), and for the overlap between alternatives (for Velissaratou 2000). We must still account for worlds where the antecedent is false when interpreting indicative CQs, but without additional motivation, it seems best to try to keep mutual exclusivity as a property of questions if possible. The pragmatic analysis of denial of the antecedent allows us to consider analyses where these worlds never participate in the alternative structure involved in the question at all. We propose such an alternative, based on the semantics of modal subordination, where the mutual exclusivity property is not abandoned, and questions can be given a very standard interpretation. We turn to our analysis in the next section.

288 Conditional Questions overview of the stack-based semantics proposed in Kaufmann (2000) to analyse modal subordination. In order to maintain temporary domainrestricted contexts, we adopt a version of Kaufmann’s semantics in our treatment of conditionals. The technical definitions are provided in section 5.3. In section 5.4, we apply these definitions to interpret CQs. In section 5.5, we return to the problems raised for previous analyses, and discuss how our analysis accounts for them.

5.1 Decomposing conditionals

(37) For any context set c, clause A, and clause B: c + If A; B ¼ c n ðc + A n c + A + BÞ The two update steps correspond to the parts c + A and c + A + B in the formula above. The rest of the machinery is simply to get the effect of the update on the temporary context to have the right effect on the resulting main context—any worlds that were lost on the second step and are present in the main context will be lost from the main context as well. The entire contents of c + A + B are never needed again—no future interpretation makes reference to them, and there is no formal mechanism for retaining them. Various complications of conditionals can be added to this kind of treatment straightforwardly, although they are for the most part irrelevant here. Heim (1992), for instance, takes the temporary context to be the set of closest worlds in the context set where the antecedent holds, not just the set of worlds in the context set where it holds: (38) For any context set c, clause /, and clause w: c + ½½if /w ¼ fw 2 c : Simw ðc + /Þ + w ¼ Simw ðc + /Þg (where Simw(c) picks out the closest worlds to w in c)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The dynamic view of conditionals (Stalnaker 1968; Karttunen 1974; Heim 1982, among others) is (reinterpreting slightly) that they involve a two-step update procedure. First, a derived context is created by updating the speech context with the antecedent of the conditional. Second, the derived context is updated with the consequent. In the standard view, at this point, the derived context is thrown away, with any relevant information in it percolating up to the main context. For instance, Heim (1983) gives the following CCP for a conditional (where M \ N stands for the intersection of M with the complement of N, i.e M \ (WN)):

James Isaacs and Kyle Rawlins 289

(39) a. A thief might break into the house. b. He would take the silver. The clearest indicator of the subordination is that the pronoun ‘he’ in (39b) cannot possibly be bound by ‘a thief ’ in (39a), but that it still covaries with the choice of thief. However, this point is independent of the presence of a pronoun, as the following possible response to (39a) from Farkas and Ippolito (2005) shows:11 (40) I would be unhappy. What (40) says, following (39a), is that, restricting ourselves to talking only about the case where a thief breaks in, the speaker would be unhappy. It says nothing about the speaker’s happiness in the case where a thief does not break in. A temporary context approach to modal subordination can capture this interpretation very directly by taking (39a) to introduce a temporary context, where hearers assume that ‘A thief will break in’ is true. Subsequent sentences that contain modal subordination cues (e.g. ‘would’ in (40)) are interpreted relative to the temporary context, not the main context. Effects of updating the temporary context percolate up to the main context, just like in the dynamic analysis of conditionals. Worlds that are lost in the temporary context due to interpreting sentences like (40) are also lost in the main 11 Farkas and Ippolito (2005) argue that the effect of modal subordination is independent of any presupposition trigger.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Additional discussion of conditionals and the interaction of conditionals and modals discussed by Kratzer (1986) is given in section 9. We propose that the dynamic account of conditionals gives us the right tools to model the interpretation of CQs, if it is acknowledged that temporary contexts may be needed outside the sentence that introduces them. In our analysis, the question in the consequent of a CQ partitions the temporary context, instead of asserting in it, rendering it inquisitive. Because the context is inquisitive, it cannot be discarded, and following answers will be interpreted relative to the temporary context. The acknowledgement that temporary contexts can persist has been made in some parts of the literature on modal subordination (Frank 1996; Kaufmann 2000) and mood (Farkas 2003). In this line of research, modal subordination introduces a temporary context in which something that is not necessarily true has been, for the moment, supposed to be true. A classic example of modal subordination from Roberts (1989) is given in (39).

290 Conditional Questions context. However, worlds that were supposed out of the way at the time of creation of the temporary context are never affected. We can see from (39a) that ‘might’sentences can introduce temporary contexts, and so can conditionals, as the following examples show: (41) a. If Edna forgets to fill the birdfeeder, she will feel very bad. b. The birds will get hungry. (Roberts 1989, example 2) (42) a. If a thief breaks into the house, he will take the silver. b. If in addition he finds the safe, he will open it. (Frank 1996, chapter 3, example 79)

5.2 The stack model We use the stack-based model of Kaufmann (2000) as the formal tool for maintaining temporary contexts. In this model, utterances are not interpreted relative to single contexts, but to stacks of contexts. Utterances that involve the creation of temporary contexts cause them to be added to the stack, where they may remain if needed. When an utterance triggers the creation of a temporary context, the utterance involves making assumptions, or the act of supposing. Kaufmann (2000) uses this model to give an account of modal subordination—sentences that introduce modal subordination (‘A thief might break in’) involve temporarily making the assumption that the subordinating possibility is true. Sentences that continue or trigger modal subordination (‘He would steal the silver’) involve interpretation relative to the temporary context. Kaufmann (2000) also makes the parallel between antecedents of conditionals and utterances that introduce modal subordination contexts. A case where there is no modal subordination will typically involve only one context on the stack, corresponding to the standard Stalnakerian context of utterance. The effect of a plain assertion on this kind of stack will be the same effect a plain assertion would have on that one context set in the standard analysis. An utterance that introduces modal subordination pushes (inserts at the top) a temporary context onto the stack, and future utterances that continue the modal subordination see that temporary context as the main one. Utterances

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In each (b) sentence, the claim is made only relative to cases where the antecedent of the conditional in (a) is true. In (41b), the birds will only get hungry if Edna forgets to fill the feeder. The whole conditional in (42b) is only true if the thief breaks in in the first place. The dynamic approach to conditionals can handle these effects, if the temporary contexts it needs anyway are allowed to stick around.

James Isaacs and Kyle Rawlins 291

5.3 Formal implementation: assertions, questions and modally subordinated questions We now give the technical definitions that make up the stack model, and in the next section, apply them to the interpretation of CQs. Utterances are interpreted relative to macro-contexts; a macro-context for our purposes is a stack of Stalnakerian contexts (which we treat as simply context sets, abstracting away from other necessary elements). A macro-context is an ordered pair, either empty or consisting of a context and a macro-context: (43) Deﬁnition: macro-context a. Ææ is a macro-context. b. If c is a (Stalnakerian) context and s is a macro-context, then Æc, sæ is a macro-context. c. Nothing else is a macro-context. 12 The use of a stack of temporary contexts has precedent in the analysis of presupposition projection in Zeevat (1992). Similar stack-based mechanisms have been used for logically modelling temporary assumptions; see the ‘Fantasy Rule’ of Hofstadter (1979: 183–5) and further development in Zeinstra (1990), as well as discussion of both in Muskens et al. (1997). See Kaufmann (2000) also for discussion of other ways that stacks have been used in modelling discourse.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

that do not contain modal subordination cues (e.g. simple past sentences, though we do not concern ourselves here with what these cues might be) cause the stack to be popped before interpretation—an operation where the top element is removed and discarded.12 A conditional, in this kind of system, behaves much like a conditional in the standard dynamic analysis. First, a derived context is created by making a copy of the current top of the stack and updating it with the antecedent of the conditional. This context is pushed onto the top of the stack. The consequent of the conditional is then interpreted relative to the stack, but the stack is guaranteed to have a temporary context as its top element, so the consequent will see that context as the main one. The stack-based analysis makes explicit what has always been implicit in the dynamic analyses of conditionals—interpretation of the antecedent involves making a temporary assumption. The difference is that the stack provides a way for the context to be retained after its immediate use. If the conditional is declarative, the update to the top context in the stack will be reductive—it will remove worlds just as any assertion does (this will also have an effect on other elements of the stack, discussed later). We propose that if the conditional is a CQ, the second step involves partitioning the temporary context, instead of reducing it.

292 Conditional Questions d. If s is a macro-context, then sn is the nth context (counting from 0 at the top) and jsj is its size (excluding its final empty element).

(44) Deﬁnition: push operator For any macro-context s and context c: push(s,c) ¼def Æc,sæ (45) Deﬁnition: pop operator For any macro-context Æc, s#æ: popðÆc; s#æÞ ¼ def Æc; s#æ if s# ¼ Ææ; s# otherwise Given a macro-context Æc#,Æc$,Ææææ and a context c, the operation push(Æc#, Æc$,Ææææ, c) results in Æc, Æc#, Æc$, Æææææ. The operation pop(Æc, Æc#, Æc$, Æææææ) would then result in Æc#, Æc$, Ææææ. In section 2.1, we discussed the definitions of update functions on single contexts. Assertions involved reductive updates—they removed some worlds entirely from the context set and increased the information content of that set. Now we consider how to extend this to stacks of contexts, i.e. discourse situations where temporary assumptions have been made. The most obvious possibility is that assertions will affect the top member of macro-context. That is, we would have a definition like: Æc, s#æ + [Assert /] ¼def Æc 4 /, s#æ. This is simple and appealing, and will work exactly right for the case where the macro-context has only one context set on it. However, it should be clear that this would not by itself be what we want. The reason is that any information gained during a conditionalized or subordinated sentence could be lost by popping the stack. This information must live on somehow—it must ‘percolate’ down to other members of the macro-context. One way of accomplishing this, which we do not develop here, is to preserve it at the time of the pop operation. The

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We refer to a macro-context either by using variables s, s#, and so on or as Æc, s#æ where c is the top element of the macro-context and s# is the rest of the macro-context. The case where no suppositions have been made is going to involve a stack like s ¼ Æc, Æææ. In this stack, c is s0, and the size of the stack is 1. A more complex stack might look like s# ¼ Æc, Æc#, Æc$, Æææææ. The size of s# is 3, and c would be s0, c# would be s1 and c$ s2. This size stack might involve a case where stacked ‘if ’-clauses have introduced two assumptions. Push and pop operators, which add to and remove from a macro-context, respectively, are defined as follows:

James Isaacs and Kyle Rawlins 293

other way, which we develop (following Kaufmann 2000), is to percolate it down immediately. In Kaufmann (2000), reductive updates center around the operation of learning in lower contexts that the top context supports the content of the update. This is formally captured by the ‘ operator. The version of the ‘ operator we use here takes three contexts as arguments, and an instance of it will look like ‘‘ (c, c#, c#4 /)’. This should be read as, roughly, ‘Discourse participants learn in a context c that another context c# supports /’. (In general, they learn that c# supports information gained between c# and c#4 /.) (46) gives the formal version of it:

( ‘ðc; c#; c$Þ ¼def

) :dw 2 W s:t: Æw1 ; wæ 2 c#or Æw; w2 æ 2 c# Æw1 ; w2 æ 2 c or Æw1 ; w2 æ 2 c$

What this amounts to is really just (c c#) [ (c \ c$), but care must be taken to use the right version of ‘ ’ as we are dealing with world pairs here. A pair of worlds is in ‘ (c, c#, c#4 /) iff (i) neither world in the pair appears in c# (but the worlds are in c) or (ii) the worlds would remain if c# were updated with /. Condition (i) corresponds to the case where the worlds have been supposed out of the way in c#; these worlds can have nothing said about them by further updates to c#. Clause (ii) corresponds to whatever the update represented by the c$ argument (for us, typically c#4 /) does to c#. Building on this, the macrocontext change potential (MCCP) of assertions is given in (47). (47) Assertive update on macro-contexts For any macro-context s and clause /: s + ½Assert / ¼def s# where js#j ¼ jsj ¼ n and s#i ¼ ‘ðsi ; s0 ; s0 4/Þ for all i; 0 < i < n This definition is very similar to Kaufmann’s (2000) ‘conclude’ operation, but here we go further in associating the operation with the act of asserting. How it is so associated can be seen in two ways: either it is just a fact about how declaratives are interpreted or the assertive force is associated with a speech–act operator that has syntactic

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(46) Deﬁnition of ‘ For any contexts c and c#, and c$:

294 Conditional Questions

(48) Inquisitive update on macro-contexts For any macro-context Æc, s#æ where c is the top member, and s# is a stack, and clause /: Æc, s#æ + [Question /] ¼ def Æc /, s#æ On this view, a question acts only on the top element of the stack, partitioning it in a straightforward manner. This allows us to keep the standard properties of the theory of questions (mutual exclusivity and exhaustivity). The idea behind (48) is that questions are asked only relative to whatever assumptions are present in the context at a given time, and the issues raised by questions are also necessarily relative to those assumptions. The answers to questions may be informative, and this information should percolate. This notion of questioning can be understood best in concert with the following constraint: (49) Inquisitiveness constraint A macro-context may not be popped if the top element is inquisitive. The idea is that issues persist only so long as they are not resolved, but they must persist long enough to be resolved or otherwise dispensed with. One other way of dispensing with them, for instance, would be to deny the presuppositions of a CQ. This constraint ensures that any

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

presence (perhaps as a force feature). For convenience, we will take the second view to be right. One key to understanding what (47) is supposed to do is the fact that the effect of an assertive update on the top member of a macrocontext is always the same as the 4 operator. This is clearest for macrocontexts of size 1: for any context c and (speech–act-less) clause /, Æc, Æææ + [Assert /] is equivalent to Æc 4 /, Æææ. Thus, this notion of assertion is simply a generalization of the standard dynamic notion of assertion (due to Stalnaker 1978)—it is generalized to handle suppositions. An example of this operation appears in section 6 below. Next, we must generalize the notion of questioning. This case is somewhat different than the case of asserting—the simplest idea can work. This is the idea that what a question update does to a stack is just what a operator would do on the top element. The reason this can work is that questions, under Groenendijk (1999), do not contribute to information gain. They raise issues by disconnecting worlds, but they do not remove any worlds themselves. In terms of licensing answers, the effect of a question is much more immediate than the effect of an assertion—intuitively, an issue is raised in an immediate and local way. What this means is that a simple definition like (48) can work.

James Isaacs and Kyle Rawlins 295

(50) Inquisitive update on macro-contexts (immediate percolation version; cf. (48)) For any macro-context s and clause /: s + ½Question / ¼def s#

where js#j ¼ jsj ¼ n and s#i ¼ ‘ðsi ; s0 ; s0 /Þ for all i; 0 < i < n

This new notion of questioning is more complicated, but it is also much more homogeneous with the definition of Assert above. We do not know of any empirical ways of deciding between (48) and (50).13 Percolating issues does require abandoning the strong form of mutual exclusivity (that all context sets always contain only alternatives

13

It is certainly true that the (meta) information that a particular issue was raised at all may need to be available. But this is also true of assertions, and the present system models none of these metainformation (i.e. it does not model the history of the discourse at all).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

answers to a CQ will be interpreted with the temporary context on top. At this point, an apparent difference between questions and assertions appears. Assertions provide information, and this information persists. Questions raise issues, and in our analysis so far, these issues do not percolate down the stack, and therefore do not persist. This might be perceived as a stipulative difference between questions and assertions. The difference, however, is really between information and issues. Our present analysis percolates information immediately on assertion, but this is not a necessary property of a stack-based analysis—percolation could happen at the point where the stack was popped. (This would be more like the stack-based logic of Zeinstra 1990; Muskens et al. 1997, where we do not learn the non-local impact of hypothetical assumptions until we apply the definition for the pop operation.) So the claim implicit in (48) is that information percolates and issues do not. This allows us to keep the mutual exclusivity and exhaustivity properties in a very strong form—if issues percolated they would percolate to a point on the stack where there are worlds not in any cell of the original partition. At that point, one would be forced to treat them as being in every cell (following Velissaratou 2000) or as being in no cell (abandoning exhaustivity). This would involve applying the ‘ operator to the definition of questioning as well:

296 Conditional Questions

(51) A: B: (52) A: B:

A thief might break in. Would he steal the silver? A thief might break in. Would you be upset?

In each of these cases, the question is asked relative to the subordinate context. In (51), the questioner is not asking about what anyone would do in the case where a thief did not break in. In (52), the questioner is only asking whether the A speaker would be upset in the case where a thief did break in—discarding entirely the issue of upsettedness in the case where a thief did not. On the surface, these data would seem to cause problems for the theory of questions, since (51) and (52) obviously cannot be interpreted exhaustively relative to the entire context set or domain of possible worlds. This difficulty parallels the problems centering around denials of the antecedent (see sections 3–4), and our account has a simple solution. Modal subordination involves pushing a temporary context onto the stack (assuming Kaufmann 2000 or a similar analysis); in the case of (51), this is the previously top context, updated with the assumption that a thief will break in. The modally subordinated question partitions only the top context of the stack, and the alternatives it creates are exhaustive and mutually exclusive relative to that context. Therefore, the question in (51) or (52) would partition a context where it has been temporarily assumed that a thief does break in. The issue raised by the question automatically has nothing to do with cases where a thief does not break in, a natural consequence of our use of the stack model.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

that are mutually exclusive). However, the notion in (50) does retain a limited version of mutual exclusivity, at least in concert with the Inquisitiveness Constraint—the currently ‘active’ context will always be partitioned. Either way, mutual exclusivity is retained as a property of the operation, and so plausibly, of the meaning of questions themselves. Because there is no deciding empirical choice, we continue the present section and example with the simpler version in (48). However, we return to the issue in section 7, and we make use of the more complicated notion in (50) in our discussion of embedded CQs in section 8. There it turns out to simplify the analysis of embedded CQs. Before we turn to the analysis of CQs, note that this semantics for questions also gives us a straightforward account of questions in modal subordination contexts:

James Isaacs and Kyle Rawlins 297

5.4 Conditional questions We are now in a position to discuss the interpretation of conditionals.14 The dynamic theory of conditionals decomposes their interpretation into two updates: one corresponding to the antecedent and one corresponding to the consequent. Our version of this, focusing on indicative conditionals, is given in (53): (53) MCCP of an indicative conditional: For any macro-context s, ‘if ’-clause [if /], and clause w: s + [if /, w] ¼def s + if / + w

(54) MCCP of an ‘if’-clause For any macro-context s and ‘if ’-clause [if /]: s + if / ¼def push(s, s0 4 /) Admittance conditions: ‘If / ’ is admissible in a macro-context s iff s0 4 / 6¼ ; The admittance conditions provide the standard presupposition for indicative conditionals (going back to Stalnaker 1968) that the content of the antecedent is possible.15 Our analysis is a truth-conditional, suppositional theory of conditionals: the antecedent of a conditional directly introduces an assumption into the context. Suppositional theories of conditionals (see Adams 1965; Mackie 1973; Ga¨rdenfors 1986, among others) are often traced back to the following remark in Ramsey (1931): (55) ‘‘If two people are arguing ‘If p, will q?’ and are both in doubt as to p, they are adding p hypothetically to their stock of knowledge and arguing on that basis about q . . .’’ (Ramsey 1931: 247) For us, adding p hypothetically involves (for indicative conditionals) creating a temporary context by adding p to a copy of the main 14 Our analysis is set up to account for CQs regardless of their type. However, creating temporary contexts for counterfactual CQs involves further complications. For the sake of simplicity, we concentrate on non-counterfactual CQs in the formal component. See section 7.2 for more discussion. 15 We treat this as a presupposition, not as a general felicity condition in the style of Stalnaker. It is not clear that the difference matters, as long as it is subject to presupposition projection. This admittance condition is often accompanied by another felicity condition, which we do take to be general: a ban on updates that do not reduce the context set.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This is a two-step update based on the standard dynamic semantics for conditionals. It has been generalized to work in the stack model (following Kaufmann 2000), and to handle question complements. The MCCP of ‘if ’-clauses is given in (54):

298 Conditional Questions

5.5 Explaining the data We are now in a position to return briefly to the problems raised in section 3. Some of these have already been dealt with in section 4, namely those involving denials of the antecedent, but there are a few remaining. The availability of ‘would’ in answers to CQs now has a straightforward explanation, or at least, we have reduced it to the problem of why ‘would’ signals modal subordination in general. The grammatical mechanisms for interpreting answers to CQs are the same as those involved in modal subordination. The issues of what polarity particles such as ‘yes’ and ‘no’ mean, and whether answers to CQs in general involve a covert ‘if ’-clause, also have a linguistically straightforward solution. Because of the modal subordination, there is no need to assume any covert ‘if ’-clauses—the use of the stack model for domain restriction sees to this. It also allows polarity particles to have the same meaning one would assign to them on the basis of non-conditionalized questions. In general, answers are

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

context, and pushing the temporary context onto the stack. How we treat the consequent depends on what kind of speech–act it involves—i.e. what its force is. If it is an assertion, we see the effect on the context just as in (47)—the operation is reductive and affects both the top context (the supposed context) and the original top, s1. If the consequent is a question, the top of the context will be partitioned. In this case, the context cannot be popped without resolving (or discarding) the issue that the interpretation of the consequent raises. Subsequent answers are automatically interpreted with respect to the unpopped stack, where the antecedent of the conditional is still supposed. Note that this treatment of conditionals drastically complicates the task of giving a definition of entailment, as conditional sentences will lead to stacks of different sizes than non-conditional sentences, and many entailments between the two will trivially not go through under a standard dynamic notion of entailment. We postpone this task until our treatment of embedded CQs (see section 8.2). This account of conditionals, and therefore CQs, inherits the standard dynamic treatment of presupposition projection. This is our technical analysis of the points raised in sections 3 and 4. Responses to CQs that deny the presuppositions of the antecedent do so prior to the interpretation of the question. This indicates presupposition failure on the part of hearers of the CQ which happens during the act of supposing, before the context could even be partitioned.

James Isaacs and Kyle Rawlins 299

6 EXAMPLE COMPUTATION OF A CQ This section goes through an example of the interpretation of a CQ. Note that here we use the simpler definition of questioning from (48). We assume an input macro-context s ¼ Æc, Æææ for some context c in (56). The facts of the worlds in the context are as follows: Alfonso comes to the party in w1, w2 and does not come in w3, w4 and Joanna leaves the party in w1, w3 and does not leave in w2, w4. This is shown in (56) with a direct representation of the stack on the left, and a picture of the stack on the right. The picture shows the worlds in a top-down way—the white area represents the active top context of the stack. (56)

Asking a CQ relative to the initial context (s + [If [Alfonso comes to the party]], [will Joanna leave?]) leads to a two-step process of

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

interpreted relative to the top element of the stack, including fragment answers. The analysis of counterfactual CQs also has a solution in sight. Affirmations of the antecedent have already been explained (as presupposition denials). The interpretation of counterfactual CQs then reduces to the general problem of understanding counterfactuals, and all that is needed is to integrate an existing dynamic analysis of counterfactuals into the stack model. Potential candidates include von Fintel (2001) and Veltman (2005). We return to the topic of this integration in section 7.2. There is one issue that is not immediately clear under our analysis, and this is a version of one of Velissaratou’s (2000) objections to the tripartition account. It is unclear why full conditionals should be available as answers to CQs. In fact, examples of this sort often seem mildly infelicitous, but this may be because of a general dispreference for redundant answers. If interpreted relative to the temporary context, a conditional answer will do no harm, but it will involve a vacuous update when interpreting the antecedent, which should perhaps be independently ruled out. One possibility is that answers of this kind are actually ‘factual conditionals’, similar to the one found in the following discourse: A: ‘I’m tired.’ B: ‘Well, if you’re tired, take a nap’ (Zaefferer 1990, 1991; Iatridou 1991). We leave this issue for future consideration. The next section walks through a sample computation of a CQ.

300 Conditional Questions interpretation. The first step involves creating a temporary context by making a copy of the main context and altering it to ensure that the antecedent is true: c 4 Alfonso is coming to the party. The temporary context is then pushed onto the stack resulting in the stack shown in (57) in which the temporary context is the active (top) context. Since, by stipulation, the antecedent is true in worlds w1 and w2 but not in w3 and w4, only world pairs containing w1 and w2 are present in the temporary context. This is shown in the picture to the right as well, where the dashed oval represents the lower context on the stack and the solid oval the top context.

The consequent question is then asked relative to the temporary context. In w1, Joanna leaves the party but in w2 she does not. Since these worlds resolve the issue raised by the question differently, they populate different cells of the partition. The polar question induces the following bipartition on the temporary context: (58)

An answer is interpreted with respect to the temporary context. A positive answer eliminates the cell containing the pair Æw2, w2æ. The temporary context is rendered uninquisitive after the ‘yes’ answer. The answer (which is an assertion) also has an effect on other members in the stack—the information gained must percolate down to the other contexts on the stack before the temporary context is popped (see the definition in (47)). In s#1$ below, the world pairs that have been removed are removed as the result of learning in s$1 that the top of the stack supports the ‘yes’ answer—of updating with s$1 ‘s$0 yes. This amounts to removing worlds where both the assumption created by the antecedent is true, and the answer is false.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(57)

James Isaacs and Kyle Rawlins 301

(59)

At the point at which the temporary context is no longer inquisitive, it can be popped off the stack, assuming that no further subordination cues appear. (60)

7 THE DYNAMICS OF INFORMATION AND ISSUES The main analysis assumed so far treats information and issues in a heterogeneous way. On assertive update, information percolates from temporary contexts to main contexts immediately, and issues do not percolate at all. This distinction serves three purposes: (i) it allows us to maintain a partition semantics; (ii) it captures the (debatable) intuition that questions are ‘local’ to temporary contexts in a way that assertions are not and (iii) it allows for a very simple statement of the kind of belief revision that occurs when the presupposition of a CQ is denied (the stack is popped). However, it is worth considering what the cost would be of homogenizing the treatment of issues and information. We have already given a definition which makes this happen—the definition in (50). In this section, we return to the question of what this involves. We also discuss in more detail what happens to information and issues in counterfactual CQs.

7.1 Percolating issues Allowing issues to percolate forces us to give up several of the principles we have argued for, but it may be that the cost is worth it. The result is effectively a dynamicized version of Velissaratou’s (2000) static analysis of CQs. The only required change from the simpler analysis (where

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This is the same state that would be reached by a declarative conditional ‘If Alfonso comes to the party, Joanna will leave’.

302 Conditional Questions issues do not percolate) lies in the definition of a question update. What we need to ensure is that, if a question removes world pairs from the top context, it does so from every context on the stack. In the definition for Question (repeated from (50) above), we did this by using the ‘ operator in a way exactly parallel to the definition for Assert. (61) Inquisitive update on macro-contexts (immediate percolation version) For any macro-context s and clause /: s + ½Question / ¼def s#

This alternative MCCP for questioning should still be taken in concert with the Inquisitiveness Constraint from (49): (62) Inquisitiveness Constraint A macro-context may not be popped if the top element is inquisitive. In the alternative system, where issues percolate immediately, this constraint now enforces partition-hood only on the top context. Lower contexts are now guaranteed to be non-partitions following the raising of an issue (similar to question meanings in Velissaratou’s account). However, as long as an issue is resolved before popping the macrocontext, the new top will be a safe, non-inquisitive, partition. This version of the semantics shares with Velissaratou’s semantics that mutual exclusivity is not absolutely enforced. However, our dynamic (stack-based) version, unlike the static version, does partially retain Hamblin’s exhaustivity/mutual exclusivity constraint: the top element of the stack is always a partition, though lower elements never are. The meaning of a question also retains exhaustivity and mutual exclusivity. The dynamic effect of a question (and thus its meaning) always has a partitioning effect on the top context on the stack, and its effect on other contexts is always mediated through that top context. This is not so on Velissaratou’s static version, where the meaning of a question is directly not a partition. Note that in a plain context with one CQ asked, the lower member of the stack will correspond to the meaning of a CQ in Velissaratou’s analysis. In this sense, the immediate percolation version of our analysis is a dynamic reconstruction of Velissaratou’s analysis (modulo the addition of a mechanism for presupposition denial).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

where js#j ¼ jsj ¼ n and s#i ¼ ‘ðsi ; s0 ; s0 /Þ for all i; 0 < i < n

James Isaacs and Kyle Rawlins 303

The general conclusion of this section is that a version of the partition theory can be retained regardless of whether issues percolate. If they do, the analysis of questioning is complicated. We will see shortly that this simplifies the treatment of embedded CQs. First, there are some loose ends about counterfactuals to wrap up.

7.2 Percolation in counterfactual conditionals

16 Note that this is in some sense separate from the effect of the antecedent on the truth conditions of the counterfactual: the antecedent also causes us to ignore worlds where it is false.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In our analysis, information gained during the interpretation of a counterfactual conditional never percolates to the main context set, including information gained from answers to counterfactual CQs. This is achieved by maintaining that information can only percolate to a lower context if it is compatible with that context (see the definition of ‘ in (46)). However, it is worth considering where, if anywhere, information gained during the interpretation of a counterfactual is retained. At present, our system predicts that it goes nowhere. This amounts to the claim that counterfactuals lack truth conditions, which is certainly a respectable view, but it is not one we happen to agree with. We have also been non-specific about which context gets pushed onto the macro-context when updating with the antecedent of a counterfactual conditional. It clearly cannot be a subset of the main context, as we have been assuming for non-counterfactuals. Our hope is that a ready-made dynamic theory of counterfactuals can be applied directly. One prime candidate is the analysis proposed by von Fintel (2001) (also, Veltman 2005). In von Fintel (2001), counterfactuals update a context that has been expanded to what is called the modal horizon. von Fintel (2001) provides the following dynamic procedure for choosing the context for a counterfactual: ‘If a conditional is accepted as an assertion, the context will first be changed to expand the modal horizon if the antecedent was not already considered a relevant possibility. Then, the conditional will be interpreted in the new context.’ This expansion of the context involves adding worlds that make the antecedent true (in line with the sphere condition of Lewis (1973), i.e. closest in some relevant sense).16 This approach provides a straightforward way of choosing the appropriate context to push onto the macro-context when interpreting a counterfactual CQ. The remaining issue is where the information gained ends up at the end of a counterfactual. We have been implicitly treating contexts on the macro-context as ‘views’ of some part of the lowest (main) context on the stack. They are what this main context looks like given some

304 Conditional Questions

8 EMBEDDED CQS Just as plain interrogatives can appear as the complement of certain verbs (‘know’, ‘wonder’, ‘ask’, etc.), conditionals are also possible. And like in matrix cases, embedded conditionals seem to have an effect on the scope of possible (or actual) answers to embedded questions. For example, when Alfonso knows the answer to a CQ, as in (63), what he knows is much more restricted than if he knows the answer to the plain question. Similarly, if he wonders about the answer to a CQ, as in (64), he wonders something quite different, and more restricted, than wondering about the answer to the plain question. (63) a. Alfonso knows whether the party will happen. b. Alfonso knows whether the party will happen if it rains. (64) a. Alfonso wonders whether the party will happen. b. Alfonso wonders whether the party will happen if it rains. The goal of this section is to show that one version of our analysis of CQs can be extended to work in such cases. We focus on the verb ‘know’, and we ignore the question of whether the complement of extensional interrogative selecting verbs really denotes a question act, as opposed to something like a question radical (see Krifka 1999; McCloskey 2006 and work cited therein). As far as we can see, our analysis extends to intensional verbs regardless. The analysis works by defining a cross-categorical notion of support (which can be thought of as entailment) in the stack model, and taking

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

temporary assumptions. However, given expansion to parts of the modal horizon as a possibility, this cannot possibly be right, because the modal horizon is typically not a part of the lowest context on a stack. Perhaps, then, every context on the stack simply provides a view of some larger information store. By default, we work with views that are known to be factual—ones that involve mutual public beliefs—but this is not the whole of the information store. Information does not really percolate down the stack, but percolates to the larger information store, and so all views of that store might change. The Inquisitiveness Constraint on popping the stack (see (49)) might then be generalized to a constraint on any view of the information store. This generalized Inquisitiveness Constraint states that any view of the store must involve a partition. This is yet another way of realizing Hamblin’s Picture, the idea that questions involve a mutually exclusive set of exhaustive alternatives. We leave formalization and exploration of this idea for the future, and turn now to the analysis of embedded CQs.

James Isaacs and Kyle Rawlins 305

‘know’ to mean that the knower’s epistemic state (cross-categorically) supports the meaning of the complement of ‘know’. Our approach is derived from Aloni and van Rooy (2002) and has been generalized to allow for conditionals on the stack semantics. We define the relevant notion of support using the version of our semantics for CQs discussed in section 7.1 where issues percolate immediately on update, and lower contexts are not partitions (but rather are broken into Velissaratou-style alternatives). The appropriate notion of support turns out to be simpler to define on this version of the analysis, for reasons discussed below.17 First, some preliminary facts. Embedded CQs are better if the ‘if ’clause is not fronted, as in (65) and (66).

The ‘if ’-clause is possible in the fronted position, but needs to be uttered with a very particular intonation. (67) Isabella knows, if Alfonso showed up, whether Joanna would leave. (68) Isabella wondered, if it rained, whether the picnic would happen. There appears to be no interpretive difference caused by the position of the ‘if ’-clause (at least, no more than is caused in matrix conditionals), and we assume that if examples like (67) and (68) are degraded, the issue is a syntactic one. We further assume that semantically, the ‘if ’-clauses in sentences like (65) are interpreted outside the scope of the question operator. (This assumption is an obvious potential point of contention.)

8.1 A dynamic, cross-categorical semantics for ‘know’ We follow Aloni and van Rooy (2002) in giving a Hintikka (1962)style semantics for ‘know’, that is dynamicized and based on crosscategorical support. We will start from Groenendijk’s (1999) simpler notion of support. Here is the essence of semantics by Aloni and van Rooy (2002), not yet adapted for the stack theory: (69) Let Kw(x) be the set of epistemically accessible worlds to individual x from world w.18 17 Thanks to Jeroen Groenendijk for comments on this section, which pushed us towards the current (much simpler) analysis. 18 One simple candidate for Kw(x) would be (informally) \{p : x believes p at w ^ w 2 p}—as far as we can tell this gets the right results for our purposes. Of course, there is also much discussion in the literature of the fact that the objects of beliefs are not propositions, but we will ignore this here.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(65) Isabella knows whether Joanna would leave the party angry if Alfonso showed up. (66) Isabella wondered whether the picnic would happen if it rained.

306 Conditional Questions (70) Preliminary CCP for ‘know’ sentences For any individual-denoting DP a, clause /, and context c: c + [a knows /] ¼def {w 2 cj Kw([[a]]w,c) supports /}

(71) Support (Groenendijk 1999 version) For any sentences / and w: / supports w iff "c 2 PðW 3 WÞ : c + / ¼ c + / + w This is a fairly standard dynamic notion of support/entailment, which has updated to work for contexts as equivalence relations, allowing it to handle questions. Intuitively, an assertion will support a question when it is redundant to ask the question after the assertion has been made. This is exactly the case where the assertion would have answered the question, if it had occurred in the opposite order. Following Aloni and van Rooy (2002), we generalize the epistemic accessibility function to handle contexts of world pairs and give a revised CCP for ‘know’ sentences.20 The epistemic accessibility function K* in (72) makes accessible any world pairs if both the worlds are epistemically accessible from the index provided to K*. (72) For any world w and individual x: ( !) 2 K ðxÞ w 1 w Kw ðxÞ ¼def Æw1 ; w2 æ ^ w2 2 Kw ðxÞ

19

For the sake of clarity, in this section we abuse the + notation, in not discriminating between cases where / is a set of world pairs (a proposition) and where it is a sentence (a linguistic object). In the case where / is a set, it is straightforward to define update on contexts as set intersection, and it also would be straightforward to extend this to the stack semantics. 20 Aloni and van Rooy use a more complicated notion of support than that in (71), but the differences are not important to us here.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In prose, someone knows a thing just in case their epistemically accessible worlds, Kw(a), support that thing. This idea could be spelled out with any definition of support, but it is particularly useful with a cross-categorical notion of support (following Groenendijk 1999, and much other work) which makes the notion well defined between declarative and interrogative sentences (or between assertions and questions). In particular, assertions support questions to which the assertion provides an answer. By using this notion of support, we can allow both questions and assertions to provide the value of / and w.19 Groenendijk’s (1999) definition of support is given in (71).

James Isaacs and Kyle Rawlins 307

A CCP for ‘know’ sentences making use of this is given in (73). (73) CCP for ‘know’ sentences (2nd version) For any individual-denoting DP a, declarative or interrogative clause /, and context c: Æw1 ; w2 æ 2 c ^

( c4½a knows / ¼ def

Kw 1 ð½½aw1 ;c Þ supports / Kw 2 ð½½aw2 ;c Þ supports /

)

8.2 Support in the stack model Something more needs to be said about what the partition on a’s epistemically accessible worlds looks like when / is a CQ. In particular, we need to update the notion of support to work with the stack model. Given the assumptions in section 7.1, this turns out to be relatively straightforward. With the main analysis presented, it is significantly more complicated. The way of defining support used above effectively involves comparing contexts. Given the assumption that issues percolate immediately, the most straightforward way to implement support is to do updates in parallel, and to check if the bottom contexts on the stack are the same.

21 This account can perhaps be extended to intensional question-embedding verbs such as ‘wonder’, if our private knowledge/belief/etc. states are allowed to contain issues (be partitioned). On this view, wondering is like asking a question ‘internally’, where asking is the act of rendering a context into a particular inquisitive state, and an ‘ask’ sentence reports this of some other context. One thing is crucial: if epistemic states can be partitioned, then the account above of ‘know’ will not work, as it will predict that ‘know’ and ‘wonder’ will have essentially the same semantics. Extensional verbs seem to involve a notion of not only support but also uninquisitive support, whereas intensional verbs involve inquisitive support. This is where the more classical distinction between selecting for the extension and intension of a question would be cashed out in this system.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The CCP given in (73) can handle plain questions and assertions straightforwardly. In prose, someone ‘knows whether /’ just in case internally ‘asking’ / would be redundant to their private epistemic state. Similarly, if we knew what they know, asking ‘whether /’ would be redundant. At this point, we have an account of interrogative complements of extensional question-embedding verbs, assuming we can find some suitable accessibility function/modal base to check the support relation with.21 Embedded CQs take more work.

308 Conditional Questions (74) For any stack s, let s? be sjs–1j (the bottom/tail of the stack). (75) Support (stack version) For any sentences / and w: / s–supports w iff "s 2 S : ðs + /Þ? ¼ ðs + / + wÞ?

8.3 Updating ‘know’ for the stack model We are now in a position to make some trivial changes to the Aloni and van Rooy-style CCP of ‘know’ from (73), in order to handle CQs. The interpretation of ‘know’-sentences is defined in terms of update on contexts. To get the macro-CCP, a (force-less) ‘know’ sentence is combined with a speech–act operator just as any force-less sentence would be (see section 5.3). The final CCP for ‘know’ sentences is given in (76). 22 The following technical definitions spell this out. The variables s and s# are arbitrary stacks, and / and w are sentences.

(i) supp-worldsðsÞ ¼def

w 2 W j di 2 N :

^

dw# 2 W : :dw$ 2 W :

ðÆw; w#æ 2 si _ Æw#; wæ 2 si Þ ðÆw; w$æ 2 s0 _ Æw$; wæ 2 s0 Þ

(ii) supp-match(s, s#) ¼def {Æw1, w2æ 2 s0 j w1 ; supp-worlds(s#) ^ w2 ; supp-worlds(s#)} (iii) s ¼stack s# iff supp-match(s, s#) ¼ s#0 (iv) Support (supposition-matching version) / s-supports w iff ½"s : s 2 S s+/ ¼stack s+/+w This version is probably not general enough to account for cases where / introduces suppositions (cases that are not relevant to the use s-support is put here).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This new notion does exactly the same thing as the old one, i.e. we check to see if new information is added by w on top of /. It also checks if new issues are raised by w. If / resolves the issue that w would raise, the comparison comes out true. What is important is that the comparison abstracts away from all the temporary assumptions made in the discourse by either / or w. It does so by looking at the bottom element of the stack. If issues did not percolate immediately, such a straightforward treatment of supporting would not be available. We would still need a mechanism to abstract away from temporary assumptions in order to check the results of w on the output of /. Such a notion is technically possible, but it is complicated. It would basically import any temporary assumptions made by / into the context that is the output of w. The idea is that someone ‘knows whether / if w ’ when, if we first knew what they knew and then supposed w, / would follow.22 In the simpler treatment sketched above, we avoid this complexity by allowing issues to percolate. As a result, the bottom element of the macro-context represents the results of importing the suppositions already.

James Isaacs and Kyle Rawlins 309

(76) CCP for ‘know’ sentences (ﬁnal version) For any individual-denoting DP a, declarative or interrogative clause /, and context c: ( ) Kw 1 ð½½aw1 ;c Þs-supports / c4½a½knows /¼def Æw1 ; w2 æ 2 c w ;c ^ Kw 2 ð½½a 2 Þs-supports /

(77)

w1 w2 w3 w4 w5 w6 w7 w8

Alfonso goes

Joanna goes

A & J fight

+ + + + – – – –

+ + – – + + – –

+ – + – + – + –

23 It appears that only one ‘if ’-clause can be easily left adjoined, and only one can be easily right adjoined.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Someone ‘knows whether / if w ’ when their knowledge supports the question ‘whether /’. More generally, someone ‘knows /’ when their knowledge supports /. The semantics is not defined on any properties (i.e. the presence of a conditional) of the complement of ‘know’—so any arbitrary assumptions introduced in any way could be handled. Thus, this analysis extends readily to embedded double conditionals, to the extent that they are syntactically well formed.23 The analysis retains the property from previous treatments of embedded questions of straightforwardly handling conjoined questions and assertions, in addition to conjoined CQs and assertions, conjoined questions and conditional assertions, and so on. The idea behind this analysis might be clearer with an example. Ferdinand, who is close to being omniscient, knows at least that we are in one of eight worlds, w1, . . ., w8. Suppose that a party has to be planned and that Joanna is invited. Suppose that Alfonso has also been invited, and may or may not attend. In this scenario, Joanna and Alfonso are sometimes on bad terms, and sometimes on good terms. The full range of possibilities for these eight worlds is shown in (77).

310 Conditional Questions Let us further assume that Ferdinand also knows that if Alfonso and Joanna are on bad terms, there is no way that both of them would be at the party. That is, he knows that w1 and w4 are out from the start. He also knows that if they are on good terms, Joanna will attend the party of Alfonso does. Consider first the case where Ferdinand happens to learn that they are fighting. He now knows that we must be in one of w3, w5, w7. (Kw(ferd) ¼ {w3, w5, w7} for w in the context set.) In this scenario, it seems entirely true to say:

Intuitively, this should be true because Ferdinand’s knowledge will make unnecessary the question of whether Joanna will come to the party if Alfonso does, and this is what the present analysis captures. The interpretation of (78) involves an update which reduces the context set to pairs of worlds where Ferdinand’s knowledge state in both worlds ssupports the embedded CQ. To find out if his knowledge state ssupports the embedded CQ, we have to test the following (pseudo)formula: s + Ferdinand#s knowledge state ð79Þ ½"s : s 2 S ¼ ðs + Ferdinand#s knowledge state + Whether Joanna will go if Alfonso goesÞ

!

The updates can be calculated straightforwardly—on the left side of the equal sign we get the intersection of the (top element of) the stack s with Ferdinand’s knowledge state, and on the right side we get the result of updating the stack containing Ferdinand’s knowledge with the CQ, just as if it were a matrix CQ. This is done for arbitrary stacks, but let us focus on a case where the arbitrary stack does not tell us much and has only one level. The result of adding Alfonso’s knowledge state in this scenario is shown here:

(80)

The antecedent of the conditional causes Alfonso to set aside worlds where Alfonso does not go. This results in a stack s#, with a simplified s#. 0 Stack s# 1 is identical to the s0 above.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(78) Ferdinand knows whether Joanna will go to the party if Alfonso goes.

James Isaacs and Kyle Rawlins 311

(81)

(82)

Just as before, we then accept the content of the antecedent of the conditional, that Alfonso is going. This results in a stack t# with t#0 identical to t0. (83) In this case, updating with the consequent question does have an effect. It causes us to disconnect the worlds w2 and w3. This disconnection must then percolate immediately and will have an effect as well on the tail of the resulting stack t$.

(84)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Next, we update this stack with the consequent, partitioning depending on whether Joanna attends to the party. This has no effect on the context set containing just w3 (i.e. s#). 0 Correspondingly, the update will have no effect on the tail of the stack, s#, 1 since there is no issue to percolate. Thus, the tails of s and s# are identical, and the sentence comes out true. Consider now the case where Ferdinand does not learn whether Alfonso and Joanna are fighting. In this version of Ferdinand’s knowledge state, worlds w2, w4 and w6 have not been excluded. The result of adding Alfonso’s knowledge state to some arbitrary stack will look like t in (82):

312 Conditional Questions

9 DYNAMIC SEMANTICS AND DOMAIN RESTRICTION Viewed from an abstract perspective, our analysis of CQs is really an analysis of domain restriction in questions and answers, and the stack semantics of Kaufmann (2000) is really a tool for modelling temporary but persistent domain restrictions. Much of our analysis boils down to the claim that in CQs, the ‘if ’-clause restricts the domain of the question operator. More generally, we propose that an ‘if ’-clause restricts the domain of any speech–act operator that is present in a clause that it adjoins to. This claim could be cast in a variety of ways and is not at all contingent on using a dynamic semantics. However, there is some reason to choose a dynamic account over a static account. The domain restrictions in question are temporary, but are clearly seen to persist past the end of a single sentence. (85) a. If Alfonso comes to the party, will Joanna go home angry? b. Yes, she will. The antecedent of the conditional in (85a) clearly provides a domain restriction not only for the consequent of the conditional but also for a following answer uttered by an entirely different speaker. Despite the temporariness of the restriction, it can potentially persist indefinitely— in our analysis it is exactly because the issue raised by the CQ in (85a) is still on the table that the temporary context cannot be discarded. A dynamic approach, using stacks to model temporary domain restrictions, directly provides the tools to analyse these effects. A static account of domain restriction struggles, because the effects are clearly discourselevel effects, not just sentence-level effects, and are clearly tied to

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The tail of t$ is clearly not identical to the tail of t, and so the sentence comes out false. Note that the structure in t$1 is not a partition on worlds w1, . . ., w8, though it does contain a Velissaratou-style alternative structure. While it is possible to implement a version of this analysis that involves partition structures all the way down (as per section 7), using the simple notion of inquisitive updates (see (48), and fn. 22), it is somewhat complicated to do in a general way. In this example, it would involve taking any suppositions introduced between t and t$, and backing them out in t so that the top element of t$ (containing the new issue) is comparable with the top element in t. In the following section, we turn to a brief discussion of how our analysis meshes with the standard analysis of modals.

James Isaacs and Kyle Rawlins 313

(86)

This analysis provides a welcome result for conditionals with no overt modal. Kratzer (1986, and earlier work by Kratzer) argues that a covert modal must be present in any indicative conditional that lacks an overt modal. In our analysis, something very similar, but not quite the same, happens. Restricting the (covert) speech–act operator does all the work that restricting a covert modal would do, and a speech–act operator would need to be present in any case. One could think of speech–act operators as a kind of covert modal with an epistemic flavour, in fact—to assert / is to reduce the context set to a state where / is necessarily true given that context set. To question / (considering a polar question) is to partition the context into cells where either / is necessarily true or it necessarily does not follow. More generally, a question induces a partition into cells where in each cell, some complete answer is necessarily true.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

discourse-level dynamics. We contend that at least for this sort of domain restriction, a dynamic account provides the best model. The claim that ‘if ’-clauses restrict the domain of a speech–act operator can be reconciled with the standard (linguist’s) analysis of conditionals, due to Lewis and Kratzer (see Lewis 1975; Kratzer 1977, 1981, 1986, 1991). According to the standard analysis, an ‘if ’-clause restricts the domain of an operator in the clause it adjoins to. In the case of the examples under discussion here, that operator would appear to be a modal. The resolution is straightforward. In our analysis, every full clause contains a speech–act operator (or, alternatively, has the interpretive force of some speech act), regardless of whether there is a modal in the sentence. If a clause has an adjoined ‘if ’-clause, the ‘if ’-clause restricts whatever speech–act operator is present. If a modal is also present, it will receive its domain restriction from the speech-act operator. If the speech–act operator has been restricted by an ‘if ’-clause, then the modal’s domain will be indirectly but straightforwardly restricted by that same ‘if ’-clause. This is illustrated schematically in (86).

314 Conditional Questions

(87) [[must]]w,c ¼ kp 2 DÆstæ . "w# 2 Ds[w# 2 \ fc(w) ^ w# 2 csc] p(w#) This denotation leaves out reference to an ordering source, but it could be added in straightforwardly. The denotation in (87) is really only trivially different from a Kratzer-style modal. In place of ‘w# 2 csc’, the Kratzer-style modal would have w# 2 p#, where p# is the proposition denoted by the content of the ‘if ’-clause. It is clear that there may be much more to examine, in the interaction of a modal with a CQ. Our suggestion for the means of interaction between ‘if ’-clauses, modals and question operators is only a tentative one. For instance, Cohen (2006) argues (based on data we have not considered here, involving conjoined consequents) that the logical structure actually places the modal outside the question operator. We will leave for the future determining the best way to go.

10 CONCLUDING REMARKS In our analysis, CQs involve the straightforward combination of a dynamic semantics for conditionals and a partition semantics of questions. The interpretation of the antecedent and the consequent is decomposed into two steps. In the first step, a temporary copy of the current context is made, and is updated to entail the propositional content of the antecedent. The question in the consequent is then taken to raise an issue relative to the temporary context. This happens in a way exactly parallel to modal subordination. The resulting analysis is technically complex, but it is complex in an expected way—it arises from the combination of independently motivated analyses. The semantic analysis we provide works in tandem with a pragmatic analysis of responses which address the content of the antecedent. In

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We now sketch what the semantics of an overt modal looks like under these assumptions. Modal words, in Kratzer’s system, are sensitive to two contextual parameters: a modal base and an ordering source. We abstract away from the reasons for using an ordering source in our discussion, but our analysis is entirely compatible with it. The modal base corresponds to the accessibility function of standard modal logic, and the ‘if ’-clause provides a restrictor to it. We derive the restriction indirectly. The modal base provided by the context is further restricted by the context set. Here is a denotation for ‘must’, assuming that fc is the modal base provided by the context, and that csc is the (onedimensional) context set provided by the context (for use in static interpretation below the clause level).

James Isaacs and Kyle Rawlins 315

(88) If you go to the store, buy some milk. (89) If the sun comes out, what a nice day it will be!

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

previous analyses, responses that deny the truth of the antecedent of indicative CQs have been incorporated into the semantics, because they seem to dispel the issues raised by the question. We argue that this issue-dispelling effect arises because such responses deny the presuppositions of the conditional structure. This pragmatic analysis allows us to maintain a standard partition semantics of questions and also provide an account which can generalize to counterfactual CQs. We have presented two ways of retaining mutual exclusivity: one that retains it for all contexts and one only for the context most immediately affected by a question. We showed that adopting the second assumption simplifies the analysis of embedded CQs. However, the first assumption does not make it technically impossible to analyse embedded CQs. The choice between these alternatives does not yet appear to make any empirical predictions, and one question for the future is whether it does. A recently revived debate in the semantics of declarative conditionals revolves around the scope a speech–act operator takes relative to an ‘if ’-clause (see Lycan 2006; Stalnaker 2005 for recent discussion). On one view, the Asserted Conditional Theory, the utterance of a conditional with a declarative consequent amounts to the assertion of the entire conditional. On the other, the Conditionalized Assertion Theory, the utterance of the consequent alone counts as the assertion. The antecedent simply provides information about when the assertion obtains. Historically, this debate has focused almost exclusively on conditionals with declarative consequents. Stalnaker (2005) argues that, for the case of assertions, the two approaches achieve the same results. When considering our analysis of CQs, the picture changes. Our analysis crucially involves conditionalized questions—this allows the ‘if ’-clause to prepare a restricted context for the question, and the denial of the antecedent to deny a presupposition prior to the question. Cohen (2006) also argues against the questioned conditional approach. The general implication is that ‘if ’-clauses might always be scopally higher than speech–act operators (following similar proposals about embedded speech–act operators in Krifka 2001, 2004). This suggests not only that conditional exclamations and commands should be possible but also that there is nothing special about the fact that they are conditionalized. Both are indeed possible:

316 Conditional Questions We predict that the ‘if ’-clause in each case should simply restrict an imperative and exclamative operator, respectively. This hypothesis provides a broad scope for future investigation. Acknowledgements

JAMES ISAACS Department of Linguistics University of California Santa Cruz 1156 High Street Santa Cruz CA 95064-1077 USA e-mail: [email protected]. KYLE RAWLINS Department of Linguistics University of California Santa Cruz 1156 High Street Santa Cruz CA 95064-1077 USA e-mail: [email protected].

REFERENCES Adams, E. (1965), ‘The logic of conditionals’. Inquiry 8:166–97. Aloni, M. & R. van Rooy (2002), ‘The dynamics of questions and focus’. In B. Jackson (ed.), Proceedings of SALT 12. Cornell University, CLC Publications. Ithaca, NY, 20–39. Bittner, M. (2001), ‘Surface composition as bridging’. Journal of Semantics 18: 127–77. Cohen, S. (2006), Conditional Interrogatives as Expressing Conditionalized Questions. Unpublished MS, University of Massachusetts, Amherst.

Farkas, D. (2003), ‘Assertion, belief and mood choice’. Presented at ESSLLI, Conditional and Unconditional Modality Workshop, Vienna. Farkas, D. & M. Ippolito (2005), Discourse Subordination. Unpublished MS, University of California Santa Cruz and University of Toronto. Frank, A. (1996), Context Dependence in Modal Constructions. Ph.D. thesis, Institut fu¨r maschinelle Sprachverarbeitung, Stuttgart. von Fintel, K. (1999), ‘The presupposition of subjunctive conditionals’. In

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We would like to thank Cleo Condoravdi, Donka Farkas, Hans-Martin Ga¨rtner, Jim Higginbotham, Michela Ippolito, Ruth Kramer, Bill Ladusaw, Frank Veltman and the editors, as well as audiences at University of California Santa Cruz, the Language Under Uncertainty workshop at Kyoto University, ZAS, the University of Frankfurt and S-Trend 2006 for discussion and comments on various stages of this work. We are especially grateful to Jeroen Groenendijk for extensive written comments on earlier versions.

James Isaacs and Kyle Rawlins 317 (eds.), The View from Building 20. The MIT Press. Cambridge, MA. Higginbotham, J. & R. May (1981), ‘Questions, quantifiers, and crossing’. Linguistic Review 1:41–79. Hintikka, J. (1962), Knowledge and Belief: An Introduction to the Logic of Two Notions. Cornell University Press. Ithaca, NY. Hofstadter, D. (1979), Go¨del, Escher, Bach: An Eternal Golden Braid. Basic Books. New York. Hulstijn, J. (1997), ‘Structured information states. Raising and resolving issues’. In A. Benz and G. Ja¨ger (eds.), Proceedings of MunDial97. University of Munich. Munich. Iatridou, S. (1991), Topics in Conditionals. Ph.D. dissertation, MIT. Cambridge, MA. Karttunen, L. (1973), ‘Presuppositions of compound sentences’. Linguistic Inquiry 4:167–93. Karttunen, L. (1974), ‘Presuppositions and linguistic context’. Theoretical Linguistics 1:181–93. Karttunen, L. (1977), ‘Syntax and semantics of questions’. Linguistics and Philosophy 1:3–44. Kaufmann, S. (2000), ‘Dynamic context management’. In M. Faller, S. Kaufmann, and M. Pauly (eds.), Formalizing the Dynamics of Information. CSLI. Stanford, CA. Kratzer, A. (1977), ’What ‘must’ and ‘can’ must and can mean’. Linguistics and Philosophy 1:337–55. Kratzer, A. (1981), ‘The notional category of modality’. In H.-J. Eikmeyer and H. Rieser (eds.), Words, Worlds, and Contexts: New Approaches in World Semantics. Walter de Gruyter. Berlin. 38–74. Kratzer, A. (1986), ‘Conditionals’. In Proceedings of CLS 22. University of Chicago Press, 1–15.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

U. Sauerland and O. Percus (eds.), The Interpretive Tract. MIT Working Papers in Linguistics 25. MITWPL. Cambridge, MA. 29–44. von Fintel, K. (2001), ‘Counterfactuals in a dynamic context’. In M. Kenstowicz (ed.), Ken Hale: A Life in Language. MIT Press. Cambridge, MA. 123–152. Ga¨rdenfors, P. (1986), ‘Belief revision and the ramsey test for conditionals’. Philosophical Review 95:81–93. Groenendijk, J. (1999), ‘The logic of interrogation’. In T. Matthews and D. L. Strolovitch (eds.), Proceedings of SALT IX. CLC Publications. Ithaca, NY. Groenendijk, J. & M. Stokhof (1984), Studies in the Semantics of Questions and the Pragmatics of Answers. Ph.D. dissertation, University of Amsterdam, Amsterdam. Groenendijk, J. & M. Stokhof (1997), ‘Questions’. In J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language. Elsevier. Amsterdam. 1055–124. Hamblin, C. L. (1958), ‘Questions’. Australasian Journal of Philosophy 36: 159–68. Hamblin, C. L. (1973), ‘Questions in Montague English’. Foundations of Language 10:41–53. Heim, I. (1982), The Semantics of Deﬁnite and Indeﬁnite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Heim, I. (1983), ‘On the projection problem for presuppositions’. In M. Barlow, D. Flickinger, and M. Wescoat (eds.), WCCFL 2: Second Annual West Coast Conference on Formal Linguistics. CSLI. Stanford, CA. 114–25. Heim, I. (1992), ‘Presupposition projection and the semantics of attitude verbs’. Journal of Semantics 9:183–221. Higginbotham, J. (1993), ‘Interrogatives’. In K. Hale and S. Keyser

318 Conditional Questions Muskens, R. (1996), ‘Combining Montague semantics and discourse representation’. Linguistics and Philosophy 19:143–86. Muskens, R., J. van Benthem, & A. Visser (1997), ‘Dynamics’. In J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language. Elsevier. Cambridge, MA. 587–648. Ramsey, F. P. (1931), ‘General propositions and causality’. In R. Braithwaite (ed.), The Foundations of Mathematics: Collected Papers of Frank P. Ramsey. Routledge. London. 237–55. Roberts, C. (1989), ‘Modal subordination and pronominal anaphora in discourse’. Linguistics and Philosophy 12:683–721. Stalnaker, R. (1968), ‘A theory of conditionals’. In N. Resher (ed.), Studies in Logical Theory. Blackwell. Oxford. Stalnaker, R. (1978), ‘Assertion’. In P. Cole (ed.), Pragmatics. Academic Press. New York. 315–32. Stalnaker, R. (2005), ‘Conditional propositions and conditional assertion’. Paper presented at the 2005 LSA Summer Institute Formal Pragmatics Workshop. Velissaratou, S. (2000), ‘Conditional questions and which-interrogatives’. Master of Logic thesis, University of Amsterdam, ILLC Publications. Veltman, F. (2005), ‘Making counterfactual assumptions’. Journal of Semantics 22:159–80. Zaefferer, D. (1990), ‘Conditionals and unconditionals in universal grammar and situation semantics’. In R. Cooper, K. Mukai, and J. Perry (eds.), Situation Theory and its Applications I. CSLI. Stanford, CA. 471–92. Zaefferer, D. (1991), ‘Conditionals and unconditionals: cross-linguistic and logical aspects’. In D. Zaefferer (ed.),

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Kratzer, A. (1991), ‘Modality’. In A. von Stechow and D. Wunderlich (eds.), Semantics: An International Handbook of Contemporary Research. Mouton de Gruyter. Berlin. 639–50. Kratzer, A. & J. Shimoyama (2002), ‘Indeterminate pronouns: the view from Japanese’. In Y. Otsu (ed.), Proceedings of the 3rd Tokyo conference on psycholinguistics. Hituzi Syobo. Tokyo. 1–25. Krifka, M. (1999), ‘Quantifying into question acts’. In T. Matthews and D. Strolovitch (eds.), Proceedings of SALT IX. CLC Publications. Krifka, M. (2001), ‘Quantifying into question acts’. Natural Language Semantics 9:1–40. Krifka, M. (2004), ‘Semantics below and above speech acts’. Talk handout, 9 April 2004. Stanford University. Lewis, D. (1973), Counterfactuals. Blackwell. Oxford. Lewis, D. (1975), ‘Adverbs of quantification’. In E. L. Keenan (ed.), Formal Semantics of Natural Language. Cambridge University Press. Cambridge, UK. Lycan, W. (2006), ‘Conditional-assertion theories of conditionals’. In J. Thomson and A. Byrne (eds.), Content and Modality: Themes from the Philosophy of Robert Stalnaker. Oxford University Press. Oxford, 148–63. Mackie, J. L. (1973), Truth, Probability and Paradox. Clarendon Press. Oxford. McCloskey, J. (2006), ‘Questions and questioning in a local English’. In E. H. Raffaella Zanuttini, Hector Campos, and P. Portner (eds.), CrossLinguistic Research in Syntax and Semantics: Negation, Tense and Clausal Architecture. Georgetown University Press. Washington DC, 87–126. Merchant, J. (2004), ‘Fragments and ellipsis’. Linguistics and Philosophy 27:661–738.

James Isaacs and Kyle Rawlins 319 Semantic Universals and Universal Semantics. Foris Publications. Berlin, 210–36. Zeevat, H. (1992), ‘Presupposition and accommodation in update semantics’. Journal of Semantics 9:379–412.

Zeinstra, L. (1990), ‘Reasoning as discourse’. Master’s thesis, Department of Philosophy, University of Utrecht. First version received: 01.11.2006 Second version received: 02.11.2006 Accepted: 04.08.2007

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Journal of Semantics 25: 321–344 doi:10.1093/jos/ffn002 Advance Access publication June 3, 2008

Branching from Inertia Worlds TIM FERNANDO Trinity College Dublin

Abstract

1 INTRODUCTION Sentences such as (1) illustrate the observation in Dowty (1979) that an event in progress at time I in world w may fail to culminate in w. (1) Pat was losing when the match took an unexpected turn. Dowty employed the notion of an (I, w)-inertia world to insure that an event e in progress at I in w culminates somewhere—namely, in every (I, w)-inertia world. We may expect e to culminate in w inasmuch as we may expect w to be an (I, w)-inertia world. As there is no law making w an (I, w)-inertia world, however, e need not culminate at w. But then can we assume (I, w)-inertia worlds exist? Surely we may assume that there are events other than e in progress at I in w. Can we be certain not one of them clashes with e down the line? Landman (1992) credits the following example to Roger Schwarzschild. Suppose I was on a plane to Boston which got hijacked and landed in Bismarck, North Dakota. What was going on before the plane was hijacked? One thing I can say is: ‘I was flying to Boston when the plane was hijacked.’ This is reasonable. But another thing I could say is: ‘I was flying to Boston. Well, in fact, I wasn’t, I was flying to Bismarck, North Dakota, but I didn’t know that at the time.’ And this is also reasonable. (pp. 30–1) With this (or perhaps a different example) in mind, let us suppose another event e# were to be in progress at I in w that culminates only in a world where e does not. Then, the requirement that e and e# culminate at every (I, w)-inertia world would mean there is no (I, w)inertia world (rendering talk of inertia worlds pointless). This suggests The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected].

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The notion of inertia is explicated in terms of forces recorded in snapshots that are strung together to represent events. The role inertia worlds were conceived to serve in the semantics of the progressive is assumed by a branching construct that specifies what may follow, apart from what follows.

322 Branching from Inertia Worlds

(i) the notion of inertia is fleshed out against fluents (as opposed to worlds), some of which are assumed to be inertial while others describe forces, and (ii) a world is formed from the events that happen in it where an event is formulated as a string of sets of fluents (Fernando 2004). The analysis of inertia and worlds in (i)–(ii) stops short of addressing modal matters such as the implication (2b) of a natural reading of (2a). (2) a. Pat stopped the car before it hit the tree. b. The car did not hit the tree but it might well have. To interpret (2b), we need structures recording what may happen, over and above what does happen. Accordingly, strings are extended to branching strings, which are closed under not only concatenation but also a second binary operation. That second operation yields branches recording some of the ways things may turn out under historical necessity (e.g. Thomason 1984). Branching is, I claim, implicit in the step from a world w to the set of (I, w)-inertia worlds. What’s more, it is branching, and not some notion of inertia world, that figures in the semantics of the progressive formalized below. To motivate that formalization, section 2 examines entailments ‘ of progressive forms that are associated with the appropriateness of temporal modification by for, as opposed to in (Vendler 1967). (3) a. Pat was walking ‘ Pat walked b. Pat walked for (?in) five minutes. (4) a. Pat was walking home 0 Pat walked home b. Pat walked home in (?for) five minutes. As for inertia, section 3 investigates its role in the contrast between the simple past and the present perfect (e.g. Steedman 2000).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

refining the relativization (I, w) on inertia worlds to discriminate between events e and e# that are headed for a conflict. The alternative is to deny that such events can be in progress at the same pair (I, w). Stepping back, we might ask what (on earth) an inertia world is. Are our intuitions about inertia worlds coherent and reliable enough that we can be comfortable with a semantic account that treats inertia worlds as primitive? Rather than take inertia worlds for granted, the present work pursues a constructive approach, building worlds bottom-up from temporal propositions. I shall refer to these propositions as ﬂuents, following the custom in artificial intelligence since McCarthy & Hayes (1969) and more recently in linguistic semantics (van Lambalgen & Hamm 2004). Very briefly,

Tim Fernando 323

(5) a. Pat left Dublin but is back (in Dublin). b. ?Pat has left Dublin but is back (in Dublin).

def p4q ðas sets of worldsÞ, p‘q 5 we refine 4 to a subsumption relation V (defined in the next section) between sets of strings. The refinement needed here is the key to the analysis of branching in section 4.

2 EVENT TYPES AS SETS OF STRINGS Underlying our representation of events as strings is the intuition that an event is ‘a series of snapshots’ (Tenny 1987). Equating a snapshot with a set of fluents that describe it, we represent an event as a string a1 a2 . . . an of sets ai of fluents describing n successive moments. We finesse questions about the choice of successive moments and string length n by working with a set of strings. For instance, we might associate rain from dawn to dusk with the set ½rain; dawn½rain*½rain; dusk ¼ f½rain; dawn½rainn ½rain; duskjn > 0g of strings [rain, dawn][rain]n[rain, dusk] with (n + 2) snapshots, all of rain, the first at dawn and the last at dusk. The different values of n correspond to different levels of temporal granularity (the larger the n, the finer the grain).1 We use square brackets [ and ], instead of curly 1

Whether or not these infinitely many strings should be thought of as distinct events depends on how we interpret the temporal relation between successive snapshots. We will sidestep this question by focusing less on the conception of an event-as-string and more on that of an event-type-as-string-set.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Inertial laws are laid down specifying the persistence of inertial fluents in the absence of forces. We work with events as in Parsons (1990) and Landman (1992), and follow the latter in looking at what might become of an event in progress, for a glimpse at modal implications of the progressive related to (2). That is, we explore ‘continuations branches’ as in Landman’s semantics of the progressive, but without appealing to ‘a Stalnaker-style theory of subjunctives’ or (for that matter) worlds. Instead, continuation branches are construed perspectivally on the basis of the structure on events induced by their string representations. Entailments can be read directly off strings, making it unnecessary to resort to worlds to define entailments in terms of truth at worlds. Rather than sets of worlds as propositions, we study sets of strings as event types. And rather than defining entailment ‘ between propositions p and q by the subset relation 4

324 Branching from Inertia Worlds

ncðUÞ

def

fa 2 PowðUÞjð"u 2 aÞ u;ag

of non-contradictory snapshots (assembled from U). Nevertheless, we will define operations and relations on languages over the alphabet Pow(U), as it is easy enough to intersect a language L 4 Pow(U)* with nc(U)* (and indeed other constraints) to weed out spurious descriptions. A natural conjunction of languages L, L# over Pow(U) is the superposition L & L# of L and L# obtained from the componentwise union of strings in L and L# of the same length L & L#

def

[ fða1 [ b1 Þ . . . ðan [ bn Þja1 . . . an 2 L and b1 . . . bn 2 L#g

n>0

(Fernando 2004). For example, we have (6). (6) ½rain; dawn½rain* ½rain; dusk ¼ ½rain+ &½dawnh* ½dusk ¼ ½rain+ & ð½dawnh+ & h+ ½duskÞ To capture the growth of information from &, let us say that L subsumes L# and write L V L# if the superposition of L and L# includes L L V L#

def

5

L 4 L&L#

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

braces, to enclose a set of fluents when it is intended as a snapshot. Boxes are arguably preferable to square brackets, but are a nuisance to typeset. Hence, we will refrain from drawing boxes except for the empty snapshot, which we write h instead of []. Beyond reinforcing the filmstrip metaphor, this helps distinguish the string h of length 1 (containing no fluents) from the empty string set ; (containing no strings). We refer to a string set as a language (as in formal language theory), and write * for Kleene star, L+ for LL* and + for nondeterministic choice. We collect the fluents in a set U, turning snapshots into elements of the powerset Pow(U) that serves as our alphabet. Snapshots here are (unlike McCarthy & Hayes 1969) partial: is the negation of u. h is as much a part of [u] as it is of [ u], where u Henceforth, we assume that every fluent u comes with another fluent ¼ u. A snapshot may be u-underdefined in that it 6¼ u and that u u (e.g. h), or else it may be u-overdefined, contains neither u nor u . Whereas u-underdefinedness simply reflects containing both u and u the partiality of observations, u-overdefinedness indicates that the observation has gone awry. This suggests restricting the alphabet Pow(U) to the family

Tim Fernando 325

(roughly: L is at least as informative as L#). Conflating a string s with the singleton language {s}, we get (7) and (8). (7) states that V holds between strings of the same length related componentwise by inclusion. (8) says L subsumes L# exactly if each string in L subsumes some string in L#. (7)

a1 . . . an V b1 . . . bm iff n ¼ m and ai bi for 1 < i < n

(8)

L V L# iff ð"s 2 LÞðds# 2 L#Þs V s#

As (9) illustrates, we may regard a comma inside a box as conjunction, and plus between strings as disjunction. ½u; w V ½u V ½u + ½w

That is, as a type with instances s 2 L, a language L is essentially a disjunction ¤s2Ls of conjunctions s (Fernando 2004). Subsumption V and superposition & go hand in hand in specifying languages LðAÞ for English sentences A such as those in (3) and (4), repeated here. (3) a. Pat was walking ‘ Pat walked b. Pat walked for (?in) five minutes. (4) a. Pat was walking home 0 Pat walked home b. Pat walked home in (?for) five minutes. The general idea is to reduce entailment to subsumption and trace the unacceptability of A to the disjointness of LðAÞ from nc(U)*. (Note that a language L is disjoint from nc(U)* iff every string a1 . . . an in L belong to is contradictory in that for some u e U, both u and u some ai.) (10) a. A ‘ B because L(A)VL(B) b. ?A because LðAÞ \ ncðUÞ* ¼ ; Leaving inflection from tense and aspect out for the moment, we assume a fluent home(p) (saying Pat is home) is negated in the language L(Pat walk home) until the very end. +

(11) LðPat walk homeÞ V ½homeðpÞ ½homeðpÞ (11) says LðPat walk homeÞ is home(p)-telic, where we call a language L u-telic if L V ½ u* ½u.2 More generally, we collect fluents that 2 This analysis makes sense only if we assume some bounded temporal granularity, with discernibly different increments. A discrete as opposed to continuous conceptualization of change saves us from pinning the precise moment of change (or measuring the duration of a state/event) to an arbitrary (arguably meaningless) degree of precision.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(9)

326 Branching from Inertia Worlds appear at the end of every L-string in the set xL of fluents u such that LVh* ½u xL

def

½u jLVh* ½u

and define L to be telic if LVxL * h, where the negation a4PowðUÞ of a 4 U is (in accordance with De Morgan) the disjunction (given by +) of negations of fluents in a ½u1 ; . . . ; un

def

½u1 + + ½un

def

(12) Lðfive minutesÞ V ½0ðsÞh+ ½5minðsÞ Then, we can put down the contrast between (3b) and (4b) to (13), and an interpretation of temporal in- and for-modification by the superpositions in (14) with a language I representing the temporal interval. (3) b. Pat walked for (?in) five minutes. (4) b. Pat walked home in (?for) five minutes. (13) a. LðPat walk homeÞ is telic. b. LðPat walkÞ is iterative. (14) a. inðL; IÞ b. forðL; IÞ

def def

L & I & xL * h L & I & hxL*

According to (14), in contributes xL * h, whereas for contributes hx*L . Let us say L is durative if L V hhh+ (that is, every string in L has length >3), and let us write [ for the intersection of V with its converse L [ L#

def

5

LVL# and L#VL:

Assuming I is durative, we derive (15) and (16), matching telic and iterative languages with temporal in- and for-modification, respectively. (15) If L is telic then inðL; IÞ [ L & I and forðL; IÞ \ ncðUÞ* ¼ ;. (16) If L is iterative then forðL; IÞ [ L & I and inðL; IÞ \ ncðUÞ* ¼ ;.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(with h ;). Clearly, if L is u-telic then L is telic. As for LðPat walkÞ, let us call L iterative if LVhx*L . Now, suppose we analyse the interval ﬁve minutes as in (12), with fluents 0(s) and 5min(s) saying that zero time and five minutes have (respectively) passed since time s.

Tim Fernando 327

Next, for tense and aspect, we adopt a Reichenbachian approach that locates the event time E relative to not only a speech time S but also a reference time R (Reichenbach 1947). To see how R might be useful, consider the pair in (5a,b). (5) a. Pat left Dublin but is back (in Dublin). b. ?Pat has left Dublin but is back (in Dublin). The position of the speech time S relative to Pat’s departure from Dublin is not enough to differentiate (5a) from (5b); abbreviating LðPat leave DublinÞ to L0,

R allows us to distinguish LðPat left DublinÞ from LðPat has left DublinÞ, coinciding with the end of L0 in the former case, and with S in the latter. (17) a. LðPat left DublinÞVðL0 & h* ½RÞh* ½S b. LðPat has left DublinÞVL0 h* ½R; S Treating R and S as fluents (in U) while fleshing out E as a language (e.g. L0), we derive the contrast in (5) from (17) in the next section. In general, we apply aspectual and tense operators in sequence, forming, for instance, LðPat left DublinÞ ¼ PastðSimpðL0 ÞÞ LðPat has left DublinÞ ¼ PresðPerfðL0 ÞÞ with tense operators for the past and present applied after operators for simple and perfect aspect. If an event with temporal projection E is represented by a language L, we get three aspectual operators on L. def

L & h* ½R (18) a. SimpðLÞ def b. Progo ðLÞ L & h+ ½Rh+ def c. Perfo ðLÞ L h* ½R

ði:e: R is E0 s right endÞ ði:e: R contained in EÞ ði:e: E < RÞ

As a position from which to view the event represented by L, R says in the case of SIMP(L), that the event has reached completion; in the case of PROGo(L), that it has not quite gotten there yet but is on its way; and in PERFo(L), that it is history. The subscript o on PERF and on PROG marks the need for additional operations that concern inertia and branching, respectively. We will attend to these complications in sections 3 and 4. For now, we work with truncated forms given by the following definitions. A string a1 . . . an is R-truncated if R; [1
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

LðPat left DublinÞVL0 h* ½S and LðPat has left DublinÞVL0 h* ½S:

328 Branching from Inertia Worlds (Thus, h[R] is R-truncated, whereas [R]h is not. Also, any string where R does not occur is R-truncated.) The R-truncation sR of a string s is the longest R-truncated prefix of s; that is, ða1 . . . an ÞR

def

a1 . . . ak

k

where

def

maxfj < njR;ai for 1 < i < jg:

(Thus, ([R]h)R ¼ [R].) Given a language L, let LR be the set {sR j s 2 L} of its R-truncations, and let us say L is R-truncated if all its strings are—that is, L ¼ LR. SIMP(L) and PERFo(L) are R-truncated, assuming R does not occur in (L). Not so for PROGo(L), and accordingly, we define def

ðProgo ðLÞÞR

(corresponding to the truncated progressive in Parsons (1990), as opposed to that in Landman (1992) which we reformulate in section 4). We can now take up (3a) and (4a) via the approximations (19) and (20), respectively, suggested by (10a). (3) a. Pat was walking ‘ Pat walked (4) b. Pat was walking home 0 Pat walked home (19) ing(LðPat walkÞ) V SIMP(LðPat walkÞ) (20) ing(LðPat walk homeÞ) X SIMP(LðPat walk homeÞ) (10) a. A ‘ B because LðAÞ V LðBÞ (19) and (20) leave out the speech time S contributed by the past tense in (3a) and (4a). We will see in the next section that our arguments for (19) and (20) remain intact after S is added for (3a) and (4a). An easy route to (19) and (20) is (21). (21) a. LðPat walkÞ ¼ h[walk (p)]+ b. LðPat walk homeÞ ¼ LðPat walkÞh & ½homeðpÞ* ½homeðpÞ ¼ ½homeðpÞ½walkðpÞ; homeðpÞ+ ½homeðpÞ Generalizing from (21), what is crucial for (19) is that LðPat walkÞ be divisible, where a language L is divisible if ing(L) V L. (Thus, if L is divisible, then L is iterative and ing(L) V SIMP(L).) As for (20), observe that ingðLÞ V h* xL and LVxL * h implies L ? nc(U)* whence ing(L) X SIMP(L)

if

L is telic and L 4 nc(U)*.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

ingðLÞ

Tim Fernando 329

Assuming LðPat walk homeÞ 4 nc(U)*, the telicity (13a) of LðPat walk homeÞ yields (20). Clearly, (19) and (20) hold for various refinements of the languages mentioned there.3 In the next section, we consider refinements of languages given by constraints of the form L 0 L#. These constraints are comparable in function to meaning postulates, and typically lead to the addition of fluents. A simple example (injecting a bit of world knowledge) is ½dawnh* ½dusk 0 h+ ½noonh+

def

s 2 L0L# 5 for every factor s# of s; s#VL implies s#VL#: How does a constraint C 4 Pow(U)* refine a language L 4 Pow(U)*? For the case of C ¼ ½dawnh* ½dusk0h+ ½noonh+ L ¼ ½rain; dawn½rain* ½rain; dusk; a natural refinement of L by C is LC ¼ ½rain; dawn½rain* ½rain; noon½rain* ½rain; dusk which filters out the string [rain, dawn][rain, dusk] of length 2 from L, and fleshes out the remaining strings in [rain, dawn][rain]+[rain, dusk] by inserting noon in the middle. For arbitrary languages L and C over the alphabet Pow(U), we form the application LC of C to L in three steps (s1) superpose L with Pow(U)* (s2) intersect C with the superposition L & Pow(U)*, and (s3) extract the V-minimal strings of C \ (L & Pow(U)*) def LC ðC \ ðL&PowðUÞ* ÞÞV ¼ ðfs 2 CjsVLgÞV where, in general, LV is the set of the V-minimal strings in L 3 A refinement relevant to the progressive has to do with incremental changes in the degree to which a fluent u holds. We can encode increases in the degree of u as

ðdxÞ u-degðxÞ ^ ðdy < xÞ previous u-degðyÞ using a temporal operator previous to shift back one moment in the past.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

requiring that any stretch of time that begins with dawn and ends in dusk contains noon. In general, the idea is that for any languages L and L# over the alphabet Pow(U), the constraint L 0 L# is the set of strings s 2 Pow(U)* such that every stretch of s that subsumes L also subsumes L#. The precise notion of stretch is given by that of a factor, where s# is a factor of s if s ¼ us#t for some (possibly empty) strings u and t. We then define

330 Branching from Inertia Worlds LV

def

fs 2 Ljð"s# 2 LÞ sVs# implies s ¼ s#g:

Note that L & Pow(U)* [ L, but that (s1) allows us to impose C on L by intersection in (s2). We tidy up in (s3), removing excesses from (s1), as L [ LV 4 L: Evidently, LC V L. 3 INERTIA AND FORCE Turning now to inertia, consider (2a). Does (2a) imply that the car hit the tree after Pat stopped it? Not quite. By stopping the car, Pat negates a precondition for Lðcar hit treeÞ, that precondition being that the car not be at rest

If the car is to hit the tree, some force must put the car back into motion. In the absence of an intervening force on the stopped car, (2a) suggests that the collision was avoided. To formalize such reasoning, let us fix a set Inr 4 U of inertial fluents 2 Inr and there is a non-inertial fluent fu such that whenever u 2 Inr, u marking the application of a force to make u true at the next moment. We can then express persistence of an inertial fluent u as the constraint ðperÞ [u] h 0 (h[u] + [f u] h) true (that stating that u persists unless some force is applied to make u is, u false). Forces that make u true appear in the backward persistence constraint (pbk)

h½u

0 ð½uh + ½f uhÞ

stating that if u is true, it must have been so previously or else was allows us to formulate the forced to be true. Differentiating f u from f u constraint ‘succeed unless opposed’ (suo)

hÞ ½f uh 0 ðh½u + ½f u

saying an unopposed force on u brings u about at the next moment. the same, (suo) would have no bite (denoting, as it Were fu and f u would, the universal language Pow(U)*). For u equal to car-at-rest, the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(2) a. Pat stopped the car before it hit the tree.

Tim Fernando 331

collision in (2a) is put into question from two directions: forward from car-at-rest and backward from car-at-rest. One direction is enough to block the inference that the car hit the tree—which is fortunate, as we will shortly be forced to modify (per). Consider again the minimal pair (5), which we can analyse partially via (22), where in(p, d) is an inertial fluent for Pat-in-Dublin and R and S are non-inertial fluents for reference and speech time. (5) a. Pat left Dublin but is back (in Dublin). b. ?Pat has left Dublin but is back (in Dublin).

To account for the contrast in (5) on the basis of (22), it suffices that inðp; dÞ persists forward in (22b) but not in (22a). (23) a. LðPat left DublinÞ X h+ ½inðp; dÞ; S b. LðPat has left DublinÞ V h+ ½inðp; dÞ; S To derive (23b) from (22b), we need only appeal to (per) and the assumption that in (22b) no force is applied against inðp; dÞ alongside its (displayed) occurrence. By contrast, in (22a)/(23a), it would appear that R blocks inðp; dÞ from flowing to the right, just as a force against inðp; dÞ would. Accordingly, we weaken (per) for every inertial u to (peR)

h + ½RhÞ ½uh 0 ðh½u + ½f u

adding the possibility ½Rh to the right-hand side as an alternative to inertial flow h½u.4 Independent confirmation that R is a barrier to forward persistence is provided by (24). (24) a. Pat was in Dublin 0 Pat is in Dublin b. LðPat was in DublinÞ V ½inðp; dÞ* ½inðp; dÞ; Rh* ½S The non-entailment in (24a) suggests that in(p, d) in (24b) does not flow to S, as [in(p, d), S] would mean Pat is in Dublin. 4 Instead of weakening (per) to (peR), it is tempting to assume that R applies a force against all inertial fluents u occurring alongside it.

ðRaf Þ

½R; u 0 ½f u

Unfortunately, (Raf) would wreck our account of the non-veridicality in (2a), assuming R is put at the end of LðPat stop carÞ.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(22) a. LðPat left DublinÞ V ½inðp; dÞh* ½inðp; dÞ; Rh* ½S b. LðPat has left DublinÞ V ½inðp; dÞh* ½inðp; dÞ;h* ½R; S

332 Branching from Inertia Worlds A Reichenbachian approach to tense locates S relative to R. Let us write ‘R < S’ for the set of strings where R precedes S ‘R < S’

def

fa1 . . . an jð"i such that R 2 ai ÞS; [ aj g 1<j
and ‘R ¼ S’ for the set of strings where R and S co-occur ‘R ¼ S’

def

fa1 . . . an jð"iÞ R 2 ai 5 S 2 ai g:

On languages L where R but not S occurs (e.g. SIMP(L0), PROGO(L0), PERFo(L0)), we can then define the operators def def

ðLh* & h+ ½Sh* Þ \ ‘R<S’ \ Unpad ðL & h* ½Sh* Þ \ ‘R ¼ S’ \ Unpad

where Unpad consists of non-empty strings that neither begin nor end with h Unpad

def

PowðUÞ+ ðhPowðUÞ* + PowðUÞ* hÞ:

Then, for instance, Lo ðPat left DublinÞ ¼ Pasto ðSimpðL0 ÞÞ Lo ðPat has left DublinÞ ¼ PresðPerfo ðL0 ÞÞ where L0 is LðPat leave DublinÞ, and the subscripts o indicate that adjustments may be necessary to satisfy constraints involving, for instance, inertia. These constraints are hard inasmuch as the implications L1 0 L2 are non-defeasible. Indeed, (2a) non-defeasibly implies (25). (2)

a. Pat stopped the car before it hit the tree.

(25) Unless some force put the stopped car back in motion, the car did not hit the tree. Defeasibility in inertia creeps in not through 0 but through the nondeterminism + to the right of 0— for example, in the choice between [u]h and [fu]h in (pbk)

h½u 0 ð½uh + ½f uhÞ:

To assume that in a language L, no forces apply except those explicitly appearing in L is to restrict the application LC of the inertial constraints C to L as follows. In step (s1) for LC, we restrict the fluents added to the set Inr of inertial fluents, superposing L with Pow(Inr)*, rather than all

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Pasto ðLÞ PresðLÞ

Tim Fernando 333

of Pow(U)*. If we then carry out steps (s2) and (s3), the end result refines LC to LiC

def

ðC \ ðL & PowðInrÞ* ÞÞV

(with i standing for inertial). Clearly, LiC 4LC , the difference being that LiC adds only inertial fluents (in Inr) to L, whereas LC may introduce forces into L. There are, of course, cases where inertial flow is undesirable, as in (26), which presumably does not entail Pat is in Dublin.

To block in(p, d) from flowing into the box [R, S] in (26), it suffices to add a force fluent freezing in(p, d), so that LðPat has been in DublinÞ V ½inðp; dÞ* ½inðp; dÞ; f ðinðp; dÞÞh* ½R; S: In general, we can block inertial flow (forward) of L ’s postconditions in Lh*[R] by turning L into {s js 2 L}, where s is s with all the necessary at the end force fluents f u d

s

d

def

d

ju 2 Inr and s V h* ½u s & h* ½f u

ju 2 Inr \ aÞÞ. It may be the case that for the (that is, (sa) is sða [ ½f u perfect operator to apply to a language L, some forces must be at work in L. So why in (2a) are we so reluctant to add forces enabling the car to hit the tree? d

(2) a. Pat stopped the car before it hit the tree. Perhaps because, as we will see in the next section, we are not forced to; we may branch. 4 BRANCHING AND COHERENCE Whether or not the most natural reading of (2a) implies (2b), the question arises: How are we to interpret (2b)? (2) b. The car did not hit the tree but it might well have. Under dynamic semantics, the sentence the car did not hit the tree eliminates all epistemic possibilities where the car hit the tree, making the clause the car might have hit the tree unacceptable (Veltman 1996). This is all well and good if might is read epistemically as in (2c).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(26) Pat has been in Dublin. ½inðp; dÞ* ½inðp; dÞh* ½R; S

334 Branching from Inertia Worlds (2) c. The car did not hit the tree but ?for all we know, it might have. There is, however, a metaphysical reading of might, under which (2b) is perfectly acceptable (Condoravdi 2002). (2) d. Had Pat not stopped the car, it might well have hit the tree.

(27) a. s1 b. s2 c. s3

def def def

½car-at-rest ½car-at-rest ½car-at-rest; car-tree-contact½car-tree-contact ½car-at-rest; mayðs2 Þ ½car-at-rest

Whereas information in Veltman (1996) grows solely by elimination, we not only have L4L# implies LVL# but also expansive growth such as s3 V s1 in (27). Veltman’s might depends on the full epistemic state for its interpretation (and is in this sense said to be non-distributive); the fluent may(s) is interpreted pointwise at each possibility in the epistemic state. Up to now, the points in our epistemic states have been strings. To interpret may(s), we extend strings to branching strings formed from an alphabet not only by concatentation, with ss#

saying

s# follows s,

but also by a second binary operation b, with b(s,s#)

saying s# may follow s.

We should be careful not to confuse non-deterministic choice + with b. Epistemic uncertainty is encoded by + through a set of strings (branching and otherwise), whereas b encodes metaphysical non-determinism within a branching string. (Introducing metaphysical possibilities via b adds information insofar as it eliminates epistemic possibilities in which the branches are ruled out as metaphysical possibilities.) To illustrate, we can flesh out s3 in (27) as the branching string

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To make sense of might in (2d), let us assume that we can turn a string s into a fluent may(s) saying roughly that from the next moment on, s may be observed. We will presently do away with such fluents may(s), but for now, they are convenient for making the following point. In (27), if the string s1 depicts car stop and s2 depicts car hit tree, then s3 depicts car stop but may have hit tree.

Tim Fernando 335 def

ˆs

bð½car-art-rest; s2 Þ½car-at-rest:

In ˆs, the first snapshot ½car-at-rest is followed by [car-at-rest], which clashes with the first snapshot of s2 (on whether or not the car is at rest), making s2 a counterfactual continuation of ½car-at-rest.5 We can read (28) off ˆs inasmuch as chopping off the suffix [car-a-rest] from ˆs restores s2 as a live option after ½car-at-rest. (28) Had it not stopped, the car may have hit the tree.

½car-at-rest; car-at-rest ¼ ½car-at-rest [ ½car-at-rest is that it contains clashing parts [car-at-rest] and ½car-at-rest that ought not to be union-ed, but must be kept apart, hence the branching. As with superposition &, it suffices to combine strings of the same length (padding them with empty boxes h at either end, if necessary). An example with R equal to the set nc(U) of non-contradictory snapshots is (29), where ˆs, s1 and s2 are as in (27). (29) ˆsh ¼ bð½car-at-rest; s2 Þ ½car-at-resth ¼ s1 h &ncðUÞ hs2 ¼ ½car-at-rest½car-at-resth &ncðUÞ h½car-at-rest; car-tree-contact½car-tree-contact In general, we form s &Rs# from strings s ¼ a1 . . . an and s# ¼ a#1 . . . a#n that are both in Rn for some fixed n, and let k be the largest i < n such that ða1 [ a#1 Þ . . . ðai [ a#i Þ 2 R* —that is, k 5

def

maxfi < njaj [ a#j 2 R for 1 < j < ig.

An alternative to ˆs (suggested by Stefan Kaufmann) is the branching string s_

def

bð½car-at-rest; car-tree-contact; s1 Þ½car-tree-contact:

obtained from the string s4 in (27d) by unwinding may(s1). (27)

d.

s4

def

½car-at-rest; car-tree-contact; mayðs1 Þ½car-tree-contact:

According to s_/s4, the car hit the tree (s2) but may have instead stopped (s1).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

What is the general mechanism for forming ˆs from s1 and s2? The mechanism is a generalization &R of superposition & based on a restriction R 4 Pow(U) of the alphabet Pow(U) to a fixed set R of ‘legal’ snapshots. For R ¼ Pow(U), &R is just &. But the idea behind throwing out from R a snapshot such as

336 Branching from Inertia Worlds (In the case of (29), s ¼ s1h, s# ¼ hs2, n ¼ 3 and k ¼ 1.) The R-zipper s &Rs# is defined iff k > 1, in which case s &Rs# preserves all of s and as much of s# as s and R allow, the remainder of s# attaching to s via b. That is, if k > 1, let (i) s$ be the superposition a1 . . . ak & a#1 . . . a#k of the k-length prefixes of s and s# s$

def

ða1 [ a#1 Þ . . . ðak [ a#k Þ

(so that s$ ¼ s &s# if k ¼ n) (ii) s+ and s#+ be the suffixes of s and s# from positions k + 1 to n def

ak+1 . . . an a#k+1 . . . a#n

so that s ¼ a1 . . . ak s+ and s#¼ a#1 . . . a#k s#+ (e.g. in (29), s+ is [car-atrest]h while s#+ is ½car-at-rest; car-tree-contact½car-tree-contact) (iii) s &Rs# be s$ plus any remaining parts s+ and s#+ attached by concatenation and b, respectively, s &R s#

def

s$ bðs$; s#+ Þs +

if k ¼ n otherwise

(whence (29) for R ¼ nc(U)). The requirement k > 1 ensures that s# can attach to s, with s &Rs# maximizing the attachment of s# to s. Next, given languages L, L# 4 R*, we collect R-zippers from their strings in the R-zipper L &R L#

def

[ fa1 . . . an &R a#1 . . . a#n ja1 . . . an 2 L and

n>1

a#1 . . . a#n 2 L# with a1 [ a#1 2 Rg: The ordinary (non-branching) strings in L &R L# are the strings in R* belonging to the ordinary superposition L & L#. (30)

ðL &R L#Þ \ PowðUÞ* ¼ ðL & L#Þ \ R*

The remainder (L &R L#) (L & L#) represents cases of branching and counterfactual continuations, on which we might impose additional requirements (beyond snapshots being from R). Going back to before, the idea roughly is to represent a veridical reading of A before B, in which both arguments A and B are asserted to

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

def

s+ s#+

Tim Fernando 337

hold, by certain non-branching strings from (30) where L ¼ h* LðAÞ h* and L# ¼ h* LðBÞh* .(We say certain because such strings must satisfy additional constraints; e.g. an occurrence of an Astring must precede that of a B-string.) In a counterfactual reading where B is asserted not to hold, the strings branch.6 For concreteness, let us focus on the branching reading of (2a) in which the car does not hit the tree. As a representative branching string for that reading, ˆs (i.e. bð½car-at-rest; s2 Þ½car-at-restÞ) suggests that (2a) implies (31a). (2) a. Pat stopped the car before it hit the tree.

Indeed, depending on what we make of the phrase going to, we might extract (31b) from the following modification of ˆs. Let us add the snapshot a

def

½car-at-rest; car-tree-contact

to s2, deriving (as it were) as2 from hs2 by backward inertial flow (pbk). (pbk)

h½u

0 ð½uh + ½f uhÞ

We then sharpen (29) to (32). (32) s1 h &R as2 ¼ bða; s2 Þ½car-at-resth In (32), s1 depicts car stop, while as2 depicts car hit tree. The transition in s1 from ½car-at-rest to [car-at-rest] requires a force (assuming car-at-rest is inertial) that in (32) gives rise to a branch marking car hit tree counterfactual. It is easy to see how in general branching from the zipper arises from a force disturbing inertia: no force, no change in inertial fluents. Furthermore, a force leads to branching only if it is opposed. In (33), Pat does not contribute a force to stop the car; therefore, it is natural to read (33) as entailing the car hit the tree. (33) Pat did not stop the car before it hit the tree. Next, consider (34) and (35). 6 Mixing non-branching and branching strings, we can represent the non-committal readings in Beaver & Condoravdi (2003) of A before B where B may or may not hold. That is, non-committal readings arise from epistemic uncertainty (i.e. non-deterministic choice +) between veridical and counterfactual ones. More below.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(31) a. Before Pat stopped the car, it may have been the case that the car was on course to hit the tree. b. Before Pat stopped the car, it was going to hit the tree.

338 Branching from Inertia Worlds (34) The car was at rest before it hit the tree. (35) The car came to rest before it hit the tree. How can we account for reading before veridically in (34) but counterfactually in (35)? The car-hit-tree string as2 cannot attach anywhere where the car is at rest (in (34)), but can attach before the car comes to rest (in (35)). We arrive at constraint (36), where beforec and beforet are the counterfactual and veridical forms, respectively, of before. (36) A beforec B

implies

going-toðBÞ beforet A

(c1) ð"s# 2 jðsÞÞs#R ¼ sR (where sR is the R-truncation of s as defined in section 2) and includes s (c2) s 2 jðsÞ: It is useful to think of strings in j(s) as R-continuations of s, and to expand a language L into the language Lj of R-continuations of strings in L Lj

def

[ fjðsÞjs 2 Lg:

Clearly, if j maps every s 2 R+ to {s}, then Lj ¼ L. But more generally, from (c2), we may conclude that Lj contains L L 4 Lj and from (c1), that L and Lj have the same R-truncations LR ¼ ðLj ÞR : In particular, SIMP(L) ¼ (SIMP(L)j)R and PERFo(L) ¼ (PERFo(L)j)R for languages L where R does not occur. By contrast, PROGo(L) may differ from (PROGo(L)j)R if j(s) 6¼ ðsÞ, with alternatives to s encoded in j(s) {s}. The crucial point is that R is a realis marker: beyond R, events may develop in as many different ways as j allows.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To satisfy (36), we have fleshed out hs2 to as2, under the assumption that going-to(car-hit-tree) implies car-at-rest and car-tree-contact. But just what does going-toðBÞ mean? To bring out the non-veridicality of the progressive, recall the aspectual operators SIMP, PROGo and PERFo in (18) employ the reference time R to mark what has come to pass from what is yet to come. We can specify branching at R through a function j : Pow(U)+ / Pow(Pow(U)+) mapping a string s 2 Pow(U)+ to a set j(s) of strings that are identical with s up to R

Tim Fernando 339

The presentation of the language Lj above is designed to bring out the similarity with the interpretation of metaphysical might in Condoravdi (2002). Condoravdi imposes a condition of historical necessity (Thomason 1984) on the metaphysical modal base, which is expressed above as (c1). To link Lj to branching strings, it is convenient to alter the presentation slightly. Given a string s, let sR be the suffix of s cut off by the R-truncation sR of s in that s ¼ sR sR :

jR ðsÞ

def

fs#R js# 2 jðsÞ fsgg:

Then, Lj ¼ L [ fsR s#js 2 L and s# 2 jR ðsÞg and if say, jR(s) ¼ {s#, s$} we can attach s# and s$ to s via b, encoding j(s) as the branching string bðbðsR ; s#Þ; s$Þ sR : This encoding generalizes to strings s and functions j such that j(s) is finite.7 Conversely, from say, the branching string bð½R; s#Þbða; s$Þ b we can get three R-continuations of [R]ab f½Rs#; ½Ras$; ½Rabg

4 jð½RabÞ:

The set of R-continuations obtained from a branching string is partial in the same sense that the snapshots are partial. What can we say about R-continuations? Can we, for instance, assume (37), which assigns s and s# the same R-continuations whenever they are identical up to R?

7 The order in which jR(s) is enumerated is immaterial, assuming we impose the following equations between branching strings

bðbðs; s#Þ; s#Þ ¼ bðs; s#Þ bðbðs; s#Þ; s$Þ ¼ bðbðs; s$Þ; s#Þ not to mention, sb(s#,s$) ¼ b(ss#,s$) and ðss#Þs$ ¼ sðs#s$Þ. More in Fernando (2005).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(For instance, (h[R][u])R ¼ [u].) Next, let us collect the alternatives to sR specified by j(s) in

340 Branching from Inertia Worlds (37) If sR ¼ s#R then j(s) ¼ j(s#). If (37) were true then by (c2), every string identical to s up to R must be in j(s) sR ¼ s#R

implies s# 2 jðs#Þ ¼ jðsÞ

and so by (c1), jðsÞ ¼ fs#jsR ¼ s#R g:

(38) j(s) consists only of strings that are reasonably probable given what has happened up to R. An argument against (38) is provided by the following passage from Landman (1992). if an accomplishment manages to get completed, it is unproblematic to assume (in retrospect) that the progressive is true during the development stage . . . even if the event gets completed against all odds. (p. 14) It would seem that if we are to restrict j(s) to the ‘reasonably probable’ continuations, then that probability judgment had better be made on the basis of events not only up to R but also beyond R, contra (38). But then how are we to keep that probability judgment from degenerating to an assignment of 0 to counterfactual events and 1 to actual events? It is, of course, easy enough to replace j(s) in (38) by j(s) {s}. But it is even easier to put (38) and probabilities aside, and try make do with (36). (36) A beforec B

implies going-toðBÞ beforet A

There is a vagueness in what go-to(B) means that is difficult to tease out from the progressive. Be that as it may, we can apply (36) to fix a problem pointed out in Beaver & Condoravdi (2003) for the ‘Anscombe/Landman analysis’ of before: Even if David never won a gold medal at anything in his whole life, provided he ate lots of ketchup, (x) is predicted true. (x)

David ate lots of ketchup before he made a clean sweep of all the gold medals in the Sydney Olympics.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We must reject (37) if we want more restricted sets j(s). (38) is a restriction corresponding to the normality condition in Beaver and Condoravdi (2003).

Tim Fernando 341

Assuming before in (x) is counterfactual, (36) and (x) yield (39) with before veridical. (39) Before David ate lots of ketchup, he was going to make a clean sweep of all the gold medals in the Sydney Olympics. (x) is rejected because no string describing David winning an Olympic medal attaches via b to a factual string. Does (36) hold if beforec in its left-hand side is replaced by beforet? (40) suggests perhaps not.

Evidently, there are limits to how far back going-to(B) can stretch. But then, (36) was motivated above by branching from a zipper— apologies for mixing metaphors—and not by the (veridical) case of a zipper coinciding with superposition. (What (36) adds to the zipper is illustrated above by the step from (29) to (32).) We have defined R-zippers to be as close to superposition as R allows, zipping zippers as high up as we could without straying from R. This is in line with the constraint Maximise Discourse Coherence (MDC) in Asher & Lascarides (2003) if we agree that discourse coherence increases as the zipper is zipped (bringing the two strings closer together). Just as epistemic uncertainty (represented in a language as non-deterministic choice + between its possibilities/strings) is subject to entailment, indeterminism expressed by branching (b) is subject to coherence. A further constraint one might apply to counterfactual before, suggested by the initial branch point condition in Beaver & Condoravdi (2003), is to sharpen the veridical before in the right-hand side of (36) to one moment before, strengthening (31b) to (31c). (41) A beforec B implies going-to ðBÞ one-moment-beforet A (31) b. Before Pat stopped the car, it was going to hit the tree. c. Up until Pat stopped the car, it was going to hit the tree. Translating (41) into an English test is perhaps more awkward than is the case for (36). At any rate, it has a precise enough meaning for our discrete branching string representations: in the zipper representing A beforec B, an A-string is formed by position k + 1, where k is as high up as the zipper zips up. (It is tempting to view (41) as an instance of MDC

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(40) a. Columbus discovered America before the Soviet Union collapsed. b. ?Before Columbus discovered America, the Soviet Union was going to collapse.

342 Branching from Inertia Worlds for counterfactual readings; any higher up and we get a veridical reading.) The discreteness here of time reflects the temporal granularity of the conceptualized event, and more generally, the perspectival character of branching. What one moment before and going-to(B) mean are arguably matters of perspective, not metaphysics. Similarly, the well in (2b) reflects how picky the continuations j(s) of a string s might be (one moment) before Pat stopped the car. (2) a. Pat stopped the car before it hit the tree. b. The car did not hit the tree but it might well have.

Let us sum up. Events were formulated as strings over an alphabet consisting of sets of fluents in section 2. These strings were then extended along two dimensions, with an eye to Reichenbachian treatments of the perfect and the progressive. In section 3, strings were extended to the right (to attach the reference time for the perfect), subject to inertial constraints. And in section 4, they were extended sideways to accommodate branching beyond the reference time (for the progressive). We can extend strings lengthwise and sideways indefinitely to obtain, at the limit, worlds from extensions that satisfy appropriate constraints such as U-bivalence \u2U h

0 ð½u + ½ uÞ

(deciding at each moment whether u holds or not) and U-noncontradictoriness ncðUÞ*

¼

0 ;Þ \ ð½u; u

u2U

). But we need not proceed to the (ruling out co-occurrences of u and u limit to examine entailments. Sidestepping worlds, we can work instead with superposition &, subsumption V, 0-constraints and zippers &R. The question remains: what does all this have to say about Dowty’s intuition that the progressive of an event is true at world w and time I if the event culminates in (I, w)-inertia worlds? Within our framework, an event culminates in a world containing a string representing the event. Attaching Reichenbach’s reference time R to the right end point of the interval I, we allow branching at R so that the progressive of an event might be true in a world where that event does not culminate. Can we extract from this a concept of an (I, w)-inertia world where every event in progress at (I, w) is guaranteed to culminate? Not quite.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

5 CONCLUSION

Tim Fernando 343

Consider the strings ab[u] and ab½ u where R 2 b and a, b 2 nc(U). We can combine the two strings to form the branching string u ab½u &ncðUÞ ab½

¼

abðb; ½ uÞ½u

ab½ u &ncðUÞ ab½u

¼

abðb; ½uÞ½ u

or

Acknowledgements This paper grew out of a talk titled ‘What is an inertia world?’ presented at the Language under Uncertainty workshop in Kyoto University, 21–23 January 2005. I thank the organizers and participants of this workshop and the editors of this special issue, Takao Gunji, Stefan Kaufmann and Yukinori Takubo. I am indebted to two anonymous referees as well as Stefan Kaufmann for detailed and penetrating comments that resulted in substantial revisions. TIM FERNANDO Computer Science Department Trinity College Dublin 2 Ireland e-mail: [email protected]

8

The alternative (mentioned in section 1) is to insist that no two events headed for a conflict can be in progress at the same pair (I, w). To pursue this alternative, we can impose further constraints (on the realization of the progressive) to that effect. I leave a detailed development of this for some other occasion. 9 Behind these forces lurk automata, which I have left out above but explore elsewhere (Fernando 2005, 2006).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

;ncðUÞ* , as u clashes or abðbðb; ½ uÞ; ½uÞ; but not the string ab½u; u . In other words, two different events may be partially realized at with u (I, w) even though there is no single world where they can both culminate. To guarantee the culmination of an event e, it seems we must relativize the notion of an inertia world not to (I, w) but to e.8 Whether or not this trivializes Dowty’s intuition, it is hardly surprising then that the bulk of this paper should concern not worlds but events. These events are shaped by perspectives specifying constraints that strings representing events must satisfy. The strings may, under pressure from other strings, branch due to forces that compete to overturn inertia.9

344 Branching from Inertia Worlds REFERENCES McCarthy, J. & Hayes, P. (1969), ‘Some philosophical problems from the standpoint of artificial intelligence’. In M. Meltzer and D. Michie (eds.), Machine Intelligence 4. Edinburgh University Press. Edinburgh. 463–502. Moens, M. & Steedman, M. (1988), ‘Temporal ontology and temporal reference’. Computational Linguistics 14:15–28. Parsons, T. (1990), Events in the Semantics of English: A Study in Subatomic Semantics. MIT Press. Cambridge, MA. Reichenbach, H. (1947), Elements of Symbolic Logic. London. Macmillan. Steedman, M. (2000), The Productions of Time. Draft, ftp://ftp.cogsci.ed.ac.uk/ pub/steedman/temporality/temporality. ps.gz. An update of ‘Temporality’. In J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language. Elsevier. North Holland. 895–935. Tenny, C. (1987), Grammaticalizing Aspect and Affectedness. Dissertation, Department of Linguistics and Philosophy, MIT. Cambridge, MA. Thomason, R. (1984), ‘Combinations of tense and modality’. In D. Gabbay and F. Guenthner (eds.), Handbook of Philosophical Logic. Reidel. 135–65. Veltman, F. (1996), ‘Defaults in update semantics’. Journal of Philosophical Logic 25:221–61. Vendler, Z. (1967), Linguistics in Philosophy. Cornell University Press. Ithaca, NY. First version received: 10.10.2005 Second version received: 02.10.2006 Accepted: 01.04.2007

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Asher, N. & Lascarides, A. (2003), Logics of Conversation. Cambridge University Press. Cambridge, MA. Beaver, D. & Condoravdi, C. (2003), ‘A uniform analysis of before and after’. In R. Young and Y. Zhou (eds.), Proceedings of Semantics and Linguistic Theory XIII. Cornell Linguistics Circle Publications. Ithaca, NY. 37–54. Condoravdi, C. (2002), ‘Temporal interpretation of modals: modals for the present and for the past’. In D. Beaver, S. Kaufmann, B. Clark and L. Casillas (eds.), The Construction of Meaning. CSLI. Stanford, CA. 59–88. Dowty, D. R. (1979), Word Meaning and Montague Grammar. Reidel. Dordrecht. Fernando, T. (2004), ‘A finite-state approach to events in natural language semantics’. Journal of Logic and Computation, 14:79–92. Fernando, T. (2005), ‘Events from temporal logic to regular languages with branching’. In G. Jaeger, P. Monachesi, G. Penn, J. Rogers and S. Winter (eds.), Proceedings of 10th Formal Grammar and 9th Mathematics of Language. Stanford, Edinburgh. 27–38. Fernando, T. (2006), ‘Finite-state temporal projection’. In O. Ibarra and H.-C. Yen (eds.), Proceedings of 11th International Conference on Implementation and Application of Automata (CIAA 2006). Lectures Notes in Computer Science 4094. SpringerVerlag, Berlin. 230–41. van Lambalgen, M. & Hamm, F. (2004), The Proper Treatment of Events. Blackwell. Oxford. Landman, F. (1992), ‘The progressive’. Natural Language Semantics 1:1–32.

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents