VOLUME 26 NUMBER 1 FEBRUARY 2009
CONTENTS ARIEL COHEN No Alternative to Alternatives
1
49
EUNHEE LEE AND JEONGMI CHOI Two nows in Korean
87
FORTHCOMING ARTICLES ALEX LASCARIDES AND NICHOLAS ASHER: Agreement, Disputes and Commitments in Dialogue
VOLUME 26 NUMBER 1 FEBRUARY 2009
YAEL GREENBERG Presupposition Accommodation and Informativity Considerations with Aspectual Still
JOURNAL OF SEMANTICS
JOURNAL OF SEMANTICS
issn 0167-5133
VOLUME 26 NUMBER 1 FEBRUARY 2009
Journal of
SEMANTICS www.jos.oxfordjournals.org
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 2009 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 26, 4 issues, 2009) Institutional Print edition and site-wide online access: £187/$365/=281 C Print edition only: £178/$347/=267 C Site-wide online access only: £178/$347/=267 C Personal Print edition and individual online access: £70/$137/=105 C Please note: £ Sterling rates apply in Europe, US$ 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 £187/$365/=281. 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 Linda Hann, E-mail: lhann@lhms. fsnet.co.uk. Phone/fax: 01344 779945. 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 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 2009 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 Volume 26 Number 1
CONTENTS ARIEL COHEN No Alternative to Alternatives
1
YAEL GREENBERG Presupposition Accommodation and Informativity Considerations with Aspectual Still
49
EUNHEE LEE AND JEONGMI CHOI Two nows in Korean
87
Please visit the journal’s web site at www.jos.oxfordjournals.org
Journal of Semantics 26: 1–48 doi:10.1093/jos/ffn005 Advance Access publication August 28, 2008
No Alternative to Alternatives ARIEL COHEN Ben-Gurion University
Abstract
‘Only if there are alternatives can one hope to get insight into what is truly at stake.’ Peter F. Drucker, The Effective Executive (p. 147) 1 INTRODUCTION How does focus restrict the domain of quantification? Rooth (1985, 1992) proposes a two-dimensional account. He claims that every expression u has two semantic values: in addition to the ordinary semantic value, ½½uO, u also has a focus semantic value, ½½uF, which is a set of alternatives to the focused elements of u. Rooth proposes a compositional way to compute the focus semantic value: the focus semantic value of u is a function of the focus semantic values of the parts of u. Moreover, the combinatory rules used to compute the focus semantic value are the same rules used in the computation of the ordinary semantic value. For example, if the ordinary semantic value ½½:uO is the negation of ½½uO, then the focus semantic 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
Rooth’s (1985, 1992) theory of focus requires, in addition to the ordinary semantic value of an expression, the focus semantic value, which is a set of alternatives generated by focus. Rooth claims that the union (disjunction) of the focus semantic value is accommodated into the restrictor of an adverbial quantifier. More recently, however, some researchers (Krifka 2001; Geurts & van der Sandt 2004) have argued convincingly that what is accommodated is, in fact, the existential presupposition induced by focus. It would appear, then, that there is no need for assuming the focus semantic value. However, in this paper, I argue that, although the primary effect of focus is, indeed, presuppositional, the focus semantic value cannot be dispensed with. Not only is the focus semantic value necessary but, in fact, additional semantic values are required too. Unlike focus, the analyses of these other semantic values cannot be reduced simply to existential presupposition. I will concentrate on a special reading of some quantificational sentences, the relative reading, whose adequate account, I propose, requires the use of semantic values triggered by alternatives to various elements: the focus, background marking and the world of evaluation.
2 No Alternative to Alternatives value ½½:[u]FF is a set containing the negations of the elements in ½½[u]FF. According to Rooth, the union (or disjunction) of the focus semantic value is accommodated into the restrictor of a quantificational adverb (henceforth Q-adverb). Geilfuß (1993) and de Hoop & Sola` (1995) make a similar proposal regarding determiners. For example, the preferred interpretation of (1) is that the majority of linguists who arrived did so by bus; it does not mean that the majority of all linguists in the world came by bus. (1) Most linguists came by [bus]F.
(2) People usually manage to survive a week without [food]F. If the role of focus were to provide a set of alternatives that restrict the domain of usually, we would expect (2) to mean that people who manage to survive a week without something usually survive a week without food. However, (2) has no such reading; rather, it means that people who have to survive a week without food usually make it. This is accounted for by the proposal that it is the presupposition of manage, rather than the alternatives introduced by focus, that is accommodated into the restrictor.3 Such examples abound,4 and I believe they 1
More accurately, Krifka claims that it is not focus as such that gives rise to the accommodated presupposition, but that presuppositional indefinites are characterized by de-stressing. 2 For dissenting opinions, see, among others, Rooth (1999), who argues that assuming that focus contributes an existential presupposition leads to incorrect predictions; Beaver (2001), who denies that presuppositions are accommodated into the restrictor, and von Fintel (1994), who claims that focus affects the context, rather than the restrictor. 3 An anonymous reviewer wonders why it is the presupposition of manage, rather than the existential presupposition of focus, that is accommodated. The reason might be due to the distinction between the two types of presupposition trigger (Zeevat 1992; Geurts forthcoming): lexical triggers, like manage, must be accommodated, while focus, which is anaphoric, does not have to be. 4 See also Schubert & Pelletier (1987) and Beaver & Clark (2003) for similar observations.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Krifka (2001) and, more recently, Geurts & van der Sandt (2004) have objected to this two-dimensional account. They suggest, instead, that it is the existential presupposition produced by focus that is accommodated, and there is no need for a set of alternatives.1 These proposals have attracted considerable interest. There appears to be fairly convincing evidence that what is accommodated into the restrictor really is presupposition, and not the union of the set of focus alternatives as compositionally defined by Rooth.2 Perhaps the strongest piece of evidence comes from cases where there is a clash between presupposition and focus; in such cases, it is presupposition that is accommodated into the restrictor. The following example is from Cohen (1999b):
Ariel Cohen 3
5
The article by Geurts and van der Sandt is published together with comments by several linguists. For more discussion on this issue, see these comments and the response from Geurts and van der Sandt.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
provide strong support for the positions of Krifka and Geurts and van der Sandt, in so far as they demonstrate that what is accommodated into the restrictor is a presupposition, and not Rooth’s alternatives.5 Does this mean that the device of alternatives can be done away with? Not necessarily. First, note that we do not have to eliminate alternatives from the theory. We could still maintain the use of alternatives, if we change Rooth’s compositional definition, and replace it with a set of rules for generating presuppositions, i.e. rules that reflect the projection behaviour of presuppositions (Cohen 1999b). For example, since presupposition is normally unaffected by negation, ½½[u]FF and ½½:[u]FF will be the same. Note that giving a role to alternatives need not contradict the theories of either Krifka or Geurts and van der Sandt; it would simply add another dimension to their accounts. But why would we want to go that way? Why do we need a two-dimensional account, with more than one semantic value, when we could simply use existential presupposition directly? I believe there is a reason. In this paper, I argue that if the full range of readings of quantificational structures is to be accounted for, one needs additional semantic values, indicating alternatives to other elements besides focus. The analyses of these other elements cannot be reduced simply to existential presupposition; hence, whatever one wishes to propose regarding focus, there is no alternative to a multidimensional theory. The rest of the paper is organized as follows. In section 2, I consider some semantic values that have already been proposed in the literature, and propose another one: the B semantic value. In section 3, I discuss a specific reading of some quantificational sentences: the relative reading. In section 4, I argue that a proper account of relative readings requires the semantic values from section 2. In section 5, I propose an answer to two questions: why only some quantifiers exhibit relative readings and what is the relation between the relative and non-relative reading of a quantifier. I argue that answering these questions requires two additional semantic values, which consider alternative possible worlds. Section 6 provides some speculative thoughts about the implications of the theory and concludes the paper.
4 No Alternative to Alternatives 2 ALTERNATIVES TO FOCUS AND NON-FOCUS
2.1 Focus semantic value
(3) A: Eva only gave xerox copies to the GRADUATE STUDENTS. B: No, PETR only gave xerox copies to the graduate students. In A’s utterance, there is pitch accent on graduate students, and the phrase is interpreted as focus: only associates with it, and the interpretation is that Eva gave xerox copies to nobody besides graduate students. B’s utterance contains a second occurrence of graduate students. This utterance is also interpreted as if there is focus on graduate students, and only associates with this focus: B is saying that Petr gave xerox copies to 6 Keeping in mind that, as discussed above, the focus semantic value must be calculated according to the rules of presupposition projection, rather than according to Rooth’s proposed rule.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Rooth (1985) proposes that every expression u has, in addition to its ordinary semantic value ½½uO, also a focus semantic value ½½uF. The intuition underlying the focus semantic value can be put roughly as follows: the focus semantic value is generated by replacing the focusmarked element (or elements) with its alternatives. Focus-sensitive operators interact with this focus semantic value. For example, in adverbial quantification, the union of the focus semantic value is used to restrict the domain of quantification.6 What exactly is focus? It is very hard to answer this question, especially since different sources may mean different things by this word: the word may refer to a psychological, intonational, syntactic, semantic, or pragmatic notion (cf. Beaver et al. 2004). Focus has a distinct informational role: intuitively, the focused element is ‘new’, ‘informative’, or ‘unexpected’, whereas the unfocused part of the sentence is ‘background’, ‘old’, or ‘given’. It is notoriously hard to provide a precise definition of these notions (see Schwarzschild 1999 for discussion), but the intuitive meaning, I believe, is fairly clear. For the purposes of the current paper, a precise theory of the informational role of focus is not necessary. In addition to its informational role, focus is often associated with a specific intonation pattern: what Jackendoff (1972) called the A-accent, characterized by a fall. This pattern, however, is not a necessary condition for a phrase to be interpreted as focus. Of particular relevance here is the phenomenon of second occurrence focus, exemplified by the following exchange, from Partee (1991, 179):
Ariel Cohen 5
2.2 An additional semantic value Jackendoff (1972) distinguishes between A-accent, which as we have seen is usually used to indicate focus, and B-accent. The latter is characterized by a fall–rise contour (L+H* in the notation of Pierrehumbert 1980).8 Bu¨ring (1997, 1999) discusses the interpretation of elements pronounced with this intonation contour. He claims that, like focus, they too introduce alternatives, and proposes that their adequate account requires an additional semantic value, which he calls the topic semantic value. According to Bu¨ring, this semantic value takes into account alternatives to both the B-marked and the focusmarked elements. The notation I will use for it is, therefore, ½½uB+F. This is a set of sets of elements of the type of u. In each such set, all elements share a B-marked constituent, but vary with respect to the focus. Following the standard assumption that the meaning of a question is the set of its possible answers, Bu¨ring’s semantic value can be interpreted as a set of questions. Bu¨ring argues that a question– answer pair is felicitous only if the question is a member of the topic semantic value of the answer. For example, Bu¨ring notes that (4b) is a felicitous answer to (4a). (4) a. Which book would Fritz buy? 7
Cf. Rooth (1995) who ‘would like to leave aside the question whether this expected focus marking is really phonologically defensible’ (p. 270). 8 A more accurate representation, which includes the boundary tone, is probably L+H* LH% (Pierrehumbert & Hirschberg 1990; Steedman 1991; Vallduvı´ & Zacharski 1994; Vallduvı´ & Engdahl 1996); a precise characterization of the intonation of B-accent, however, is not necessary for our purposes here.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
nobody besides graduate students. Yet, the phrase graduate students is pronounced very differently this time and does not appear to carry a pitch accent. Does this mean that focus has no obligatory phonetic realization? This question cannot be said to have been satisfactorily settled. Rooth (1996), Bartels (2004) and Beaver et al. (2004) have conducted experiments to investigate whether second occurrence focus is realized phonetically and have found no pitch accent. However, they do report finding other phonetic correlates, like a small increase in duration and a very slight (possibly insignificant) increase in intensity. Doing justice to this question lies outside the scope of this paper. Instead, I will simply treat focus as a feature, saying little about its phonological manifestation: some parts of a sentence may be marked with this feature, others would not.7
6 No Alternative to Alternatives b. Well, [I]B would buy [‘The Hotel New Hampshire’]F. The topic semantic value of (4b), according to Bu¨ring, is:
This corresponds to the following set of questions: (6) fWhich Which Which Which
book book book book
would would would would
you/I buy?, Paul Simon buy?, Fritz buy?, Fritz’s brother buy?, . . .g
Since the question (4a) is a member of this set, the exchange in (4) is felicitous. An important question to consider is whether B-accent, like A-accent (when it is associated with focus), gives rise to an existential presupposition. It is hard, however, to answer this question based on exchanges like (4). The reason is that, in addition to whatever presuppositions are induced by the B-accent itself, Bu¨ring’s constraint on felicity is also operative. In this case, this constraint requires that the context contain a question of the form (7) Which book did x buy? Since (7) presupposes that x bought some book, the felicity of (4b) requires the truth of (8). (8) Somebody bought some book. This is precisely what we would get if we assumed that B-accent induces an existential presupposition.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(5) f fI would buy ‘War and Peace’, I would buy ‘The Hotel New Hampshire’, I would buy ‘The World According to Garp’, . . .g, fPaul Simon would buy ‘War and Peace’, Paul Simon would buy ‘The Hotel New Hampshire’, Paul Simon would buy ‘The World According to Garp’, . . .g, fFritz would buy ‘War and Peace’, Fritz would buy ‘The Hotel New Hampshire’, Fritz would buy ‘The World According to Garp’, . . .g, fFritz’s brother would buy ‘War and Peace’, Fritz’s brother would buy ‘The Hotel New Hampshire’, Fritz’s brother would buy ‘The World According to Garp’, . . .g, . . .g
Ariel Cohen 7
(9) a. A: B: b. A: B:
I thought they all took one. [Ann]B did (p. 170). Can Jack and Bill come to tea? [Bill]B can (p. 173).
B’s response in (9a) does not presuppose that somebody took one. This fact can be confirmed by standard presupposition tests. For example, if B’s answer is inside the scope of a possibility modal, it no longer follows that anyone did: (10) It is possible that [Ann]B did. Now consider what happens when Ann is focused: (11) a. [Ann]F took one. b. It is possible that [Ann]F took one. In this case, (11a) does presuppose that somebody took one, a presupposition that projects from within the scope of a possibility modal, as can be seen in (11b).9 Similarly, B’s response in (9b) does not presuppose that somebody can come to tea; indeed, it does not follow from (12) that anyone can: (12) It is possible that [Bill]B can. In contrast, (13a), where Bill is focused, does presuppose that someone can come for tea, and this presupposition can survive a possibility modal: 9
Of course, (11a) would no longer be a felicitous response to A’s statement.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Does this mean that we can do away with Bu¨ring’s topic semantic value and rely on existential presupposition alone, as has been proposed for focus by Krifka (2001) and Geurts & van der Sandt (2004)? The answer is no, for two reasons. One reason why an additional semantic value is necessary is that even if B-accent does give rise to an existential presupposition, we would be able to explain why (4b) presupposes (8), but not why it presupposes a question of the form (7); the finer distinctions made possible by the topic semantic value appear necessary. The second reason is that there are good reasons to deny that B-accent, on its own, induces an existential presupposition. This can be seen more clearly when we consider examples with B-accent, but without A-accent. Consider the following examples, from O’Connor & Arnold (1973):
8 No Alternative to Alternatives (13) a. [Bill]F can come to tea. b. It is possible that [Bill]F can come to tea. Hence, if Bu¨ring’s theory is correct, it provides evidence that a multidimensional analysis of semantic value is necessary anyway, regardless of the way focus is dealt with. Hence, the motivation for doing away with the focus semantic value is substantially lessened.
2.3 Contrast semantic value
(14) a. A: What did the pop stars wear? B: The [female]B pop stars wore [caftans]F (Bu¨ring 1997, ex. 12, p. 56). b. Eight boys had read ‘City of Glass’. [Two]B boys had even read [the entire trilogy]F (Bu¨ring 1997, ex. 21, p. 100). Whether or not B-accented elements are appropriately called topics, it is clear that they belong to the background. I will therefore henceforth refer to such elements as Background-marked, or B-marked. What distinguishes these elements from other elements in the background? Here, there appears to consensus that B-marked elements are contrastive in some sense. This notion can be made more concrete in terms of what Vallduvı´ & Vilkuna (1998) call thematic contrast.10 Suppose we have a contrast set of comparable elements M ¼ fa, b, c, . . .g. In a contrastive context, if it is said, of some property P, that P(a), then it is implied that some comparable property holds for some other element of M.11 Hence, if an element requires a contrastive context, and is not a focus, then it is B-marked. 10 Actually, Vallduvı´ and Vilkuna refer to it as kontrast, to distinguish it from other notions of contrast. 11 Vallduvı´ and Vilkuna believe that only topics, or, in their terms, links can exhibit this kind of contrast, whereas, as discussed above, I think it can also apply to tails, i.e. non-topical ground.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Bu¨ring applies his theory to elements that are pronounced with a Baccent, which he claims are topics. Whether B-accented elements are really topics is controversial, however. Jackendoff (1972) agrees that they have the information status of topics, but Portner & Yabushita (1998) and Kadmon (2001) believe they are foci, and Krifka (1998) claims they have both topical and focal features. Indeed, if one takes the view (as I do) that topics are what the sentence is about (Reinhart 1981), then topics must be referential. But non-referential elements may be expressed with a B-accent, as can be seen in Bu¨ring’s own examples:
Ariel Cohen 9
(15) a. The boys were eating ice cream. They were happy. b. The boys met the girls at the ice cream parlor. They were happy. However, there are cases when the antecedent does not denote a subset of the reference of the anaphor; Hendriks and Dekker call this type of anaphora non-monotonic anaphora, and argue that this is what is expressed by the fall–rise contour. For example: (16) John fed the animals. The cats were hungry. Normally, we would interpret the sentence as an expression of monotonic anaphora: the cats and the animals refer to the same set of individuals (in context), which means that all the animals fed by John were cats. However, Hendriks and Dekker point out that if cats is uttered with a fall–rise intonation, we get a different interpretation: the set of cats does not contain the set of animals (in context), so that John fed some animals that were not cats. In many cases, non-monotonic anaphora expresses contrast; hence, one might get the impression that B-marked elements are associated with the L+H* contour. However, non-monotonic anaphora do not always express contrast, hence not all L+H* accents are cases of contrast; and, I would like to add, not all cases of contrast are nonmonotonic anaphora. Indeed, in section 3 we will see examples of elements that require contrastive contexts, and are not foci, so they are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Since it is not clear that B-marked elements are topics, but it is clear that they are contrastive, I will refer to Bu¨ring’s semantic value as the contrast semantic value, rather than the topic semantic value. As with the case of focus, an important question is the relation between B-marking and intonation. Are B-marked elements always associated with a fall–rise contour? Here, again, opinion is divided. Some would say there is a tight connection, while others would deny this, or would claim such a connection only for some cases of B-marked elements. In this context, it is interesting to consider the proposal made by Hendriks & Dekker (1996). Usually, when a discourse referent Y is anaphoric to an antecedent X, they have the same reference, as in (15a): both antecedent and anaphor refer to the set of boys. Sometimes, the reference of an antecedent is a subset of the reference of the anaphor, as in (15b), where the plural pronoun they has two antecedents, the boys and the girls, each of which denotes a subset of the reference of the pronoun.
10 No Alternative to Alternatives B-marked. Yet, they do not indicate non-monotonic anaphora; hence, they are not required to exhibit the fall–rise contour. Since B-marking is not always associated with a specific contour, I will treat B-marking, just like focus marking, as a feature; its meaning is an indication of contrast, but I will leave aside its phonological realization.
2.4 B semantic value
(17)
Semantic Value
Alternatives to F?
Alternatives to B?
½½uO ½½uF ½½uB+F ?
No Yes Yes No
No No Yes Yes
The question mark in the table begs to be replaced with yet a fourth type of semantic value: the B semantic value—½½uB. This semantic value is obtained by replacing B-marked elements with alternatives.12 Some evidence that the B semantic value is necessary comes from Artstein (2002). He works on echo questions, as in: (18) Amy: I gave flowers to George. Bill: You gave FLOWERS to George? According to Artstein, the word in capitals (FLOWERS) is expressed with a B-accent. The normal interpretation of such an echo question is that Bill misheard or misunderstood what Amy was saying, and is seeking clarification. Artstein notes that Bill’s question means roughly the same as: (19) You gave WHAT to George? He therefore concludes that the meaning of both is a set of propositions of the form ‘Amy gave x to George’: It should be emphasized that ½½uB+F is not derived compositionally from ½½uB and ½½uF. The notation merely indicates that the contrast semantic value is sensitive to both the focus and the background. 12
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us take stock. We have three types of semantic value: the ordinary semantic value, which takes no alternatives into consideration; the focus semantic value, which considers alternatives to the focus, and the contrast semantic value, which considers alternatives to both the focus and the B-marked element. The types of semantic value can be described in a table:
Ariel Cohen 11
(20) fAmy gave flowers to George, Amy gave chocolate to George, Amy gave wine to George, . . .g
3 RELATIVE READINGS In this section, I briefly discuss the theory of relative readings I have proposed in Cohen (2001b).
3.1 Many It is widely accepted that many and few are ambiguous between cardinal and proportional readings (Partee 1988).14 For example, (21) is ambiguous. (21) Many kids attended the faculty picnic. Under the cardinal reading, (21) simply means that the number of kids at the picnic was high (relative to some standard). But (21) also has a proportional reading, according to which it says that a large proportion of all kids attended the picnic. Thus, if n kids were at the picnic, and n is considered high, then (21) would be true under the cardinal reading. But if n kids constitute only a small percentage of the total number of contextually relevant kids, (21) would be false under the proportional reading. Besides the cardinal and proportional readings, Westersta˚hl (1985) claims there is an additional reading of many and few, his ½½many4,k (p. 404; the ‘4’ indicates his fourth formalization of many, and the ‘k’ is 13 I will return to Artstein’s theory in section 5.4 below, where it will be further argued why the accented expressions are not focused. 14 It has, in fact, been suggested that they have additional readings—see Lappin (2000) for an overview. Lappin proposes dealing with all these readings using a parameterized representation. I will return to this issue in section 5 below.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Artstein derives this interpretation by treating B-accent as a case of focus. Thus, (20) would be the focus semantic value of Bill’s utterance. But, as we have seen above, there are reasons to doubt that B-accent indicates focus.13 Hence, it would seem more plausible to treat Baccent as a case of B-marking; in this case, (20) would be the B semantic value of Bill’s utterance, and we would get the same result. But is there additional evidence that the B and contrast semantic values are necessary? Specifically, do they affect truth conditions? In the rest of the paper, I will argue that the answer is yes. I will discuss a class of interpretations of some quantified sentences that, I claim, provide just this kind of evidence.
12 No Alternative to Alternatives a parameter indicating what proportion counts as many). He produces the following example: (22) Many Scandinavians have won the Nobel Prize in literature.
(23) Many Scandinavians who have distinguished themselves in some fashion, have won the Nobel Prize in literature. What it means to distinguish oneself is of course quite vague, but it is unlikely that the number of Scandinavians who have done so is so small as to make 14 out of them a high proportion. Consequently, I concur with Westersta˚hl that the sentence is true under neither the cardinal nor the proportional reading; its apparent truth, then, is in need of explanation. The reading Westersta˚hl proposes may be paraphrased as follows: (24) Many of the winners of the Nobel Prize in literature were Scandinavians. This interpretation is the proportional reading, but with the order of arguments reversed; I will call this view, namely that the normal order of arguments dictated by the syntax may be reversed, the Reverse Interpretation view. Alternatively, de Hoop & Sola` (1995) propose that the reading exemplified by (22) is just the usual cardinal reading. It is the context which determines what number of Scandinavians who have won the Nobel Prize in literature would be considered ‘many’. Both these approaches suffer from serious problems—I will briefly go over a few of them.15 For either view, the number of Scandinavians who have won the Nobel Prize in literature affects the truth of (22). It 15
See Cohen (2001b) for an extensive discussion.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As of 1984, out of a total of 81 winners of the Nobel Prize in literature, 14 came from Scandinavia. Given this fact, Westersta˚hl judges (22) to be true, though, he claims, it would be false under either the proportional or the cardinal reading. Note that this interpretation cannot be explained away as a case of domain restriction. As mentioned above, the domain of a determiner is restricted by the disjunction (union) of the focus semantic value of its scope (Geilfuß 1993; de Hoop & Sola` 1995). One may suppose that, if we take this fact into account, (22) will turn out to be true under the proportional reading after all. But this does not seem to be the case. If we assume that the VP in (22) is focused and that it induces alternative forms of distinguishing oneself, (22) would mean something like:
Ariel Cohen 13
does so either directly (de Hoop and Sola`) or after being divided by the number of Nobel Prize winners in literature (Westersta˚hl). In contrast, according to both approaches, the total number of Scandinavians does not affect the truth of the sentences in any way. However, this does not seem to be correct, as can be seen in (25). (25) Many Andorrans have won the Nobel Prize in literature.
(26) a. Most/all/no/some alligators like to sunbathe. b. Most/all/no/some alligators are alligators that like to sunbathe. To say that Q is conservative is to say that when evaluating Q(w, /) we only count ws, never non-ws. For example, to determine the truth of (26a), it only matters how many alligators like to sunbathe, and no fact about non-alligators is relevant. The cardinality interpretation of de Hoop and Sola` implies that many, under the reading exemplified by (22), ought to be conservative with respect to its complement; under the Reverse Interpretation view, it ought to be conservative too, but with respect to the VP. It turns out, however, that the reading of many under discussion is not conservative with respect to either the complement or the VP. If many were conservative with respect to the first argument, (22) would be equivalent to (27), but this is clearly not the case. (27) Many Scandinavians are Scandinavians who have won the Nobel Prize in literature. For Westersta˚hl, many ought to be conservative with respect to the VP. So, (22) ought to be equivalent to (28), but, in fact, it is not.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It may be sufficient for as few as two or three Andorrans to have won the Nobel Prize in literature for (25) to be true, but such a small number would not be enough for the truth of (22). Intuitively, the reason for this judgment is that there are many more Scandinavians than Andorrans. Note that this judgment cannot be explained by either the cardinal or the proportional reading of many. According to the cardinal reading, if two or three Scandinavians are few, so should two or three Andorrans be. Turning to the proportional view, if we divide 2 or 3 by the total number of Andorrans (about 60,000) we get an extremely small number, which can hardly be considered ‘many’. Perhaps the reason for the problematic nature of examples like (22) is that they fail to exhibit conservativity. A generalized determiner Q is conservative iff, for all properties w and /, Q(w, /) is equivalent to Q(w, w \ /). For example, (26a) is equivalent to (26b).
14 No Alternative to Alternatives (28) Many Scandinavians who have won the Nobel Prize in literature, have won the Nobel Prize in literature.
16
In fact, this is very similar to yet another reading of many proposed by Westersta˚hl (1985), his ½½manyM 2 (the ‘2’ indicates his second formalization of many and ‘M’ is the universe of discourse). According to this interpretation, Many As are Bs means that ‘the frequency of Bs on A is greater than a certain normal frequency of Bs . . . we let the normal frequency of Bs simply be the frequency of Bs in the universe M’ (p. 402, original emphasis). Thus, according to this reading, given a universe M, j/j many(w, u) is true iff jw\/j jwj > jMj .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We have seen that, when evaluating many(w, /) in sentences such as (22), the number of ws affects the truth value of the sentence. Of course, the number of ws that are /s also matters. It appears, therefore, that we are dealing with some sort of proportional reading. We have also seen that this interpretation of many, unlike the standard proportional reading, is not conservative. This means that it is not sufficient to look only at ws, we need to consider non-ws too. I have therefore suggested that (22) expresses a comparison between various countries with respect to the proportion of the population who have won the Nobel Prize in literature. The sentence is true just in case this proportion is greater in Scandinavia than it is in general in the world. This does not require that the proportion of Nobel Prize winners be greater in Scandinavia than in every other country, but it does require that it be greater than the proportion among the world population in general.16 I refer to this interpretation of many as its relative proportional reading, since according to it, the meaning of many(w, /) is defined relative to the properties of non-ws; I refer to the standard proportional reading as the absolute proportional reading. Crucial to this interpretation of many is the idea that alternatives to Scandinavians are considered. Does this mean that Scandinavians is focused? Herburger (1997) believes that it does. She follows the Reverse Interpretation view for sentences like (22), but claims that this reading is only possible when Scandinavians is focused (see also Eckardt 1994). Herburger does not, however, explain how focus is expressed in her examples. How does she know that Scandinavians is focused in (22)? Presumably, Herburger feels that the noun carries some pitch accent, but it is not clear that this pitch accent is even necessary for this interpretation (Westersta˚hl does not mention any particular intonation pattern), and even if it is, what is its exact nature. Since pitch accent is, as mentioned above, neither necessary nor sufficient to determine whether an element is focused, in order to be clear about the information status of Scandinavians, we need to turn to pragmatics, rather than phonology. Note that, under the interpretation in question, (22) seems to be licensed only in a contrastive context; out of the blue, this sentence,
Ariel Cohen 15
with the reading under discussion, is actually rather odd. But in the context of the exchange in (29), it is perfectly acceptable. (29) Amy: Have many Israelis won the Nobel Prize in literature? Bill: No, many Scandinavians have won the Nobel Prize in literature.
(30) a. #The King of France is bald. b. The exhibition was visited by the King of France. His explanation is that the King of France is the topic of (30a), but not (30b). Now examine the following exchange: (31) Amy: The President of France is bald. Bill: #No, the King of France is bald. Bill’s utterance in (31) is odd, indicating that the King of France is the topic of this sentence, even when there is pitch accent on the subject. Now consider: (32) #Many Martians have won the Nobel Prize in literature. Assuming there are no Martians, (32) is odd, indicating that it carries a presupposition that Martians exist. Note that this would be the case even when we stress Martians. The descriptive content of the complement of many, then, is presupposed, even when stressed. Hence, we can conclude that the complement of many is a topic, rather than a focus.17 17 Of course, there are cases, such as (i), suggested to me by an anonymous referee, where the whole DP may be focused.
(i)
The exhibition was visited by many Scandinavians.
It is hard to judge whether (i) has a relative reading, nor is it clear what the information status of Scandinavians is (focus? subordinate topic? cf. Erteschik-Shir 1997). But the point is that for Herburger’s theory to work, the complement of many must be a focus in non-controversial cases like (22), which, if I am correct, it is not.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The complement of many, then, is interpreted contrastively. Foci may, indeed, be contrastive, but topics may be contrastive too. Is Scandinavians a focus or a topic? In order to answer this question, additional tests are needed. The clearest way, perhaps, to test for whether a phrase is a topic is to check whether its descriptive content is presupposed (Strawson 1964; Reinhart 1981). Strawson notes that while the classic (30a) presupposes the existence of a king of France, and is therefore odd, (30b) is straightforwardly false.
16 No Alternative to Alternatives In contrast, Herburger would predict that the descriptive content of the VP, rather than the subject, is presupposed, since proportional quantifiers presuppose the non-emptiness of their restrictor, and the VP is claimed to be mapped onto the restrictor. However, (33) is perfectly fine (though false), even though there is no Nobel Prize in silly walks: (33) Many Scandinavians have won the Nobel Prize in silly walks.
(34) a. Russia has the greatest number of scientists in the world, but . . . b. . . . few of the people in Russia are scientists. c. . . . #few SCIENTISTS are in Russia. We can conclude, then, that the complement of the determiner in (22), regardless of accent, is not a focus, as Herburger claims, but rather a (contrastive) topic. I will leave a discussion of the nature of contrastive topics for another occasion; it suffices for current purposes to note that a contrastive topic is contrastive, but is not a focus; hence, it must be B-marked. Scandinavians, then, is a B-marked element in (22). Note that intonation alone would not have been sufficient to determine this, since it need not be expressed with a fall–rise contour, because it does not involve non-monotonic anaphora (cf. section 2.3 above).
3.2 Often The phenomenon of relative readings is not only restricted to many and few but also extended to their adverbial counterparts: often and seldom, and their synonyms. The following example is due to de Swart (1991): (35) Paul often has a headache. De Swart observes that (35) has a reading under which in many of the contextually relevant situations, Paul has a headache. This
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Additional evidence that the complement of many in sentences such as (22) is actually a topic comes from discourses such as (34) (from de Hoop & Sola` 1995; my explanation, however, differs from theirs). Sentence (34a) establishes Russia as a topic; since (34b) is about the people in Russia, it follows felicitously. According to the Reverse Interpretation view, (34b) ought to mean the same as (34c), which consequently ought to be fine too, but the fact is that (34c) is distinctly odd in this context. The reason is that the topic of (34c) is actually scientists, which is not the topic established by (34a).
Ariel Cohen 17
interpretation of often corresponds to the absolute proportional reading of many. But de Swart notes an additional reading: ‘The sentence can also be taken to mean that the situations of Paul having a headache occur with a frequency superior to the average’ (p. 21). This sounds very much like the relative proportional reading of many: Paul has a headache more frequently than the average just in case he is more likely to have a headache than an arbitrary person is. Relative readings obtain also with the atemporal use of Q-adverbs: (36) A politician is often crooked.
(37) Amy: The main suspects are a politician, a physician, and a linguist. Who do you think did it? Bill: Well, a politician is often crooked. As the context of (37) indicates, a politician is contrastive and belongs to the background; hence, it is B-marked. Note that if the background is clearly delimited and does not include the subject, the relative reading disappears. For example, fronting the Q-adverb has precisely this effect (Cohen 2004b): (38) a. Often, Paul has a headache. b. Often, a politician is crooked. Sentence (38a) can only mean that there are many situations where Paul has a headache, not that Paul has headaches more frequently than the average; (38b) can only mean that many politicians are crooked, not that they are more likely to be crooked than other people are. Note that in (37) focus is on crooked, not politician. In fact, if we force politician to be focused, by stressing it and uttering the sentence in a context in which politicians are not part of the background, the result is quite bad: (39) Amy: We live in bad times—there is corruption everywhere. Bill: ??A [politician]F is often crooked.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Sentence (36) is ambiguous: on one reading, the absolute reading, it would be true just in case many politicians are crooked; under this reading, (36) is false. However, we may feel uncomfortable declaring unequivocally that the sentence is false. This, I suggest, is because (36) has another reading, namely that a politician is more likely to be crooked than an arbitrary person is. Under this reading, (36) may, to our misfortune, be true. The right context is important, perhaps crucial, for obtaining the relative reading:
18 No Alternative to Alternatives To the extent that Bill’s utterance in (39) is interpretable at all, it means something very different from the relative reading: namely, that many crooked individuals are politicians. This fact indicates that accounting for relative readings in terms of the Reverse Interpretation view is incorrect in the case of Q-adverbs, just as it is incorrect in the case of many and few. Similar points can be made about applying the cardinality view of de Hoop & Sola` (1995) to often.18
3.3 Generics Consider the following examples:
It is not the case that Frenchmen, in general, eat horsemeat, and yet (40a) is true. Very few Bulgarians are good weightlifters, and yet (40b) is true. How can the truth of (40a) and (40b) be accounted for? This type of sentence poses a challenge to any theory of generics. Take, for example, the commonly held view that generics express modalized universal quantification, restricted by context (Chierchia 1995; Krifka 1995; Pelletier & Asher 1997, among many others). According to this view, the generic (41a) means roughly (41b). (41) a. As are Bs b. All relevant normal As are Bs. Thus, (40a) would be true just in case all relevant normal Frenchmen eat horsemeat. But it is hard to see any plausible restriction of the domain of Frenchmen that would include only the horsemeat eaters and would filter out all others as irrelevant or abnormal. Similarly, (40b) would mean that all relevant normal Bulgarians are good weightlifters, and its truth would require that Bulgarians who are not good weightlifters be considered irrelevant or abnormal. Thus, examples like (40) pose a significant problem to the view of generics as contextually restricted modalized universals. Some researchers, most notably Carlson (1977), have used such sentences to argue that generics are not quantificational at all. Yet, a non-quantificational account of generics would leave their truth conditions stipulated, rather than explained. Moreover, it would be difficult to account for phenomena such as scope ambiguity and quantificational variability, as Carlson himself admits in a later article (Carlson 1989). 18
In fact, this is the way in which de Swart attempts to explain the relative reading of (35).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(40) a. The Frenchman eats horsemeat. b. Bulgarians are good weightlifters.
Ariel Cohen 19
Others, most notably Krifka et al. (1995), have sought to maintain the advantages of a quantificational account of generics in general, but deny such an account to examples like the sentences in (40). According to their view, such sentences express direct kind predication and are not quantificational. This is rather a disappointing move, in that it treats the truth conditions of such generics as primitive and leaves no room for explaining them in terms of properties of arbitrary individuals. Moreover, the same arguments favoring a quantificational account of generics can apply to such examples too. For example, (42) expresses a scope ambiguity: An alternative approach, proposed, most explicitly, by Wilkinson (1991), is to keep the quantificational view of generics, but apply the Reverse Interpretation view to this particular type of example. According to this approach, the meanings of the sentences in (40) can be paraphrased as follows: (43) a. Horsemeat eaters are Frenchmen. b. Good weightlifters are Bulgarian. In other words, (40a) is a generic statement about horsemeat eaters, rather than Frenchmen, and (40b) is about good weightlifters, rather than Bulgarians. However, just as with many and often, the Reverse Interpretation view is problematic, and for similar reasons. One problem is that the paraphrases in (43) fail to capture the truth conditions of the sentences in (40). Suppose that most horsemeat eaters were actually, say, Belgian, and that most good weightlifters were Russian. The sentences in (40) might still be true, but those in (43) would definitely be false. An additional difficulty for the Reverse Interpretation view comes when conservativity is taken into account. Examples like those in (40) fail to exhibit conservativity: (44) a. The Frenchman is a Frenchman who eats horsemeat. b. Bulgarians are Bulgarians who are good weightlifters. The sentences in (44) are not paraphrases of their counterparts in (40); the former are false, whereas the latter are true. The Reverse Interpretation view implies that the generic quantifier is conservative with respect to the VP, i.e. with respect to eat horsemeat for (40a) and be a good weightlifter for (40b). However, this does not seem to be correct, as (45a) and (45b) are not equivalent to (40a) and (40b), respectively, and are, in fact, necessarily true.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(42) Bulgarians are good at some sport.
20 No Alternative to Alternatives (45) a. Frenchmen who eat horsemeat eat horsemeat. b. Good Bulgarian weightlifters are good weightlifters. An additional problem is that the sentences in (40) definitely seem to be about Frenchmen and Bulgarians, respectively. Standard tests for topicality confirm this intuition: (46) a. As for the Frenchman, he eats horsemeat. b. As for Bulgarians, they are good weightlifters. The same test shows that, in contrast, the topics of the sentences in (43) are horsemeat eaters and good weightlifters, respectively:
If topics are mapped onto the restrictor, it follows that the Frenchman and Bulgarians are mapped onto the restrictor of the generic, rather than the nuclear scope, contradicting the Reverse Interpretation view. Based on the arguments above, I conclude that the Reverse Interpretation view is wrong. And yet, it cannot be denied that this view has a certain initial appeal. Why is this? I submit that the reason is that although the Frenchman is the topic of (40a), this sentence would not really be felicitous in the context of a discussion of French people. Instead, it would be more appropriate in the context of a discussion of various nationalities and their eating habits. Similarly, although Bulgarians is the topic of (40b), the sentence would not normally be said in the context of a description of Bulgarians. Rather, such a sentence might be said as part of a conversation about weightlifting, or perhaps sports in general, comparing the accomplishments of people from various nationalities. In other words, the subjects of the sentences in (40) are contrastive. Since they are topics, hence part of the background, they are B-marked. The sentences in (40), I propose, are evaluated with respect to alternative nationalities, and receive a relative reading. Sentence (40a) would be true just in case a Frenchman is more likely to eat horsemeat than a person of arbitrary alternative nationality is. Note that this might still hold if few Frenchmen eat horsemeat, or if the majority of horsemeat eaters are, say, Belgian. Similarly, (40b) is true since a Bulgarian weightlifter is more likely to be a good one than a weightlifter of some arbitrary nationality is. Again, this would be true even if a good weightlifter were more likely to be, say, Russian, rather than Bulgarian. Relative readings of generics are more common than may seem at first sight. Indeed, many naturally occurring generics are only true if given a relative interpretation. Consider (48), for example:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(47) a. As for horsemeat eaters, they are Frenchmen. b. As for good weightlifters, they are Bulgarian.
Ariel Cohen 21
(48) Tigers eat people. This sentence is false under the absolute reading: it is not the case that, in general, tigers eat people—very few do, in fact. Nor is it true that, in general, people who are eaten by some animal are eaten by tigers. However, (48) is true under the relative interpretation, since a tiger is more likely to eat people than an arbitrary animal is. In section 4.3, I will demonstrate that if we replace the view of generics as contextually restricted modalized universals with a theory that takes them to express probability judgments, the relative readings receive a straightforward interpretation.
Relative readings require an explanation; preferably, the explanation should be the same for all three constructions, and independently motivated. In this section, I intend to demonstrate that the relative readings of all three constructions can be accounted for in a general way using the four semantic values proposed in section 2. The basic idea is this:19 in both the absolute and relative reading, we look at a certain proportion (or probability), call it P. The absolute reading simply requires that P be high, whereas according to the relative reading, we compare P with a similar proportion (or probability) that is obtained after taking into account alternatives to the B-marked element.
4.1 Many Let us repeat (22), annotated for focus marking and B-marking: (49) Many [Scandinavians]B have [won the Nobel Prize in literature]F. I am assuming that many is a relation between two sets. While there are reasons to believe that many is intensional (Keenan & Stavi 1986; Kamp & Reyle 1993), the intensions of the related properties only affect the threshold for being many, not the arguments of many themselves (Fernando & Kamp 1996). We will return to the intensionality of many in section 5.1 below. The logical form of (49), then, is something like the following (where scand is the property of being a Scandinavian, and wnpil is the property of winning a Nobel Prize in literature): (50) many(scand, wnpil) 19
I thank Sigrid Beck (personal communication) for her help in formulating the idea.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4 A MULTIDIMENSIONAL ACCOUNT OF RELATIVE READINGS
22 No Alternative to Alternatives Under the absolute (proportional) reading, (49) is true iff the proportion of literature Nobel Prize winners among Scandinavians is high. This can be formalized as follows:
j½½scandO \ ½½wnpilO j j [ ½½scandB \ ½½wnpilO j > (52) j [ ½½scandB j j½½scandO j If the alternatives to scand consist of all nationalities, its union will simply be PEOPLE, the set of all people. The truth conditions then become: j½½scandO \ ½½wnpilO j jPEOPLE \ ½½wnpilO j > jPEOPLEj j½½scandO j In words, the proportion of literature Nobel Prize winners among Scandinavians is higher than the proportion of literature Nobel Prize winners among people in general. This is the desired interpretation. We can now make a first attempt at a definition of relative readings:
(53)
Definition 1 (many, relative reading, first version) many(w, u) is true iff
j½½wO \ ½½/O j j [ ½½wB \ ½½/O j . > j [ ½½wB j j½½wO j
One potential problem with this definition is that there are some reasons to believe that the first argument of the quantifier leaves, at LF, a copy at the second argument position.20 This means that w appears 20
This is argued for by Fox (1999: p. 183): A sentence such as [(i)] has [(ii)] as the output of QR, which can in turn be converted to one of the structures in [(iii)]. Economy principles determine that the interpreted structure is [(iii.b)]. (i) John1 [VP t1 likes every boy]. (ii) John1 [every boy [VP t1 likes every boy (iii) a. *John1 [every boyx [VP t1 likes x (ruled out by economy) b. John1 [every boyx [VP t1 likes boy x
Thus, in [(iii.b)], the logical form of [(i)], the predicate boy appears in both restrictor and nuclear scope. Presumably, the same considerations apply to other quantifiers, and to many in particular.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
j½½scandO \ ½½wnpilO j > q; where q is ‘large’. (51) j½½scandO j What about the relative reading? Again we are talking about the proportion of literature Nobel Prize winners among Scandinavians. But this time, instead of simply saying that it is high, we will compare it to the proportion of literature Nobel Prize winners among all people who belong to a nationality that is an alternative to scand. This set of people is the union of the B semantic value of the subject. We get the following truth conditions:
Ariel Cohen 23
not only as the first argument of many but also as part of the second argument. If this is the case, definition 1 would yield the wrong result: we would be comparing the proportion of Scandinavian winners among Scandinavians to the proportion of Scandinavian winners (rather than winners in general) among people in general. The sentence would therefore be erroneously predicted trivially true. In order to avoid this problem, let us take the union of the B semantic value of u as well.21 Here is the revised definition: Definition 2 (many, relative reading, second version) j½½wO \ ½½/O j j [ ½½wB \ [ ½½/B j > . j [ ½½wB j j½½wO j
Definition 2 is an improvement, but it is still not quite right. The reason is that it does not take focus into account; yet, as we have seen above, the union of the focus semantic value restricts the domain of quantification of a determiner. For example, in order to get the reading paraphrased by (23) above, we need to consider not simply the proportion of winners among Scandinavians, but among Scandinavians who have distinguished themselves. In general, when we evaluate many(w, u) (either absolutely or relatively), we need to consider not only j½½wO \ ½½/O j , j½½wO j but also
(54)
j½½wO \ [ ½½/F \ ½½/O j (55) . j½½wO \ [ ½½/F j For simplicity, let us assume that, in general, an expression is always a member of the set of its own focus semantic value.22 Then, [ ½½uF \ ½½uO is equivalent to ½½uO, and we consider (56)
j½½wO \ ½½/O j . j½½wO \ [ ½½/F j
For the absolute reading, (56) ought to be ‘large’. What about the relative reading? As before, we would like to take the unions of the B semantic values of the elements in (56) and place the result in the righthand side. However, since the denominator of (56) contains not only 21 22
Of course, if Fox’s theory is wrong, no harm will be done by this move. Though this is not always the case (Cohen 1999b).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
many(w, u) is true iff
24 No Alternative to Alternatives the ordinary semantic value of u but also the union of its focus semantic value, we need to take, so to speak, the B semantic values of this union: which means the union of the union of the contrast semantic value of u. The result is this: j [ ½½wB \ [ ½½/B j . (57) j [ ½½wB \ [ [ ½½/B+F j The general definition of the relative reading of many, then, is: Definition 3 (many, relative reading, third version) j½½wO \ ½½/O j j [ ½½wB \ [ ½½/B j . > j½½wO \ [ ½½/F j j [ ½½wB \ [ [ ½½/B+F j
Applying this definition, (49) would be true iff j½½scandO \ ½½wnpilO j j [ ½½scandB \ [ ½½wnpilB j (58) > j½½scandO \ [ ½½wnpilF j j [ ½½scandB \ [ [ ½½wnpilB+F j In order to evaluate this inequality, let us write ½½scandO ¼ SCAND and ½½wnpilO ¼ WNPIL. Let us assume that the alternatives to winning the Nobel Prize in literature are the properties of winning various distinguished prizes. Thus, (59) [ ½½wnpilF ¼ PRIZE-WINNERS, where PRIZE-WINNERS is the set of people who won some distinguished prize. Let us assume that, as before, (60) [ ½½scandB ¼ PEOPLE. Since the property wnpil is not B-marked, (61) [ ½½wnpilB ¼ WNPIL, and (62) [ [ ½½wnpilB+F ¼ [ ½½wnpilF ¼ PRIZE-WINNERS. Now, (49) is true iff jSCAND \ WNPILj
jPEOPLE \ WNPILj
(63) jSCAND \ PRIZEWINNERSj > jPEOPLE \ PRIZEWINNERSj In words, the proportion of distinguished Scandinavians who have won the Nobel Prize in literature is greater than the corresponding proportion in the world at large. This appears to be the correct interpretation.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
many(w, u) is true iff
Ariel Cohen 25
4.2 Often
(64) a. Mary always takes [John]F to the movies. b. [Mary]F always takes John to the movies. c. always(ks.take(Mary, John, s)) 23
Herburger (1997) disagrees; see Cohen (2001b) for arguments against her view. Emphasizing, once again, that the focus semantic value is calculated by the rules of presupposition projection. 25 Krifka develops such a system, and then discards it in favor of a system that uses existential presupposition instead of sets of alternatives. 26 To be more precise, Q-adverbs apply to sets of minimal situations (Heim 1990; von Fintel 1994), but I will not endeavour to represent minimality in my logical forms. 24
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
For the semantics of often and other Q-adverbs, I will use probabilities rather than proportions, so as to reflect their lawlike nature (Cohen 1999a). Readers who are uncomfortable with probabilities may wish to continue to think in terms of proportions; the analysis presented here will work just as well with proportions. Unlike the case of many, where syntax determines which elements go into the restrictor,23 in the case of often, syntax is not sufficient. I will therefore assume that the Q-adverb has sentential scope, and that the restriction is determined by focus. More precisely, I will follow Rooth (1985) by assuming that the union of the focus semantic value restricts the domain of a Q-adverb.24 In a probabilistic semantics, this means that the union of the focus semantic value forms the reference class of the conditional probability. Before we can discuss the logical forms of sentence involving often, we first need to determine what the Q-adverb quantifies over. One possibility is to take often (and other Q-adverbs) to quantify over cases (in the sense of Lewis 1975). It turns out, however, that it would be rather complex to provide a formal specification of focus semantic values of the resulting logical forms (though quite possible; see Rooth 1995 and Krifka 200125 for examples of such systems). Specifying the other types of semantic value would presumably be even more complex. Therefore, so as not to be distracted from the main point of the paper by overly complex formulations, I will assume that Q-adverbs quantify over situations. That is to say, propositions denote sets of situations, and therefore Q-adverbs, whose scope is a proposition, apply to sets of situations.26 Thus, for example, the logical form of both (64a) and (64b) is (64c); the difference in their truth conditions will result from different values assigned to the restrictor, as a consequence of the difference in their respective focus structures.
26 No Alternative to Alternatives Note that (64c) contains only a nuclear scope, not a restrictor; the value of the restrictor is provided by the union of the set of alternatives.27 How do we get the different interpretations of (64a) and (64b)? In a probabilistic semantics, always(/) is true iff the probability of /, given the restrictor, is 1: (65) P(½½/Oj [ ½½/F) ¼ 1 In the case of (64a), ½½/F is a set of propositions, namely a set of sets of situations. In each set of situations, Mary takes a different individual (e.g. John, Bill or Fred) to the movies:
The union of the focus semantic value is then: (67) fs j dx take(Mary, x, s)g Thus, (64a) is true iff (68) P(fs j take(Mary, John, s)gjfs j dx take(Mary, x, s)g) ¼ 1 This means that (64a) is true iff the probability that Mary is taking John to the movies, given that she is taking someone to the movies, is 1. This appears to be the correct interpretation. In (64b), ½½/F is different. Consequently, we compute different truth conditions: (69) P(fs j take(Mary, John, s)gjfs j dy take(y, John, s)g) ¼ 1 Thus, (64b) is true iff the probability that Mary is taking John to the movies, given that someone is taking him to the movies, is 1—the desired interpretation. Let us now turn to often. Under its absolute reading, we take some expression, /, and consider the probability of / given the union of the alternatives to the focused elements of /. This can be defined formally as follows: Definition 4 (often, absolute reading, first version) often(/) is true iff P(½½/Oj [ ½½/F) > q, where q is ‘large’. The account of relative readings is the same, except that the value of q is given a definite value: the same as the left-hand side, but with alternatives to the B-marked element considered. As with the absolute 27 Equivalently, we could maintain the appearance of a tripartite structure, by following authors such as Rooth (1992) and von Fintel (1994, 2004) and adding an empty restrictor, indicated by a free variable.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(66) ffs j take(Mary, John, s)g, fs j take(Mary, Bill, s)g, fs j take(Mary, Fred, s)g, . . .g
Ariel Cohen 27
reading, we take some expression /#, and consider its probability given the union of the alternatives to the focused elements of /#. In the case of the absolute reading, /# was simply / itself, but for the relative reading it is the union of the B semantic value of /. This means that we consider the probability of [ ½½/B given [ [ ½½/B+F. Hence, the definition of the relative reading is as follows: Definition 5 (often, relative reading) often(/) is true iff P(½½/Oj [ ½½/F) > P([ ½½/Bj [ [ ½½/B+F). We can now see how these definitions work. Consider again Bill’s answer in (37), repeated below, with B-marking and focus marking: Its logical form would be (where pl is the property of being a politician, and crkd is the property of being crooked): (71) often(ks.dx(in(x, s) ^ pl(x) ^ crkd(x))) Let us indicate the nuclear scope by u. Then the ordinary semantic value is: (72) ½½/O ¼ fs j dx(in(x, s) ^ pl(x) ^ crkd(x))g With regard to the focus semantic value, it is important to identify where the focus is. The adverb may be stressed, but since Q-adverbs require focus (Cohen 2004a), there must be (possibly second occurrence) focus inside the nuclear scope; since pl is B-marked, the only possibility is focus on crkd. Assuming the only alternative to crkd is honest, the focus semantic value is: (73) ½½/F ¼ ffs j dx(in(x, s) ^ pl(x) ^ crkd(x))g, fs j dx(in(x, s) ^ pl(x) ^ honest(x))gg The union of the focus semantic value is then simply (74) [ ½½/F ¼ fs j dx(in(x, s) ^ pl(x))g Assuming that the alternatives to pl are physician and linguist, the B semantic value is: (75) ½½/B ¼ ffs j dx(in(x, s) ^ pl(x) ^ crkd(x))g, fs j dx(in(x, s) ^ physician(x) ^ crkd(x))g, fs j dx(in(x, s) ^ linguist(x) ^ crkd(x))gg Its union is simply (76) [ ½½/B ¼ fs j dx(in(x, s) ^ crkd(x))g
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(70) A [politician]B is often [crooked]F.
28 No Alternative to Alternatives The contrast semantic value is: (77) ½½/B+F ¼ fffs j dx(in(x, s) ^ pl(x) ^ crkd(x))g, fs j dx(in(x, s) ^ pl(x) ^ honest(x))gg, ffs j dx(in(x, s) ^ physician(x) ^ crkd(x))g, fs j dx(in(x, s) ^ physician(x) ^ honest(x))gg, ffs j dx(in(x, s) ^ linguist(x) ^ crkd(x))g, fs j dx(in(x, s) ^ linguist(x) ^ honest(x))ggg This corresponds to the following set of questions:
These questions are implied by the context of (37), hence the felicity of the exchange. The union of the union of the contrast semantic value is just (79) [ [ ½½/B+F ¼ fs j dx(in(x, s) ^ person(x))g We can now account for the two readings of (71). According to the absolute reading, often(/) is true iff P(½½/Oj [ ½½/F) is ‘large’. In this case, this means that (80) P(fs j dx(in(x, s) ^ pl(x) ^ crkd(x))gjfs j dx(in(x, s) ^ pl(x))g) > q, where q is ‘large’. In words, the sentence is true just in case a situation involving a politician is likely to be situation involving a crooked politician, as desired. Under the relative reading, often(/) is true iff (81) P(½½/Oj [ ½½/F) > P([ ½½/Bj [ [ ½½/B+F). In this case, this is the requirement that (82) P(fs j dx(in(x, s) ^ pl(x) ^ crkd(x))gjfs j dx(in(x, s) ^ pl(x))g)> P(fs j dx(in(x, s) ^ crkd(x))gjfs j dx(in(x, s) ^ person(x))g). In words, the sentence is true just in case a situation involving a politician is more likely to be a situation involving a crooked politician than a situation involving an arbitrary person is likely to be a situation involving a crook. This is the desired relative reading.
4.3 Generics I will assume that characterizing generics, like Q-adverbs, are quantificational. Following Krifka et al. (1995), I draw a distinction
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(78) fHow honest is a politician?, How honest is a physician?, How honest is a linguist?g
Ariel Cohen 29
between direct kind predication, as in (83a), and characterizing generics, exemplified by (83b). (83) a. Lions are an endangered species. b. Lions have a bushy tail.
Definition 6 (generics, absolute reading, first version) gen(w, u) is true iff P(½½uOj½½wO \ [ ½½uF) > 0.5. To obtain the relative reading, we will employ the same principle as with the other constructions above: the right-hand side of the inequality is the same as the left-hand side, except that we consider alternatives to the B-marked element. Definition 7 (generics, relative reading) gen(w, u) is true iff P(½½uOj½½wO \ [ ½½uF) > P([ ½½uBj [ ½½wB \ [ [ ½½uB+F). For example, consider (40b), repeated below, with focus and B marking: (84) [Bulgarians]B are [good]F weightlifters. Its logical form is (where bl is the property of being a Bulgarian and gwl indicates the property of being a good weightlifter): (85) gen(ks.dx(in(x, s) ^ bl(x)), ks.dx(in(x, s) ^ gwl(x)))
28 The idea that topics are mapped onto the restrictor of a quantifier is not at all new—see, among many others, Reinhart (1981) and Chierchia (1992). For a specific discussion of topicality and the generic quantifier, see Cohen (1996, 2001a) and Cohen & Erteschik-Shir (2002). 29 See section 5.3 below for the origins of this number. Some might say that 0.5 is too low, and does not capture the intuition that generics are true of ‘almost all’ instances. In Cohen (1996, 1999a), I suggest that generics are subject to additional constraints, whose effect is to create the appearance of a much higher threshold.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Sentence (83a) is predicational, expressing the statement that the kind Leo has a certain property. In contrast, (83b) is quantificational, expressing a statement about individual lions. Unlike Q-adverbs, the generic quantifier does not have sentential scope: the topic of the sentence is mapped onto the restrictor, and the rest of the sentence is mapped onto the scope.28 However, just like the case of Q-adverbs, the focus semantic value (as well as presuppositions) can further restrict the domain of the generic quantifier. I will assume that generics, again like Q-adverbs, express probability judgments (Cohen 1999a). Under the absolute reading, a true generic requires this probability to be greater than 0.5:29
30 No Alternative to Alternatives Let w be the restrictor and u the nuclear scope. Their respective ordinary semantic values are: (86) a. ½½wO ¼ fs j dx(in(x, s) ^ bl(x))g b. ½½uO ¼ fs j dx(in(x, s) ^ gwl(x))g Assuming, for simplicity, that the only alternative to gwl is bwl (the property of being a bad weightlifter), the focus semantic value of the nuclear scope is: (87) ½½uF ¼ ffs j dx(in(x, s) ^ gwl(x))g, fs j dx(in(x, s) ^ bwl(x))gg
(88) [ ½½uF ¼ fs j dx(in(x, s) ^ wl(x))g Since u does not contain B-marking, its B semantic value is just a singleton set of its ordinary semantic value, so that its union is just the ordinary semantic value: (89) a. ½½uB ¼ ffs j dx(in(x, s) ^ gwl(x))gg b. [ ½½uB ¼ fs j dx(in(x, s) ^ gwl(x))g The contrast semantic value is: (90) ½½uB+F ¼ fffs j dx(in(x, s) ^ gwl(x))g, fs j dx(in(x, s) ^ bwl(x))ggg The union of its union is: (91) [ [ ½½uB+F ¼ fs j dx(in(x, s) ^ wl(x))g The B semantic value of the restrictor is: (92) ½½wB ¼ ffs j dx(in(x, s) ^ bl(x))g, fs j dx(in(x, s) ^ Russian(x))g, . . .g Its union is simply: (93) [ ½½wB ¼ fs j dx(in(x, s) ^ person(x))g We can now account for the two readings of (84). The absolute reading will be true just in case (94) P(½½uOj½½wO \ [ ½½uF) > 0.5 Inserting the values for (84) into this inequality, we get: (95) P(fs j dx(in(x, s) ^ gwl(x))gjfs j dx(in(x, s) ^ bl(x) ^ wl(x))g) > 0.5.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Its union, then, will be (where wl is the property of being a weightlifter):
Ariel Cohen 31
This says that a situation involving a Bulgarian weightlifter is likely to be a situation involving a good weightlifter. Keeping in mind that the situations are minimal (see note 26 above), this is the desired interpretation. The relative reading will be true iff (96) P(½½uOj½½wO \ [ ½½uF) > P([ ½½uBj [ ½½wB \ [ [ ½½uB+F) In this case, it is the requirement that
In words, a (minimal) situation involving a Bulgarian weightlifter is more likely to be a (minimal) situation involving a good weightlifter than a (minimal) situation involving an arbitrary weightlifter is likely to be a (minimal) situation involving a good weightlifter. 5 THE ORIGINS OF RELATIVE READINGS We have seen that many (and few), often (and seldom) and generics have relative readings, and traced the differences in these readings to the value of the parameter q. The idea of accounting for different readings of many by different values of parameters is not new. Lappin (2000) proposes a parameterized representation and applies it with impressive success to a variety of interpretations of many. Lappin proposes some general, plausible constraints on the possible values of the parameters; however, he cannot specify why they have the particular values they do. Moreover, Lappin offers no explanation for why other quantifiers do not receive such parameterized representations. More to the point of the present paper, other quantifiers besides many, few, their adverbial counterparts and generics do not seem to have relative readings. Consider: (98) a. Most Scandinavians have won the Nobel Prize in literature. b. Paul usually has a headache. c. A politician is almost always crooked. Regardless of context and intonation, (98a) can only mean that the majority of (relevant) Scandinavians have won the Nobel Prize in literature, not that the proportion of Scandinavians who have done so is higher than that among people in general; (98b) means that a relevant situation containing Paul is likely to be a situation where he has a headache, not that Paul has a headache more frequently than the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(97) P(fs j dx(in(x, s) ^ gwl(x))gjfs j dx(in(x, s) ^ bl(x) ^ wl(x))g) > P(fs j dx(in(x, s) ^ gwl(x))gjfs j dx(in(x, s) ^ person(x) ^ wl(x))g)
32 No Alternative to Alternatives average and (98c) only means that if someone is a politician, he or she is almost certainly crooked, not that he or she is more likely to be crooked than an arbitrary person is. Why is it that only some quantifiers, but not others, give rise to relative readings? And how are their relative readings related to their absolute readings?
5.1 Many
(99) a. There were few faculty children at the 1980 picnic. b. Few egg-laying mammals turned up in our survey, perhaps because there are few. Sentence (99a) can be true even if all faculty children came to the picnic, provided that there were few faculty children in 1980. And (99b) will be true even if the survey included all egg-laying mammals. So, how many is many? Since the quantifier itself does not give us any clue, we need to find some strategy to determine if a given proportion, in a given context, qualifies. The absolute and relative readings turn out to be two strategies to accomplish this.30 Fernando and Kamp (1996) consider this problem and suggest that, under the absolute proportional reading,31 many(w, u) is true iff it could well have been the case that fewer ws are us. In other words, there are more ws that are us than expected. For example, (100) means that more academics watched the 2006 World Cup than one would expect of academics (who watch TV). (100) Many academics watched the 2006 World Cup on TV. Note than when we change the arguments of many, the expectations might change, and the threshold will change with them: (101) Many sports fans watched the 2006 World Cup on TV. 30 31
I thank Hans Kamp for suggesting this direction to me. They do not consider relative readings.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The key to the answer is, I believe, the fact that many and few (and, correspondingly, often and seldom) are vague. They are not the only vague quantifiers: there are others, e.g. almost all, hardly any, several. But these other quantifiers differ from many and few in that, while vague, they do provide some information about the required proportion: it is reasonably clear that 99% qualifies as almost all, whereas 1% does not. In contrast, the requirements for many and few seem completely open ended. In fact, Huettner (1984) (as cited by Partee 1988) presents cases where few(w, u) can be true even if all ws are us:
Ariel Cohen 33
(102) Many people admired by David Beckham watched the 2006 World Cup on TV. Let us follow the spirit, if not the letter, of the proposal by Fernando and Kamp. The truth conditions of many depend on our expectations, in the sense that many(w, u) is true iff there are more ws that are us than expected. I propose a formalization of this notion in which the notion of alternatives is central, and which, as I will demonstrate, can be extended to handle relative readings.32 Let us propose a new type of semantic value: world semantic value, ½½uW, which takes into account alternatives to the world of evaluation. ½½uW is a set of extensions of u, one for each world. Formally, this is the range of the intension of u. If u is a property, ½½uW is a set of sets of individuals. I will assume Lewis’ (1968, 1971, 1986) counterpart theory, so that the individuals in different worlds are different.33 If we then apply union to the world semantic value, [½½uW, we get the set of all individuals that are in the extension of u in some world. For example, take (100), whose logical form is something like (103) (where wc is the 2006 World Cup, and acad is the property of being an academic). (103) many(acad, kx.watch(x, wc)) The union of the world semantic value of the restrictor, [½½acadW, is the set of possible academics, i.e. the set of individuals who are 32
The formalization of Fernando and Kamp, while appropriate for absolute readings, is not easily extended to the case of relative readings. 33 See Kratzer (1989) for using counterpart theory in situation semantics, and specifically von Fintel (2004: note 2) for an argument for using it in accounts of adverbial quantification. Alternatively, we could say that properties denote sets of pairs of worlds and individuals (cf. Schubert & Pelletier 1989), but I will not explore this possibility here.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Sentence (101) is exactly like (100), except that the predicate academic has been replaced with sports fan. But since more sports fans are expected to watch soccer matches than academics, (101) will require a higher proportion to be true. In this way, Fernando & Kamp (1996) are able to solve the following puzzle, first noted by Keenan & Stavi (1986): if we replace one of the arguments of many with an argument that is co-extensional with it, the truth conditions of the sentence may change. For example, suppose that David Beckham, the soccer star, admires all and only academics. Thus, the predicates academic and person admired by David Beckham have the same extension. And yet, since we may expect people admired by David Beckham to watch soccer on TV, we may assign to (102) truth conditions that are different from those of (100).
34 No Alternative to Alternatives academics in some world. The union of the world semantic value of the scope, (104), is the set of individuals who watched the 2006 World Cup on TV in some world. (104) [½½kx.watch(x, [wc]F)W
(105) many(acad, kx.watch(x, [wc]F)) The union of the union of the W+F semantic value of the scope, [ [ ½½kx.watch(x, [wc]F)W+F, is the set of individuals who watched something on TV in some world. Now consider the probability that something is a u in some world, given that it is a w in some world and an alternative to u in some world: (106) P([ ½½uW j [ ½½wW \ [ [ ½½uW+F) Since individuals in different worlds are different, (106) is the probability that if an individual in some world is a w and an alternative to u, then it is a u. This is precisely the expectation that a relevant w is a u, required by the theory of Fernando and Kamp. For (100), we are considering the probability (107) P([ ½½kx.watch(x, [wc]F)W j [ ½½acadW \ [ [ ½½kx.watch (x, [wc]F)W+F). This is the probability that someone who is an academic and watched something on TV in some world watched the 2006 World Cup in that world. So, the absolute reading of many is defined as follows: Definition 8 (many, absolute reading) many(w, u) is true iff j½½wO \ ½½/O j W W W+F Þ. O F >Pð[½½/ j [ ½½w \ [ [ ½½/ j½½w \ [½½/ j 34 Just like in the case of contrast semantic value, this notation is not meant to indicate that ½½uW+F somehow ought to be derived compositionally from ½½uW and ½½uF.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We can also combine the world semantic value and the focus semantic value, in a way analogous to the way the contrast semantic value combines the B semantic value and the focus semantic value. ½½uW+F is a set of sets; each set is the focus semantic value of u in some world. If, again, u is a property, [ [ ½½uW+F is a set of individuals; each one is a member of the extension of some focus alternative to u in some world.34 For example, consider (100) again. Let us assume that there is focus on the 2006 World Cup on TV, so that its logical form, annotated for focus, is:
Ariel Cohen 35
Applying this definition to (100), the sentence is true iff the proportion of academics who watched the 2006 World Cup on TV among academics who watched something on TV is higher than the expectation that an academic who watched something on TV watched the 2006 World Cup. These appear to be the correct truth conditions. Now, note what happens if we change W to B: we get precisely the relative reading of many, except that on the right-hand side we have a probability rather than a proportion. I suggest that this is, in fact, the relative reading of many:
j½½wO \ ½½/O j B B B+F . Þ O F >Pð[½½/ j [ ½½w \ [ [ ½½/ j½½w \ [½½/ j By having probabilities, rather than proportions, on the right-hand side of the inequality, we extend the idea of Fernando & Kamp (1996) to the case of relative readings: many is intensional, yet it expresses a relation between sets, and its intensionality only affects the threshold of the inequality. Thus, for example, sentence (49), repeated below, means (under the relative reading) that the proportion of Scandinavians who have won the Nobel Prize in literature is greater than the probability (rather than proportion) that just anybody will win this prize. (108) Many [Scandinavians]B have [won the Nobel Prize in literature]F. We can conclude, then, that there are two strategies to decide how many is many: one involves looking at alternative worlds and the other involves looking at alternative values for the B-marked element. If we pick the first strategy, we get the absolute reading, whereas if we pick the second strategy, we get the relative reading. B-marking facilitates the relative reading; if no element is B-marked, we get absolute readings instead.
5.2 Often The relation between the absolute and relative readings of often is similar. Often is a vague quantifier, which gives no clue as to the required threshold probability. Hence, one of two strategies is employed: either the one that utilizes the world semantic value or, if an element is B-marked, the one that utilizes the B semantic value. We can now rephrase definition 4 using the new semantic values, so that the definition of the absolute reading of often is:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Definition 9 (many, relative reading, final version) many(w, u) is true iff
36 No Alternative to Alternatives Definition 10 (often, absolute reading, final version) often(u) is true iff P(½½uOj [ ½½uF) > P([ ½½uWj [ [ ½½uW+F). And, as in the case of many, if we change W to B, we get the relative reading, whose definition I repeat below for completeness: Definition 11 (often, relative reading) often(u) is true iff P(½½uOj [ ½½uF) > P([ ½½uBj [ [ ½½uB+F). To see how the revised definition works, consider again (70) and its logical form (71), repeated below:
We have seen that this sentence is true, according to the absolute reading, iff (110) P(fs j dx(in(x, s) ^ pl(x) ^ crkd(x))gjfs j dx(in(x, s) ^ pl(x))g) > q, where q is ‘large’. We can now give a value to q. Note that (111) [ ½½uW ¼ fs j s is a situation in some world containing a crooked politiciang, and (112) [ [ ½½uW+F ¼ fs j s is a situation in some world containing a politiciang. According to definition 10, u is equal to the conditional probability of (111) given (112). In words, u is the probability that in some world, a situation containing a politician contains a crooked politician, hence the expectation that a politician will be crooked. Thus, (109) is true just in case the probability in the actual world for a politician to be crooked is greater than expected. Since the definition of the relative readings has not changed, its calculation proceeds as before.
5.3 Generics Generics are not usually considered vague.35 Indeed, in the above, I have proposed that their absolute reading is an inequality with a precise threshold 0.5. But where did this value come from? And why do generics have relative readings, like vague quantifiers? 35 Though they are by some researchers: for example, according to Farkas & Sugioka (1983) generics express the vague quantifier significantly many.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(109) a. A [politician]B is often [crooked]F. b. often(ks.dx(in(x, s) ^ pl(x) ^ crkd(x)))
Ariel Cohen 37
Let us assume for a moment that generics really are vague. If so, how are they interpreted? The absolute reading of generics should be defined just like the absolute reading of often (except that topic of the generic is mapped onto the restrictor): Definition 12 (generics, absolute reading, second version) gen(w, u) is true iff P(½½uOj½½wO \ [ ½½uF) > P([ ½½uWj [ ½½wW \ [ [ ½½uW+F).
(113) a. Academics like to watch sports on TV. b. People admired by David Beckham like to watch sports on TV. It appears, therefore, that the threshold for both (113a) and (113b) is the same, hence the expectations are the same. What does it mean to have the same expectations about academics as about people admired by David Beckham? If we know something about either academics or David Beckham’s object of admiration, surely we will have some expectations about them, and these expectations will differ. If our expectations about them are the same, this means that we have no prior beliefs about what academics or people admired by David Beckham are likely to do. Thus, when we calculate the threshold for the absolute reading of generics, we do so as if we have no relevant prior knowledge. What is the reason for this? Elsewhere (Cohen 1996, 1997) I have proposed that the use of a generic is to state a default rule. A default rule is a rule that one has to resort to in the absence of relevant information. For example, someone who utters (114) suggests a rule of reasoning, 36 In fact, they are co-extensional throughout a long period of time; this is necessary when we apply this test to generics, since they are parametric on time (Cohen 1999a).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
How can we put a value on the right-hand side of the inequality, namely the threshold? In the case of many(w, u) (and its adverbial counterpart), the threshold represented the expectation that a w is a u, following Fernando & Kamp (1996). As we have seen, one of the arguments for this proposal is that when we replace one of the arguments of many with another that is co-extensional with it, the truth conditions of the sentence may change. However, this is not the case with generics. Let us assume, once again, that David Beckham admires all and only academics. Then the predicates academic and person admired by David Beckham are coextensional.36 Nonetheless, (113a) and (113b) appear to have the same truth conditions:
38 No Alternative to Alternatives according to which, if we know that Nevermore is a raven, and we do not know anything else about Nevermore, we may conclude that it is black. (114) Ravens are black.37
Definition 13 (generics, absolute reading, final version) gen(w, u) is true iff P(½½uOj½½wO \ [ ½½uF) > P([ ½½uWj [ ½½wW \ [ [ ½½uW+F) ¼ 0.5. As before, if we change W to B, we get the relative reading as defined in 7, repeated below: Definition 14 (generics, relative reading) gen(w, u) is true iff P(½½uOj½½wO \ [ ½½uF) > P([ ½½uBj [ ½½wB \ [ [ ½½uB+F).
5.4 Independent evidence for the world semantic value? The world semantic value appears to provide us with a uniform account of both absolute and relative readings. But is there independent evidence 37 The meaning and use of a generic are not unrelated: the default rule is sound (or ‘useful’) iff the generic sentence is acceptable and true. 38 Of course, prior knowledge is necessary in order to evaluate the truth value of the generic, e.g. whether ravens are, in fact, black.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Therefore, a generic, unlike many, does not assume any prior knowledge from the hearer in order to determine its threshold.38 The value of the threshold is that which indicates no knowledge. In order to calculate this value, we can use The Principle of Indifference. Keynes (1921) describes the principle this way: ‘The Principle of Indifference asserts that if there is no known reason for predicating of our subject one rather than another of several alternatives, then relatively to such knowledge the assertions of each of these alternatives have an equal probability’ (p. 42, original emphases). In the case of the generic gen(w, u) we have two alternatives: either ws are us or they are not. Since we calculate the probabilities as if we had no prior knowledge about w and u, both options will receive equal probability, namely 0.5. The difference, then, between the absolute readings of often and generics is that, with the latter, the threshold is fixed at 0.5. The final version of the truth conditions of generics expresses this fact:
Ariel Cohen 39
that it is a real phenomenon? And is it, like focus and the contrast semantic values, typically, though perhaps not obligatorily, associated with a specific intonation contour? In this admittedly speculative section, I would like to argue that the answer to both questions is, in fact, yes. Specifically, I propose that the world semantic value plays a role in the interpretation of incredulity questions. Suppose Bill hears Ann uttering (115a); in response, Bill utters (115b).39 (115) a. Ann: John is going to get the job. b. Bill: JOHN is going to get the job?!
(116) fJohn is going to get the job, Mary is going to get the job, Julie is going to get the job, . . .g This set of alternatives corresponds to a question inquiring which of these alternatives was asserted. Artstein can therefore account for the question aspect of an echo question: it is used when one interlocutor failed to 39 Capitals indicate pitch accent. The interpretation of this pitch accent will be discussed momentarily.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
How are we to interpret Bill’s utterance? On the one hand, it looks like a question—specifically, an echo question. Bill’s utterance ends with a rising intonation, and it can get the same sort of answer that a genuine question would elicit. Thus, ‘yes’ is a perfectly reasonable response to (115b). On the other hand, however, Bill’s utterance is not really a question. Bill is not seeking information; we can safely assume that Bill understood what Ann was saying. The point of Bill’s utterance is to express incredulity. Bill does not question the fact that John will, indeed, get the job, but expresses surprise—this is not at all what Bill expected, so much so that it is hard for Bill to believe it. For instance, Bill may believe that John is an extremely unsuitable choice, so that his appointment is incredible. Alternatively, Bill can express indignation: Bill may be interpreted as saying that John’s appointment is bad, unethical, unjust or the like. For example, Bill may have received a promise to get the job himself, and John’s appointment breaks this promise. Incredulity questions are often treated as a kind of echo questions, because they share many syntactic properties (Authier 1993). Looking at the semantics of echo question may therefore help us figure out the meaning of incredulity questions. We have already seen in section 2.4 above the theory of echo questions proposed by Artstein (2002). According to him, the meaning of (115b) is a set of alternatives of the form:
40 No Alternative to Alternatives
(117) fJohnw1 , Johnw2 , Johnw3 , . . .g. We combine the world semantic value of John with the other elements of the sentence, using the standard semantic combinatory rule. The resulting world semantic value of (115b) is a set of propositions of the form: (118) fJohnw1 is going to get the job, Johnw2 is going to get the job, Johnw3 is going to get the job, . . .g40 40 We need to be careful about what propositions are. In this case, it makes little sense to equate them with sets of possible worlds, and we probably want what Lewis (1986) calls a singular proposition: (ia) is identified with the ordered pair (ib), and is true iff the first element is a member of the second (see Dorr 2005 for arguments for using singular propositions in counterpart theory in general).
(i)
a. Johnwi is going to get the job. b. Æ Johnwi , the set of individuals who are going to get the job in some worldæ
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
understand or hear clearly what the other one is saying. Thus, if we interpret (115b) as an echo question, the implication is that he did not hear clearly and is seeking confirmation about the identity of the person who will get the job. Artstein acknowledges that in addition to clarification seeking echo questions, there are also cases where an echo question is used to express incredulity or indignation about some proposition, usually the previous utterance or an entailment of it. However, he does not explain how these particular aspects of the meaning follow from his system: how does it follow that if Bill is inquiring about the identity of the person who got the job, that Bill knows it is John, but expresses incredulity or indignation about the fact? Moreover, incredulity questions differ in their intonation from pure echo questions. In fact, they have a tune similar to that of ordinary declaratives, except that, being questions, they have a final rise rather than a final fall (Moulton 1987). As Gunlogson (2003) points out, this, in itself, is not sufficient for the interpretation of incredulity: rising declaratives can express a wide range of speaker attitudes, and incredulity is only one of them. The meaning of incredulity is expressed not through the tune but via expanded pitch range (Hirschberg & Ward 1992; Herman 1996; Jun & Oh 1996; Ladd & Morton 1997; Gunlogson 2003; Lee 2005). Nonetheless, I believe that Artstein’s insight, namely that incredulity questions, just like echo questions, involve reference to a set of alternatives, is correct. However, the relevant semantic value here is not the focus semantic value, but the world semantic value. Expanded pitch range is used to indicate that the world semantic value of John ought to be considered. This is the set of counterparts to John in each one of Bill’s belief (or normative) worlds:
Ariel Cohen 41
What Bill is actually asking, then, is this: in which world is John going to get the job? The domain of worlds that Bill is asking about depends on the modal base (which, in turn, is dependent on the context). The modal base can be epistemic, i.e. the alternative worlds are Bill’s belief worlds: in each one of these worlds, some candidate is getting the job. Bill is then asking us to find a world among them in which John gets the job. This is a rhetorical question, because Bill already knows the answer—he hardly needs us to tell him what is in his belief worlds! Normally, a rhetorical wh-question is interpreted as implying that the answer is the empty set.41
Thus, (119a) implies that nobody believes such nonsense, (119b) implies that he has never spoken against his mother, (119c) implies that it makes no difference. Therefore, when Bill is asking about his belief worlds, he is implying that the answer is the empty set: i.e. in none of his belief worlds does John get the job. Hence, he getting the job is incredible.42 Alternatively, the modal base may be deontic. In this case, Bill refers to worlds that are permissible, according to his norms. Once again, this is interpreted as a rhetorical question, since Bill’s norms are obviously known to himself. Therefore, Bill is implying that in none of these worlds does John get the job. This is how the indignation interpretation is generated: the appointment of John to the job constitutes a violation of Bill’s norms of conduct. Thus, an incredulity question has a dual aspect. On the one hand, it really is a question: Bill is asking in which world John is going to get the job. If Ann answers yes, she is indicating such a world—the actual world (though this world may not be among Bill’s belief or normative worlds). On the other hand, it is a statement of incredulity or indignation: by being rhetorical, the question implies that none of the possible answers are true, i.e. that the echoed statement is incredible or outrageous. If this account is on the right track, it provides an additional case where the world semantic value plays a role in interpretation and is associated with a distinct form of intonation: expanded pitch range. 41 I am using the neutral term ‘implying’, since it is not relevant to our discussion here whether this is an entailment, a presupposition or an implicature. 42 Of course, we are referring to Bill’s belief worlds before Ann spoke; after he heard Ann and accepted what she said, Bill’s belief worlds will obviously contain the fact that John will get the job.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(119) a. Who believes such nonsense? (Bolinger 1957, 158) b. When has he ever said a word against his mother? (Horn 1978: 151) c. What difference does it make? (Quirk et al. 1985: 826)
42 No Alternative to Alternatives 6 CONCLUSIONS
(120) [I]B would buy [‘The Hotel New Hampshire’]F. Then ½½uF+B would be (121).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In this paper, I have proposed that there are two strategies to interpret vague quantifiers: both strategies use the same principle of utilizing alternatives, but one involves taking alternative worlds into account, while the other considers alternatives to B-marked elements. Neither of these strategies involves use of existential presuppositions; hence, I conclude that the device of alternatives is necessary in order to explain the range of interpretations of quantificational structures. The conclusion is, then, that the focus semantic value, far from being a spurious addition to the semantics, is only one of an arsenal of semantic values: in fact, six such values have been proposed in this paper. If the analysis presented here in on the right track, the phenomenon of various semantic values, generated by considering alternatives to various elements (focus marked elements, B-marked elements, worlds) is quite ubiquitous. In hindsight, this is perhaps not so surprising. Multidimensional theories are constantly being proposed, to account for a variety of phenomena at the interface of semantics and pragmatics. Indeed, Potts (2007) identifies so many such accounts, that he sees the need to classify them. And as we go even more deeply into the interaction between semantics and pragmatics, it is to be expected that such multidimensional theories will require more and more semantic values. There were times, not too long ago, when it was thought that focus merely expressed some pragmatic notion (‘newness’ or some such), and had no influence on truth conditions. In those days, there was indeed no need for additional semantic values. But when the effect of focus on truth conditions began to be recognized, the time was ripe for Rooth’s additional semantic value. Later, when Bu¨ring attempted to incorporate an additional pragmatic phenomenon into semantics, he saw the need for yet another semantic value. It is important to note that the semantic values are not arbitrarily chosen: there seems to be some internal structure to the set of semantic values. ½½uB+F is, in a sense, a combination of ½½uB and ½½uF; similarly, ½½uW+F combines ½½uW and ½½uF. But are there alternatives to other elements? And are there other combinations of semantic values? Is there, for example, ½½uF+B? This would be a set of B semantic values for each alternative to the focus. For example, suppose u is (4b), repeated below as (120).
Ariel Cohen 43
Is there such a semantic value? The answer ought to be empirical—is there a phenomenon whose explanation requires it? As far as I have been able to determine, the answer is no. There also does not appear to be any evidence for a ½½uF+W semantic value. It seems, then, that there is a certain order in the combination of the background or the world semantic value with focus, so that the focus semantic values is, in a sense, ‘more basic’ than the other two. Why is this? A complete answer would have to await future research, which will chart the domain of semantic values and identify the principles dictating the kind of things to which alternatives can be considered, and the sanctioned combinations of semantic values to create new ones. For now, I will offer a tentative speculation. The beginnings of an explanation might be in the realization that focus really is very basic, in that it is used to indicate new information, or the ‘point’ of the sentence, or some such. It might even be the case that, as argued in Erteschik-Shir & Lappin (1987), focusing is an integral part of the computational systems of all modes of perception, and is a single task-specific mechanism that identifies the foregrounded constituent in representations of all modular systems. If that is correct, then focus, in a sense, comes for free. Seen in this light, the ½½uB+F or ½½uW+F semantic values are natural: they denote a set of sets. Each such set corresponds to the focus semantic value of a sentence that contains focus marking. In contrast, the ½½uF+W or ½½uF+B semantic values, had they existed, would be rather unnatural. They, too, would denote a set of sets. But now, each set would correspond to the background/world semantic value of a sentence without any focus. In other words, the background or world
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(121) f fI would buy ‘War and Peace’, Paul Simon would buy ‘War and Peace’, Fritz would buy ‘War and Peace’, Fritz’s brother would buy ‘War and Peace’, . . .g, fI would buy ‘The Hotel New Hampshire’, Paul Simon would buy ‘The Hotel New Hampshire’, Fritz would buy ‘The Hotel New Hampshire’, Fritz’s brother would buy ‘The Hotel New Hampshire’, . . .g, fI would buy ‘The World According to Garp’, Paul Simon would buy ‘The World According to Garp’, Fritz would buy ‘The World According to Garp’, Fritz’s brother would buy ‘The World According to Garp’, . . .g, . . .g
44 No Alternative to Alternatives semantic values would be treated as more basic than the focus semantic value. Acknowledgements I thank David Beaver, Sigrid Beck, Hans Kamp and two anonymous referees for many helpful comments and suggestions. This work has been made possible in part by a grant from the Alexander von Humboldt Foundation.
REFERENCES Artstein, R. (2002), ‘A focus semantics for echo questions’. In A´gnes BendeFarkas and A. Riester (eds.), Proceedings of the Workshop on Information Structure in Context. IMS, University of Stuttgart. Stuttgart, Germany. 98–107. Authier, J. M. (1993), ‘Nonquantificational wh and weakest crossover’. Linguistic Inquiry 24:161–8. Bartels, C. (2004), ‘Acoustic correlates of ‘‘second occurrence’’ focus: toward an experimental investigation’. In Kamp and Partee (eds.). 354–61. Beaver, D. (2001), Presupposition and Assertion in Dynamic Semantics. CSLI. Stanford, CA. Beaver, D. & B. Clark (2003), ‘‘‘Always’’ and ‘‘only’’: why not all focus sensitive operators are alike’. Natural Language Semantics 11:323–62. Beaver, D., B. Clark, E. Flemming, T. F. Ja¨ger & M. Wolters (2004), ‘When semantics meets phonetics: acoustical studies of second occurrence focus’. Unpublished MS, Stanford University.
Bolinger, D. (1957), Interrogative Structures of American English: The Direct Question. University of Alabama Press. Tuscaloosa, AL. Bosch, P. & R. van der Sandt (eds.) (1999), Focus: Linguistic, Cognitive, and Computational Perspectives. Cambridge University Press. Cambridge. Bu¨ring, D. (1997), The Meaning of Topic and Focus—The 59th Street Bridge Accent. Routledge. London. Bu¨ring, D. (1999), ‘Topic’. In Bosch and van der Sandt (eds.). 142–65. Carlson, G. (1977), Reference to Kinds in English. Ph.D. thesis, University of Massachusetts, Amherst, MA. Also published in 1980, Garland. New York. Carlson, G. (1989), ‘On the semantic composition of English generic sentences’. In Chierchia et al. (eds.). 167–92. Carlson, G. & F. J. Pelletier (eds.) (1995), The Generic Book. University of Chicago Press. Chicago, IL.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
ARIEL COHEN Department of Foreign Literatures and Linguistics Ben-Gurion University of the Negev Beer Sheva 84105 Israel e-mail:
[email protected]
Ariel Cohen 45 Linguistics—WCCFL XIV. CSLI Publications. Stanford, CA. 155–167. de Swart, H. (1991), Adverbs of Quantification: A Generalized Quantifier Approach. Ph.D. thesis, Groningen University. Also published in 1993, Garland. New York. Dorr, C. (2005), ‘Propositions and counterpart theory’. Analysis 65: 210–8. Eckardt, R. (1994), Adverbs in Focus. DYANA/Universita¨t Tu¨bingen. Germany Erteschik-Shir, N. (1997), The Dynamics of Focus Structure. Cambridge University Press. Cambridge. Erteschik-Shir, N. & S. Lappin (1987), ‘Dominance and modularity’. Linguistics 25:671–85. Farkas, D. F. & Y. Sugioka (1983), ‘Restrictive if/when clauses’. Linguistics and Philosophy 6:225–58. Fernando, T. & H. Kamp (1996), ‘Expecting many’. In Galloway and Spence (eds.). 53–68. Fox, D. (1999), ‘Reconstruction, binding theory, and the interpretation of chains’. Linguistic Inquiry 30:157–96. Galloway, T. & J. Spence (eds.) (1996), Proceedings of the Sixth Conference on Semantics and Linguistic Theory. Cornell University. Ithaca, NY. Geilfuß, J. (1993), ‘Nominal quantifiers and association with focus’. In P. Ackema and M. Schoorlemmer (eds.), Proceedings of the Workshop on the Semantic and Syntactic Analysis of Focus. OTS Working Papers. Utrecht, The Netherlands. 33–41. Geurts, B. (forthcoming), ‘Specific indefinites, presupposition, and scope’. In R. Ba¨uerle, U. Reyle and T. E. Zimmermann (eds.), Presuppositions and Discourse. Elsevier. Oxford. Geurts, B. & R. van der Sandt (2004), ‘Interpreting focus’. Theoretical Linguistics 30:1–44.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Chierchia, G. (1992), ‘Anaphora and dynamic binding’. Linguistics and Philosophy 15:111–83. Chierchia, G. (1995), ‘Individual-level predicates as inherent generics’. In Carlson and Pelletier (eds.). 176–223. Chierchia, G., B. H. Partee & R. Turner (eds.) (1989), Properties, Types and Meaning. Kluwer. Dordrecht, The Netherlands. Cohen, A. (1996), Think Generic: The Meaning and Use of Generic Sentences. Ph.D. thesis, Carnegie Mellon University. Published in 1999, CSLI. Stanford, CA. Cohen, A. (1997), ‘Generics and default reasoning’. Computational Intelligence 13:506–33. Cohen, A. (1999a), ‘Generics, frequency adverbs and probability’. Linguistics and Philosophy 22:221–53. Cohen, A. (1999b), ‘How are alternatives computed?’ Journal of Semantics 16:43–65. Cohen, A. (2001a), ‘On the generic use of indefinite singulars’. Journal of Semantics 18:183–209. Cohen, A. (2001b), ‘Relative readings of many, often, and generics’. Natural Language Semantics 9:41–67. Cohen, A. (2004a), ‘Existential generics’. Linguistic and Philosophy 27: 137–68. Cohen, A. (2004b), ‘Fronted quantificational adverbs’. In B. Shaer, W. Frey, and C. Maienborn (eds.), Proceedings of the Dislocated Elements Workshop, vol. 1. ZAS. Berlin. 63–76. Cohen, A. & N. Erteschik-Shir (2002), ‘Topic, focus, and the interpretation of bare plurals’. Natural Language Semantics 10:125–65. de Hoop, H. & J. Sola` (1995), ‘Determiners, context sets, and focus’. In J. Camacho, L. Choueiri and M. Watanabe (eds.), Proceedings of the 14th West Coast Conference on Formal
46 No Alternative to Alternatives Linguistic Meaning. Elsevier. Amsterdam, The Netherlands. Kamp, H. & U. Reyle (1993), From Discourse to Logic. Kluwer Academic Publishers. Dordrecht, The Netherlands. Keenan, E. L. & J. Stavi (1986), ‘A semantic characterization of natural language determiners’. Linguistics and Philosophy 9:253–326. Keynes, J. M. (1921), A Treatise on Probability. Macmillan. London. Kratzer, A. (1989), ‘An investigation of the lumps of thought’. Linguistics and Philosophy 12:607–53. Krifka, M. (1995), ‘Focus and the interpretation of generic sentences’. In Carlson and Pelletier (eds.). 238–64. Krifka, M. (1998), ‘Scope inversion under the rise-fall pattern in German’. Linguistic Inquiry 29:75–112. Krifka, M. (2001), ‘Non-novel indefinites in adverbial quantification’. In C. Condoravdi and G. R. de Lavalette (eds.), Logical Perspectives on Language and Information. CSLI Publications. Stanford, CA. 1–40. Krifka, M., F. J. Pelletier, G. Carlson, A. ter Meulen, G. Link & G. Chierchia (1995), ‘Genericity: an introduction’. In Carlson and Pelletier. 1–124. Ladd, R. D. & R. Morton (1997), ‘The perception of intonational emphasis: continuous or categorical?’ Journal of Phonetics 25:313–42. Lappin, S. (2000), ‘An intensional parametric semantics for vague quantifiers’. Linguistics and Philosophy 23:599–620. Lee, O. J. (2005), The Prosody of Questions in Beijing Mandarin. Ph.D. thesis, Ohio State University. Columbus, OH. Lewis, D. (1968), ‘Counterpart theory and quantified modal logic’. Journal of Philosophy 65:113–26.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Gunlogson, C. (2003), True to Form: Rising and Falling Declaratives as Questions in English. Routledge. New York. Heim, I. (1990), ‘E-type pronouns and donkey anaphora’. Linguistics and Philosophy 13:137–77. Hendriks, H. & P. Dekker (1996), ‘Links without locations: information packaging and non-monotone anaphora’. In P. Dekker and M. Stokhof (eds.), Proceedings of the Tenth Amsterdam Colloquium. University of Amsterdam. Amsterdam, The Netherlands. 339– 58. Herburger, E. (1997), ‘Focus and weak noun phrases’. Natural Language Semantics 5:53–78. Herman, R. (1996), ‘Final lowering in Kipare’. Phonology 13:171–96. Hirschberg, J. & G. Ward (1992), ‘The influence of pitch range, duration, amplitude and spectral features on the interpretation of the rise-fall-rise intonation contour in English’. Journal of Phonetics 20:241–51. Horn, L. R. (1978), ‘Some aspects of negation’. In J. H. Greenberg, C. A. Ferguson and E. A. Moravcsik (eds.), Universals of Human Language, Volume 4: Syntax. Stanford University Press. Stanford, CA. 127–210. Huettner, A. (1984), ‘Semantics seminar paper on few and many’. Amherst, MA. University of Massachusetts. Jackendoff, R. S. (1972), Semantic Interpretation in Generative Grammar. MIT Press. Cambridge, MA. Jun, S. A. & M. Oh (1996), ‘A prosodic analysis of three types of wh-phrases in Korean’. Language and Speech 39:37– 61. Kadmon, N. (2001), Formal Pragmatics: Semantics, Pragmatics, Presupposition, and Focus. Blackwell. Oxford. Kamp, H. & B. H. Partee (eds.) (2004), Context-Dependence in the Analysis of
Ariel Cohen 47 Potts, C. (2007), ‘Into the conventionalimplicature dimension’. Philosophy Compass 4:665–79. Quirk, R., S. Greenbaum, G. Leech & J. Svartvik (1985), A Comprehensive Grammar of the English Language. Longman. London. Reinhart, T. (1981), ’Pragmatics and linguistics: an analysis of sentence topics. Philosophica 27:53–94. Rooth, M. E. (1985), Association with Focus. Ph.D. thesis, University of Massachusetts, Amherst. Rooth, M. E. (1992), ‘A theory of focus interpretation’. Natural Language Semantics 1:75–116. Rooth, M. E. (1995), ‘Indefinites, adverbs of quantification and focus semantics’. In Carlson and Pelletier (eds.). 265–99. Rooth, M. E. (1996), ‘On the interface principles for intonational focus’. In Galloway and Spence (eds.). 202–26. Rooth, M. E. (1999), ‘Association with focus or association with presupposition?’ In Bosch and van der Sandt (eds.). 202–26. Schubert, L. K. & F. J. Pelletier (1987), ‘Problems in the representation of the logical form of generics, plurals, and mass nouns’. In E. LePore (ed.), New Directions in Semantics. Academic Press. London. 385–451. Schubert, L. K. & F. J. Pelletier (1989), ‘Generically speaking, or using discourse representation theory to interpret generics’. In Chierchia et al. (eds.). 193–268. Schwarzschild, R. (1999), ‘Givenness, AvoidF and other constraints on the placement of accent’. Natural Language Semantics 7:141–77. Steedman, M. (1991), ‘Structure and intonation’. Language 67:260–96. Steinberg, D. D. & L. A. Jakobovits (eds.) (1971), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Lewis, D. (1971), ‘Counterparts of persons and their bodies’. Journal of Philosophy 68:203–11. Lewis, D. (1975), ‘Adverbs of quantification’. In E. L. Keenan (ed.), Formal Semantics of Natural Language. Cambridge University Press. Cambridge, MA. 3–15. Lewis, D. (1986), On the Plurality of Worlds. Blackwell. Oxford. Moulton, W. G. (1987), ‘On the prosody of statements, questions, and echo questions’. American Speech 62:249–61. O’Connor, J. D. & G. F. Arnold (1973), The Intonation of Colloquial English: A Practical Handbook. Longman. London. Partee, B. H. (1988), ‘Many quantifiers’. In J. Powers and K. de Jon (eds.), ESCOL’88: Proceedings of the Fifth Eastern States Conference on Linguistics. Ohio State University. Columbus, OH. 383–402. Partee, B. H. (1991), ‘Topic, focus and quantification’. In S. Moore and A. Z. Wyner (eds.), Proceedings of the First Conference on Semantics and Linguistic Theory. Cornell University. Ithaca, NY. 159–87. Pelletier, F. J. & N. Asher (1997), ‘Generics and defaults’. In J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language. Elsevier. Amsterdam, The Netherlands. 1125–77. Pierrehumbert, J. (1980), The Phonology and Phonetics of English Intonation. Ph.D. thesis, MIT, Cambridge, MA. Pierrehumbert, J. & J. Hirschberg (1990), ‘The meaning of intonational contours in discourse’. In P. Cohen, J. Morgan and M. Pollack (eds.), Intentions in Communication. MIT Press. Cambridge, MA. 271–311. Portner, P. & K. Yabushita (1998), ‘The semantics and pragmatics of topic phrases’. Linguistics and Philosophy 21:117–57.
48 No Alternative to Alternatives von Fintel, K. (1994), Restrictions on Quantifier Domains. Ph.D. thesis, University of Massachusetts, Amherst. von Fintel, K. (2004), ‘A minimal theory of adverbial quantification’. In Kamp and Partee (eds.). 137–75. Westersta˚hl, D. (1985), ‘Logical constants in quantifier languages’. Linguistics and Philosophy 8:387–413. Wilkinson, K. (1991), Studies in the Semantics of Generic Noun Phrases. Ph.D. thesis, University of Massachusetts, Amherst. Zeevat, H. (1992), ‘Presupposition and accommodation in update semantics’. Journal of Semantics 9: 379–412. First version received: 04.08.2006 Second version received: 10.12.2007 Accepted: 12.04.2008
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Psychology. Cambridge University Press. Cambridge. Strawson, P. F. (1964), ‘Identifying reference and truth values’. Theoria 30. Reprinted in Steinberg and Jakobovits (1971). Vallduvı´, E. & E. Engdahl (1996), ‘The linguistic realization of information packaging’. Linguistics 34:459–519. Vallduvı´, E. & M. Vilkuna (1998), ‘On rheme and kontrast’. In P. W. Culicover and L. McNally (eds.), The Limits of Syntax. Academic Press. New York. 79–108. Vallduvı´, E. & R. Zacharski (1994), ‘Accenting phenomena, association with focus, and the recursiveness of focus-ground’. In P. Dekker and M. Stokhof (eds.), Proceedings of the 9th Amsterdam Colloquium. ILLC. Amsterdam. 638–702.
Journal of Semantics 26: 49–86 doi:10.1093/jos/ffn009 Advance Access publication December 1, 2008
Presupposition Accommodation and Informativity Considerations with Aspectual Still YAEL GREENBERG Bar Ilan University
This paper deals with a newly observed phenomenon which lies at the interface of the semantics and pragmatics of aspectual still (as in John is still asleep), namely the fact that still is infelicitous when it appears in past tense sentences whose reference time is not specified by some temporal adverbial or the utterance context. The main claim of the paper is that in such sentences, the truth of the assertion and that of the ‘prior time’ presupposition this particle triggers can be both inferred from the truth of the minimally contrasting still-less counterpart. Moreover, in such cases the presuppositional status of the ‘prior time’ claim is lost. Hence, the use of still in such sentences is uninformative and thus infelicitous. The analysis has several more general theoretical implications. Concerning the semantics of still, it shows that the novel data cannot be accounted for by using current definitions of the presupposition triggered by still. Instead, a modified definition of still is developed, which, following Ippolito (2007), uses one eventuality variable in both the assertion and the presupposition of sentences with still but, unlike this theory, does not require the denotation of this variable to be contextually salient. As for the analysis of tense and temporal structure of clauses, the interaction of still with frame adverbials supports the view that such adverbials denote intervals identical to the reference time of the sentence, rather than including it. In addition, the paper argues that the contrast between felicitous and infelicitous cases of still can only be explained if, contrary to many current analyses, we assume that past tense is not necessarily anaphoric but can be represented in some cases as a new variable bound by existential closure. Given the proposed analysis, the felicity or infelicity of still in past tense sentences can be seen as a diagnostic for determining whether or not the reference time in such sentences is anaphoric or not.
1 INTRODUCTION This paper deals with a newly observed phenomenon which lies at the interface of the semantics and pragmatics of aspectual still (as in John is still asleep), namely the fact that still is systematically infelicitous when the time described by the sentence (or more technically, the reference time) is not given a specific characterization by a temporal adverbial or the utterance context. The main claim I make is that this kind of infelicity results from the fact that in such constructions the use of still is uninformative. 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
Abstract
50 Presuppositions and Informativity with Aspectual still
2 THE NEWLY OBSERVED DATA: FELICITOUS V. INFELICITOUS OCCURRENCES OF STILL I will be concerned with felicity contrasts exemplified in (1): (1)
A: How’s John? B: Well, he is (still) unemployed (but we hope he will have a job soon) B#: Well, he was (#still) unemployed (but now he has a job)
As seen in (1), the still-less versions of both present and past tense sentences are felicitous.1 However, whereas when adding still to the 1 One may claim here that a past tense answer to a present tense question as in (1B#) is, or should be, infelicitous even without still, due to the tense mismatch which could lead to incoherent discourse. However, all my informants accepted such discourses with no problems. One potential explanation of this could be that A is not asking how John is at the moment of utterance, but more generally, how John is ‘these days’, to which an answer encompassing the near past is relevant. Another explanation might be that the discourse is coherent because B wants to emphasize John’s present situation (having a job) by contrasting it with the past (where he was unemployed). Thanks to an anonymous reviewer for pointing out this potential difficulty. Notice, though, that some past tense sentences may be perfectly felicitous without having clear relevance to the present. For example, a teacher can felicitously utter the out of the blue (ii). Notice that still is infelicitous here too:
(ii) Shakespeare (#still) lived in England.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The analysis, which is couched within a Discourse Representation Theory (DRT) framework (Kamp 1981), also argues for two more general theoretical points. First, I show that the data cannot be accounted for by using current definitions of the presupposition triggered by still. I propose a modified definition of still which, following Ippolito’s (2007) analysis, uses one eventuality variable in both the assertion and the presupposition of sentences with still, but, unlike this theory, does not require the denotation of this variable to be contextually salient. Second, as opposed to many current theories, I show that the data on still support a view where tense in past tense sentences is not necessarily anaphoric or pronominal, but rather potentially existentially closed and novel. The novel data are described in section 2. Section 3 gives background about the anaphoric vs. novel characterization of tense and reference times and about the semantics of still. After examining some potential explanations of the data in section 4 and pointing out their shortcomings, I turn to my own proposal for the semantics of still in section 5 and show how it can account for the novel data. Section 6 examines the compatibility of the analysis with current claims on presuppositions and informativity. Section 7 concludes the paper and examines some remaining questions and directions for further research.
Yael Greenberg 51
present tense (1B) this felicity is maintained, adding it to the past tense (1B#) renders it strikingly infelicitous. Similar felicity contrasts are found in other minimally contrasting present/past tense sentences, for example (2)–(4): (2)
(3)
What is the reason for this contrast? No direct answer to this question can be found in the literature on still, simply because these kinds of data have not yet been examined in any theory dealing with this particle. Of course, one might be tempted to stipulate that still is simply bad with the past tense and fine only with the present tense. However, not only is this generalization unmotivated, but as the felicity of still in (5B)–(8B) shows, it is simply wrong: (5) (6) (7) (8)
A: How’s John? B: Well, I saw him last month. He was (still) unemployed. A: Look at this poor building! It was so much nicer in the 1980s! B: Were you (still) living in it then? A: I heard lots of stories about Mr. Smith. B: Me too. During the war he was (still) very rich. A: There is an important meeting tonight. B: Yes. John came in and told me about it. But I was (still) ill, so I told him that I must rest and that I won’t be able to arrive.
The occurrence of still in these past tense sentences is a lot better than in (1B#)–(4B#) above. The difference between the two types of sentences seems to be related to the specification of the reference time. Intuitively, what all felicitous past tense sentences in (5B)–(8B) seem to share is that their reference time is specified by an adverbial in the previous sentence [as in (5)] and in the sentence itself [as in (6) and (7)] or by the eventuality time of the previous sentence [i.e. the time when John told me about the meeting, in (8)]. In contrast, no specification of the reference time seems to exist in the past tense (1B#4B#), and A’s
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(4)
A: There is an important meeting tonight. B: I won’t be able to be there. I am (still) ill, and I must rest. B#: I won’t be able to be there. I was (#still) ill, and I must rest. A: Look at this poor building! B: Are you (still) living in it? B#: Were you (#still) living in it? A: I heard lots of stories about Mr. Smith. B: Me too. He is (still) very rich. B#: Me too. He was (#still) very rich.
52 Presuppositions and Informativity with Aspectual still utterances in these sentences do not supply any exact specification of that time either. It should become clear now what the felicitous past tense sentences in (5B)–(8B) share with the felicitous present tense sentences in (1B)– (4B) above: in the latter case too, the reference time can be said to be specified, namely, to refer to the speech time of the sentence. These intuitions can be informally summarized as in (9): (9) still can only be felicitous in a sentence whose reference time is specified by/refers to another time in the linguistic or discourse context.2
3 SOME BACKGROUND: CHARACTERIZING REFERENCE TIMES AND THE SEMANTICS OF STILL
3.1 Reference times, eventuality times and speech times 3.1.1 Basics. I follow here ideas developed and used in, for example, Partee (1973, 1984), Hinrichs (1986), Kamp & Reyle (1993), Ogihara (1994) and Kratzer (1998). Following the tradition of Reichenbach (1947), these theories analyse the temporal structure of sentences using three temporal parameters: the speech time, the reference time and the eventuality time. In simple sentences, the tense node denotes a time argument which stands for the reference time of the sentence and whose position is determined relative to the speech time.3 Specifically, in present tense 2
As pointed out to me by a reviewer, it is interesting to note that the situation with the perfect aspect seems to be the opposite, namely we get infelicity when the reference time is specific or anaphoric, as in (i). This ‘opposite’ correlation with the specificity/anaphoricity of the reference time may warrant further exploration. (i)
#John has been sick yesterday. 3
Notice that although the use of the term ‘reference time’ in these theories is inspired by Reichenbach’s notion, it is not necessarily identical to the Reichenbachian usage. In this paper, I use the term as it is used in Partee (1973, 1984), Hinrichs (1986), Kamp & Reyle (1993), Ogihara (1994), Kratzer (1998), and as defined above. Comparing this and the Reichenbachian uses of the term is beyond the scope of this paper, but see, for example, Partee (1984) and Nelken and Francez (1995) for some comparisons.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The immediate aim of this paper is to explain this novel generalization. In the next section, I examine some background assumptions concerning the two main components of the analysis, namely the characterization of reference times and the semantics of still. In section 5, I show that once these two components are defined appropriately, their interaction leads to a natural explanation of the generalization in (9).
Yael Greenberg 53
(10)
a. Yesterday afternoon John saw Mary/wrote a letter b. Yesterday afternoon John was very ill/was writing a letter
3.1.2 Are reference times necessarily anaphoric/pronominal? Unlike the reference times of present tense sentences, whose characterization is quite simple (equal to the speech time), the situation with past tense sentences is more complicated. Prior (1967) analysed past tense as existential quantification over times. However, as Partee (1973) shows in her famous example in (11a), taking such a sentence to talk about ‘some time in the past’ would come out wrong, no matter whether we have wide scope negation, as in (11b), which is too strong and seems false, or narrow scope negation, as in (11c), which is too weak and seems trivially true: (11)
a.
I didn’t turn off the stove
4 Given this view, the present tense is anaphoric and not merely ‘indexical’. This is part of a general tendency in the modern semantic literature, which assumes that the traditional distinction between anaphoric (or more precisely, co-referring) and indexical (or deictic) uses of pronouns need not be maintained and does not seem to have real linguistic significance (see, e.g. the discussion in Heim & Kratzer 1998, p. 239–42). A similar position regarding anaphoric and indexical uses of tense is held in, for example, Partee (1973, 1984) and reviewed in section 3.1.2. 5 Heim (1993) and others take this information to be a presuppositional component of tense. For simplicity reasons, I will not attempt to represent this presuppositional status in the DRT-based analysis I propose below (although this is perfectly compatible with what I will suggest).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
sentences, the reference time equals the speech time,4 and in past tense sentences, it precedes it.5 Locating temporal adverbs update the reference time. For example, in ‘John woke up at 9.00’, the denotation of at 9.00 is processed in the beginning and updates the reference time to be a time prior to the speech time, equal to 9.00. The speech time (now) is a contextually salient time, and like other contextually salient arguments, its value is given by a contextually supplied assignment function gc (see, e.g. Heim 1994). As for the eventuality time, represented as s(e), aspectual factors determine its position relative to the reference time. In simple sentences, the time of eventive/perfective predicates is taken to be included in the reference time, whereas the time of stative or imperfective (progressivized) predicates is taken to include or overlap it. For example, in (10a), with eventive predicates, the eventuality times of seeing Mary and writing the letter are included in yesterday afternoon, whereas in (10b) the stative/imperfective eventualities of being very ill and writing the letter temporally overlap yesterday afternoon and may even include that time:
54 Presuppositions and Informativity with Aspectual still b. :dt#[t#< t ^ I turn off the stove at t#] c. dt#[t#< t ^ :I turn off the stove at t#]
(12)
a. At 3 p.m. June 21st, 1967, Mary had a brilliant idea b. Mary woke up sometime during the night. She turned on the light. c. When Mary telephoned, Sam was always asleep.
Following Partee, the view that tense can behave as pronominal/ anaphoric is by now standard. What is not agreed upon, however, is whether tense must be anaphoric/pronominal. Unlike the supporters of the ‘reference time as only pronominal’ approach (e.g. Hinrichs 1986; Kratzer 1998; Avrutin & Reuland 2002; Beck 2006, inter alia), there are theories which represent the reference time in some past tense sentences as existentially closed (such as Comrie 1985; Ogihara 1994; Bonomi 1995; Musan 1997; Kehler 2000; von Fintel & Iatridou 2002; Pancheva & von Stechow 2004).6 I tend to agree with the latter approach, since besides cases where reference times are indeed anaphoric to temporal antecedents, there also seem to be felicitous sentences where no such antecedent is found, that is where, using Heim’s (1982) terminology, the reference time is novel. One such construction is ‘out of the blue’ past tense questions, as in (13a) and (13b), discussed in Kratzer (1998) and Partee (1984), respectively: (13)
a. Who built this church? Borromini built this church. b. Who killed Julius Caesar?
Kratzer (1998) points out that ‘. . .the English question [in (13a)] is acceptable out of the blue. If past tense is pronominal, this is surprising. There is no contextually salient past time in this context’ (p. 16). To solve the problem, Kratzer suggests that what seems to be a simple past sentence in (13a) is, in fact, a sentence with present tense and perfect 6 Other theories are neutral with respect to this question. For example, Abusch (1997) assumes that tense can be pronominal but does not explicitly say whether it must be so.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Instead, Partee claims that the reference time of (11a) is anaphoric to a contextually supplied time, for example the time immediately before I left home. The sentence then asserts that I didn’t turn off the stove at that (contextually supplied) time. This parallels nominal anaphoricity of ‘free’ pronouns, as in ‘She left me’, where the pronoun she is anaphoric to a contextually salient antecedent, for example an individual pointed at. In addition, Partee shows that tense can have other kinds of antecedents that pronouns are known to have, namely definite, indefinite (i.e. existentially closed) and quantified antecedents, as in (12a–c), respectively:
Yael Greenberg 55
(14)
John didn’t build this church.7
Besides ‘out of the blue’ questions and negative sentences, there are also affirmative past tense sentences, uttered in context, whose reference time seems existentially closed. I believe that (15B), the still-less version of (1B#) above, is one such sentence: (15)
A: How’s John? B: Well, he was unemployed (but now he has a job).
Although (15B) is not uttered out of the blue, but rather against the context of (15A), this context does not give us any information about the location of the past reference time of (15B). Intuitively, this reference time is novel, meaning ‘some time in the past’, as in (16) 7
Thanks to a reviewer for pointing out this example to me.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
aspect (not to be confused with the ‘perfect’ auxiliary have). Thus, the reference time of (13a) is indeed pronominal: it is anaphoric to the utterance time (due to the present tense) and, due to the perfect aspect, it asserts that the event time is over by the reference (utterance) time, that is the event is in the past. This suggestion, however, is problematic if we want to maintain the more intuitive view that sentences like (13a) express simple past and not present perfect. Partee (1984) admits that the reference time of (13b) seems to be ‘some time in the past’ but maintains that it is, in fact, anaphoric and suggests that it is ‘large, vague, and possibly even irrelevant . . . (and) . . . could potentially be ‘‘the whole of the past’’’ (p. 314). According to her, the reason we understand the killing of Julius Caesar to happen ‘some time before now’ is because of the well-known condition that the event time is included in the reference time. This explanation, however, would not work for questions with state verbs like ‘Who admired Julius Caesar?’, which are also understood as asking about a state which took place some time in the past. Unlike events, which are assumed to be included in the reference time (here ‘the whole of the past’), states are supposed to include or overlap it. It is not clear, then, why we necessarily get the existential reading in such cases too. In addition, the reference time of some negative sentences does not seem to be anaphoric, but rather existentially closed and novel. (14), for example, neither means that John didn’t build that church in a contextually salient past time nor means that there is a contextually salient past time where he didn’t build it, but rather that he never built this church, that is there is no past time which overlaps a Johnbuilding-this-church eventuality:
56 Presuppositions and Informativity with Aspectual still (where, here and henceforth, ‘O’ is the temporal overlap relation and ‘s(e)’ gives the running time of the event e): (16) de,t[John-unemployed (e) _ t < n ^ t O s(e)]
(17) de,t [John-unemployed (e) ^ t D ^ t < n ^ t O s(e)]
3.2 The semantics of still 3.2.1 The ‘prior time’ presupposition and a potential problem with it Turning now to the semantics of still, virtually all theories analysing this particle (e.g. Ko¨nig 1977; Lo¨bner 1989; Michaelis 1993; Mittwoch 1993; van der Auwera 1993; Krifka 2000; Ippolito 2007) take still U to implicate that U continues at the reference time of U. According to these theories, (18), for example, implicates that John continues to be asleep at the speech time: 8 In this sense, we may say that the use of past tense in (15B) is similar to the use of a specific indefinite Noun Phrase as in John spoke with some/a woman. As with the existential reading of the past tense, the semantic structure of such sentences contains existential quantification over an individual woman. Even if the speaker has a certain individual woman in mind when uttering it, this information is not part of the semantic structure, and crucially, unlike what happens with real pronouns, the felicity of the sentence does not depend on the listener’s ability to identify this individual. 9 Compare with claims in, for example, Bonomi (1995). 10 One may also claim that this restriction should not be part of the semantic structure of (15B) at all but is provided by the pragmatics (e.g. by the need to make a true and relevant assertion). I will not try to develop this approach here but continue to take (17) as a possible representation of (15B#).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
There are two potential objections to this representation. First, one may say that although the context in (15A) does not seem to supply a unique past reference time, the speaker has a certain time in mind which he is referring to when uttering (15B). However, notice that even if the speaker knows that the listener has no way of knowing when exactly John was unemployed, his use of the past tense in (15B) is felicitous. In contrast, if a speaker knows that the listener has no way of identifying the denotation of the pronoun she in ‘She is nice’, he will not use the pronoun, or else, his use of the pronoun will be infelicitous.8 A second objection is that (16) is too weak as a representation of (15B): Clearly, the reference time in this sentence is not any time in the past. (It will not be, e.g. an interval in the prehistoric period.) Rather, it is restricted by the presuppositions and implications of the sentence, for example it is expected to be an interval which appears within John’s life time, and most probably when he is already grown up (where being unemployed is relevant). This can be represented in (17), where the reference time is restricted to occur within a relevant period D (the period where John is alive and grown up):9,10
Yael Greenberg 57
(18)
John is still asleep
(19)
a. John is writing a letter. b. John entered the room. Mary was in the living room c. John entered the room. Mary kissed him.
Given such observations, some theories (e.g. Moens 1987; Moens et al. 1987) have taken this temporal ‘surrounding’ intuition to be, in fact, a defining property of stative predicates (sometimes referred to as ‘the superinterval property’). The problem for the approach to still described above is that if these theories are right, then the ‘prior time’ presupposition triggered by still comes out trivial. For example, the presupposition that John was asleep before 6.13 [in (20a)] would be completely trivial if the still-less (20b) independently entails that the sleeping state holds both before and after 6.13: (20)
a. At 6.13 John was still asleep b. At 6.13 John was asleep
Since still does not seem trivial in (20a), then either the assumption that the ‘prior time’ presupposition is the only contribution of still11 or 11 That is, one can claim that still has an additional contribution to the semantics of the sentence. This kind of view is found in, for example, Michaelis (1993) and van der Auwera (1993), who suggest that still also adds the implication that the state is expected to cease at the reference time. I deal with this suggestion in section 4 below.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This intuition is usually captured using an assertion and a presupposition. (18) is taken to assert that John is asleep now and presuppose that he was also asleep for some time before and up to now. I will henceforth call this latter component the ‘prior time’ presupposition and a large part of the following discussion will be concerned with its formalization. Before doing that, however, let us examine a potential problem for this approach to still in general. As mentioned above, eventualities denoted by stative or progressivized predicates (with which still typically occurs) are standardly taken to include or overlap the reference time. However, when the reference time is denoted by point adverbials, like ‘at 6.13’, ‘when the bell rang’ or even ‘now’, the intuitions tend to be stricter: in such cases, the states seem not merely to overlap but to surround the point reference times, that is to obtain both before and after such points of time. For example, (19a) seems to imply that the writing of the letter started before now. Similarly, unlike the eventive predicate in (19c), the stative predicate in the second sentence of (19b) seems to temporally surround the entering event, so it does not ‘move the narrative forward’ (see, e.g. Partee 1984; Hinrichs 1986):
58 Presuppositions and Informativity with Aspectual still the assumption that states must surround point reference times should be rejected. I believe it is the second assumption which should be rejected, since, as has been convincingly shown by other theories (e.g. Partee 1984; Dowty 1986; Hinrichs 1986; Lascarides & Asher 1993; Ogihara 1994; de Swart & Verkuyl 1999), the ‘surrounding’ effect, though intuitively strong, is, in fact, not entailed and sometimes not even implicated by the use of stative predicates. Consider, for example, (21a,b), from Hinrichs (1986) and de Swart and Verkuyl (1999), respectively: (21)
Here the states get an ‘inceptive’ or an ‘inchoative’ reading, and they do ‘move the narrative forward’.12 Similar examples are found in Dowty (1986), who claims that the ‘superinterval property’ of statives is a cancellable implicature and not part of their Semantics. Even more relevant to us are cases in which the stative or progressive predicates are used with point adverbials or in present tense. Consider, for example, the sentences in (22) (with the stative or progressive predicates in italics): (22)
a. b. c. d. e. f.
We weren’t sure whether John will participate in the race or not, but at 2.00, when the gun went off, he was running.13 The bomb exploded exactly at 14.31 John was inside his room then. That was very lucky—half a second before he was still outside. We did as much as we could, but at 12.34 John was dead. The minute I saw her I was disappointed. When John left the room too I was all alone. John dialed 911 at 17.08. At 17.19 sharp the firefighters were there.
None of these sentences entails that the state denoted by the underlined predicate holds before the time denoted by the point adverbial (e.g. that John was running before the gun went off, that I was all alone before John left the room, etc.). In fact, given the contexts constructed in these examples, such readings would be quite odd. Assuming that states necessarily surround point reference times would wrongly predict such sentences to be infelicitous. 12
This seems to be related to the understood causal relation between the first and second eventualities in (21) (see Lascarides and Asher 1993). 13 Thanks to S. Rothstein (personal communication) for pointing out this example to me.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
a. Sue switched off the light. It was pitch dark in the room. b. Hilary entered the room. Phil was happy to see her.
Yael Greenberg 59
Less extreme cases are illustrated in (23). Here the states can be understood as surrounding the reference time. But crucially, this is not necessary: (23)
When John heard that his father died he was miserable. (The pilot to the passengers): We are in the air! Oh no! The baby is awake! John entered the boss’s room at 2.00. At 2.01 he was unemployed. e. At 6.13 John was asleep.
a. b. c. d.
3.2.2 Two formulations of the assertion and presupposition of sentences with ‘still’ I will focus here on two formulations of the assertion and 14 Of course, this conclusion is dependent on the assumption that there is no subclass of stative predicates which necessarily entail (and not only strongly implicate) the truth of the eventualities at a time before the reference time. If such stative predicates are found, then we should expect still to be systematically infelicitous with them. Felicity of still with such hypothesized predicates will indeed pose a difficulty for the theories followed in this paper, which take the contribution of still to the sentence to be its ‘prior time’ presupposition. However, I am not aware of such a subclass. Notice, for example, that although Ogihara (1994) claims that progressive predicates cannot be inchoative (i.e. they must have the ‘super interval property’), this characterization does not seem right in light of the perfect felicity of (22a) above, with a progressive verb. But further research should look more closely at this question.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(23a), for example, is potentially ambiguous between a reading where John was already miserable when he heard the bad news and a more salient one where he wasn’t (he became miserable as a result of hearing the bad news). Similarly, (23b) can be uttered just as the plane takes off (so ‘being in the air’ is not necessarily true before now), (23c) can be true if the baby was asleep a second ago, (23d) can be true even if a second before 2.01 John was not unemployed yet and (23e) can, in fact, be true even if John was not asleep before 6.13. This is exactly why adding still to such sentences nontrivially entails that the states hold also before the reference time. For example, adding still to (23a) (‘When John heard that his father died he was still miserable’) clearly disambiguates it, excluding the reading where John became miserable when he heard the bad news. The same holds for the other sentences in (23). To summarize, still indeed triggers a ‘prior time’ presupposition, and this presupposition is not trivial, exactly because in reality states do not necessarily surround their reference time. Rather they only overlap or include it.14 We can now turn to a closer and more precise examination of this ‘prior time’ presupposition.
60 Presuppositions and Informativity with Aspectual still presupposition of sentences with still found in the literature. First, following Lo¨bner (1989), Krifka (2000) defines the assertion and presupposition of sentences like (18) as in (24), where U(t) is true iff U is true throughout an interval t, and t } t# iff t# began before t and abuts it (cf. Ko¨nig 1977; Mittwoch 1993): (24)
a. Assertion: John is asleep (t) b. Presupposition: dt# t# } t [John is asleep (t#)]
(25)
Assertion: de,t [t ¼ now ^ asleep (e, john) ^ t O s(e)] (¼ John is asleep) b. Presupposition: de,t#[asleep (e, john) ^ t# } now ^ t# O s(e)] a.
A somewhat different semantics for still is proposed in Ippolito (2007). (26) is a simplified version of Ippolito’s truth conditions for the sentence ‘John is still cooking’, where e1 is a free eventuality variable, whose value is supplied by the context: (26)
‘John is still cooking’ is defined if dt’ < now [t# 4 e1 and time(e1) is a time at which John is cooking] b. If defined, ‘John is still cooking’ is true iff now 4 e1 and time(e1) is a time where John is cooking a.
It is crucial for Ippolito that the eventuality variable in both the presupposition in (26a) and the assertion in (26b) denotes the same (contextually salient) eventuality. According to her, this guarantees the ‘continuity’ feeling found with aspectual still. The sentence ‘John is still cooking’ presupposes that there is a time earlier than the speech time such that the running time of a salient eventuality of John’s cooking includes this past time and asserts of that 15 It is important to notice that the overlap relation between the eventuality time and the reference time will hold for all sentences with still, since, as seen in (i) and (ii), still is compatible with lexical stative or progressivized verbs but not with eventive verbs (see, e.g. Michaelis 1993):
(i) John was (still) on the roof/running (ii) John (#still) spoke with Mary/ran.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Using the more detailed temporal framework reviewed above with speech times, reference times and eventuality times, the assertion of (18), which is identical to the assertion of the still-less sentence ‘John is asleep’, will be represented as in (25a), asserting that there is a sleeping eventuality of John whose time overlaps the speech time (now). Given this, the presupposition of (18) in (24b) can be rephrased as in (25b), saying that there is a sleeping eventuality of John whose time overlaps some time prior to and abutting now:15
Yael Greenberg 61
4 SOME POTENTIAL EXPLANATIONS OF THE DATA Before presenting my own account of the contrast between felicitous and infelicitous occurrences of still, repeated in (27), let us briefly look at some alternative explanations of it: (27)
A: B: B#: B$:
How’s John? He was (#still) unemployed He is still unemployed Last June, he was still unemployed
First, one might try to use the contextual saliency of the eventuality argument, required in Ippolito (2007). Ippolito explicitly claims that ‘out of the blue’ utterances with still, where no such contextually salient eventuality is present, are infelicitous: the sentence [‘John is still cooking’] will be felicitous only if the common ground entails that (a) there is a salient eventuality of cooking by John, and (b) the time of this eventuality includes a past time. Therefore, the sentence John is still cooking cannot be felicitously uttered out of the blue. [p. 10] One might want to claim, then, that the reason for the infelicity of still in (27B) is that although the sentence is not uttered out of the blue, but against the context of A’s question, this context does not supply any salient past eventuality of John being unemployed. However, given the felicity of (27B#) and (27B$) this cannot be right. The only difference between (27B) and (27B#) is tense. Crucially, both are uttered against the same context, and in neither of them is there any reference to a salient eventuality of John being unemployed.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
eventuality that its running time includes the speech time. [. . .] because the assertion is about that very salient eventuality, the sequence [‘Two days ago John was cooking. He is still cooking’] is understood as talking about a single event stretching over 2 days, for the second clause requires that John’s current cooking be salient in the context and that it overlaps a past time (Ippolito 2007: 11). In contrast, in the reformulation of Krifka’s definition in (25) above, the assertion and the presupposition need not necessarily talk about the same eventuality of John’s sleeping (or cooking), and the continuity intuition is captured by explicitly using the abutting relation. In addition, unlike Ippolito, who insists that the eventuality argument in sentences with still is contextually salient, no such requirement is made in the definition in (25).
62 Presuppositions and Informativity with Aspectual still
(28)
The baby is still asleep. Please be quiet!
In addition, we cannot attribute the infelicity of (27B) to a general difficulty in accommodating presuppositions, of the sort which has been claimed for the presuppositions of too and again (see, e.g. Zeevat 2003; Geurts & van der Sandt 2004). The felicity of (27B#) and (27B$) indicates that no such general problem is found with still. The felicity of (28) also shows that the ‘prior time’ presupposition triggered by still can be accommodated like other types of presuppositions, for example those triggered by definite (like the baby). Another promising idea is to try and attribute the contrast in (27) to the tendency of sentences with still to express surprise that the state (referred to by the VP) continues and has not stopped yet.16 For example, ‘John is still asleep’ seems to implicate or presuppose surprise at the fact that John is asleep now, although he was no longer supposed to be asleep. One might want to claim, then, that this kind of ‘surprise’ implication can only arise when the reference time is specified. For example, it is not surprising that there is a time at which John continued to be asleep, if he was asleep, but it may be surprising now, or at 5 p.m. or when Harry arrived. Perhaps, the lack of ‘surprise’ effect is what causes the infelicity of (27B). However, as has been already noted in, for example, Lo¨bner (1999) and Krifka (2000) inter alia, the ‘surprise’ effect seems to be cancellable implication of sentences with still. Consider (29a,b): (29) 16
a.
Unsurprisingly, John was still angry this morning. (This was expected given the way he was treated yesterday).
See, for example, van der Auwera (1993), who formulates this implication as falsity of the asserted state in the ’worlds of expectations’. Thanks to an anonymous reviewer for pointing out this potential explanation.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Similarly, (27B$) differs minimally from (27B) in that the reference time is said to be last June (instead of being unspecified). This sentence too does not make the eventuality of John being unemployed any more salient than (27B) does. The problem with (27B), then, has to do with the saliency of the reference time not that of the eventuality argument. In general, then, we see that the eventuality referred to by the Verb Phrase (VP) need not be salient in order for still to be felicitous. In fact, this can be seen even in completely ‘out of the blue’ utterances with still. (28) can be the first thing I say to a guest of mine as I open the door for her. Crucially, even if the guest doesn’t know in advance that the baby was asleep (or, for that matter, that I have a baby), she could easily accommodate this information:
Yael Greenberg 63
b. c.
Given the fact that you went to sleep so late yesterday night, it is not surprising that you are still very tired. Usually John is still asleep at 7.00 (Unexpectedly today, he is awake)
(30)
A: How’s John? B: Unsurprisingly, he was (#still) unemployed. B#: Unsurprisingly, he is (still) unemployed.
It seems, then, that a different direction is needed in order to explain the infelicitous occurrence of still. In the next section, I turn to develop such a direction. 5 THE ANALYSIS
5.1 A modified, DRT style, definition of still Since we will be dealing here with updating reference times in different contexts, it will be very useful to couch the semantics of still within a DRT framework, in which such pieces of information are easily represented. I assume familiarity with DRT and will make the following simplifications and notational decisions: I will use a bracket notation, as in Geurts (1999), in which the variable e is used for all 17 Both attribute it to the interaction between the semantics of still and the Gricean maxim of relevance. 18 One may argue that the ‘surprise’ effect with still should be always evaluated with respect to a certain perspective, so that even with explicit adverbials like ‘unsurprisingly’, it is possible to have in mind a perspective with respect to which it is in fact surprising that, for example, John continues to be angry this morning or continues to be asleep at 7.00 [in (29a) and (29c), respectively]. If this direction is taken, however, we should be able to say very precisely why it is that the existence of a perspective for a surprise effect is always possible with specific or anaphoric reference times, but systematically blocked with existentially closed ones. Such problems may be solved once a precise theory of the surprise effect with still is developed. At the present stage, however, this direction does not seem to provide a satisfying way of accounting for the newly observed data described in section 2.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The sentences in (29) are perfectly felicitous, despite the fact that the ‘surprise’ implication seems to be explicitly cancelled in them. Thus, as argued in Lo¨bner and Krifka, this kind of component does not seem to be an inherent part of the semantics of still but is a conversational implicature.17 It would be strange to claim, then, that the striking infelicity of (27B) is due to the fact that this cancellable implicature does not arise. Moreover, as seen in (30), the felicity contrast between past tense and present tense sentences with still remains the same even in cases in which the ‘surprise’ implication is cancelled. This is unexplained if the reason for this contrast is the existence or absence of the ‘surprise’ implication:18
64 Presuppositions and Informativity with Aspectual still
(31)
a. John is asleep b. [n, t, e: john asleep(e), t ¼ now, t O s(e)]
In (31b), the reference time of the clause, namely t, is equated with the speech time (‘now’), since the sentence is in the present tense, and is taken to overlap the eventuality time, since the type of eventuality is stative. After existential closure, the sentence asserts that there is a sleeping eventuality of John whose time overlaps the speech time interval (i.e. now). Having these notational issues in mind, I propose the following modified definition of the assertion and presupposition of still (where tr and tps stand for the reference time of the sentence and the time variable introduced in the ‘prior time’ presupposition, respectively): (32) Assuming a clause U, with reference time tr and a predicate P with an eventuality e s.t. P(e), (a) still U is defined iff the universe of U has a time interval tps. which meets the following two conditions: tps. < tr, tps. O s(e) (b) If defined, then still U is true iff [tr,e: P(e), tr O s(e)] Notice that, as in the reformulation of Krifka’s definition in (25) above, here too the assertion of still , given in (32b), is equivalent to the assertion of the corresponding U without still. Thus, the only thing we have to add to the DRS of a sentence with still is the ‘prior time’ presupposition, given by the definedness condition in (32a). Following Geurts’ (1999) notation of underlying presupposed material, then, the DRS of the present tense (33a) will be (33b), which after existential closure will give us (33c):
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
eventualities and (ignoring tense for a moment) the Discourse Representation Structure (DRS) of John ran will be [e: john-run(e)] instead of the more precise [e,x: x ¼ john, run(e,x)] (or the even more precise [e,x: x ¼ john, run(e), agent(e) ¼ x]). This is because the focus of this paper is on the relationship between times, not individuals. For this reason, I will not try to systematically capture nominal anaphora. For example, I will represent the second sentence of John left the room. He was tired as [e: john-tired(e)] (again ignoring tense). I will take the relationship between the reference time and the imperfective eventuality time (e.g. running time of lexical statives and progressives) to be O(overlap) instead of inclusion (4) (though using inclusion will be just as compatible with my claims below). For illustration, let us look at the representation of the present tense (31a) in (31b):
Yael Greenberg 65
(33)
a. John is still asleep b. [ n, t, e, tps.: john-asleep(e), t ¼ now, t O s(e), tps.< t, tps.O s(e)] c. de[John-asleep(e) ^ n O s(e) ^ dt# t# < n ^ t# Os(e)]
5.2 Accounting for the data 5.2.1 ‘Still’ with existentially closed reference times We are now in a position to explain the infelicitous occurrences of still described above. In section 3.1.1 above, we suggested that the reference time of the past tense still-less (34a) is existentially closed. Starting from the ‘unrestricted’ version of the DRS in (16) above, repeated here as (34b), adding still [as in (34c)] will give us DRS (34d): (34)
a.
(How is John?)—He was unemployed (but now he has a job) b. [n, e, t: John-unemployed (e), t < n, t O s(e)] c. (How is John?)—#He was still unemployed (but now he has a job) d. [n, t, e, tps.: john-unemployed (e), t < n, t O s(e), tps.
According to (34d), (34c) asserts that an eventuality where John is unemployed temporally overlaps some past time and presupposes that 19
van der Sandt (1992) and Geurts (1999) assume that accommodated presuppositions are projected as high as possible. This is irrelevant in (33b) since the home DRS of the presupposition is also the top DRS here.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(33b) asserts that some eventuality where John is asleep overlaps now and presupposes that this eventuality overlaps some time prior to now as well. Since, using van der Sandt’s (1992) and Geurts’ (1999) terminology, there is no ‘antecedent’ for the ‘prior time’ presupposition in the context or in previous utterance, it is accommodated, that is added to the DRS of the clause.19 The proposal is very similar to previous proposals about the semantics of still, but it is not identical to them. Specifically, the definition of still in (32) follows Ippolito’s (2007) idea that the presupposition and the assertion necessarily talk about the same eventuality. In addition, as in Ippolito, the temporal relationship between the ‘prior time’ (tps.) and the reference time is defined as anteriority and not as the stricter ‘abutting’ relationship used in Krifka’s (2000) definition. On the other hand, given our discussion in section 4, the definition is unlike Ippolito’s and like the reformulation of Krifka’s definition in (25), in allowing the e variable to be existentially closed and not requiring it to denote a contextually salient eventuality.
66 Presuppositions and Informativity with Aspectual still
(35a)
Unemployed (j) ————————— - - - - - - - - - - - - - -t- - - - - - - - - - - - - - - - - - - - -now
Crucially, this information is enough to guarantee that both the assertion and the ‘prior time’ presupposition of (34c) (‘John was still unemployed’) are met. More specifically, given the information in (35a) one can automatically infer (35b): (35b)
Unemployed (j) ————————— - - - - - - - - ————————— - - - - - -t- - - - - - - - - - - - - - - - - - - - -now
t$j t# ————————— Given the information in (35b), we can infer both (a) that there is a past subinterval of t, t#, which overlaps a ‘John-unemployed’ eventuality (namely, the assertion of John was still unemployed), as well as (b) that there is another past subinterval of t, t$, such that t$ < t# which also overlaps a ‘John-unemployed’ eventuality (namely, the presupposition of ‘John was still unemployed’). The result is, then, that by accepting the truth of (34a) (‘John was unemployed’), we can automatically infer both the assertion as well as the ‘prior time’ presupposition of (34c) (‘John was still unemployed’). But, crucially, this ‘prior time’ presupposition is the contribution of still to the sentence (remember that the assertion of still p is just like that of p).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
this eventuality temporally overlaps some time prior to that past time as well. More intuitively, the meaning we get is that a state of John being unemployed continues at some past time. Why is this meaning infelicitous, then? The reason, I suggest, is that it is not informative. Intuitively, if the denotation of t is not salient, then saying that a state P continues at t (the information of the sentence with still) does not add anything to saying that a state P holds at t (the information of the sentence without still). More precisely, in such a case, both the assertion and the prior time presupposition of still U can be met in the still-less U as well. Consequently, the use of still is unjustified and thus its presence is infelicitous. To see this more clearly, consider the schema in (35a) for the stillless (34a), where there is some past time t which temporally overlaps a John-unemployed eventuality. Given the indicated context, the denotation of t is not being made salient in any way. We further assume that states are not temporally instantaneous, for example that the time of being unemployed is a non-singleton interval:
Yael Greenberg 67
Thus, both this presupposition, as well as the assertion, are met without still, so using still is unjustified and vacuous, and hence, infelicitous. Notice that the same problem remains when we take the existential quantification over the reference time of (34a) to be restricted, as in (17) above, repeated here as (36): (36)
[n, e, t: John-unemployed (e), t < n, t D, t O s(e)]
(37a)
Unemployed (j) ————————— - - - - - - -————————— – - - - - -t- - - - - - - - - - - - - -now j________________________. . . . ) period where John is alive and grown up
Now adding the prior time presupposition to the ‘restricted’ (36) will give us DRS (38): (38)
[n, t, e, tps.: john-unemployed (e), t < n, t D, t O s(e), tps.
(38) says that some past time t, inside a relevant period D, overlaps some John-unemployed eventuality and that some time tps., prior to t, which is also in that period D,20 overlaps this eventuality as well. The problem is that, as with the unrestricted version of (34a), here too the use of still is uninformative. This is because here too we can always ‘divide’ the reference time of the still-less version of the sentence into a past time t# and a past time prior to t#, t$, both overlapping the time of the unemployed eventuality. This is seen in (37b): 20 Notice that since tps is required to overlap the unemployed eventuality as well, it must be included in D, just like t. This is because of the nature of the restriction D, as being a period where being unemployed can be true and relevant at all. Assuming that tps. is not in D will lead to the unreasonable scenario where John is unemployed before he is alive and/or grown up.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As explained in section 3.1.2 above, the idea of having the restriction D in (36) is that the unrestricted existential quantification in (34b) seems too permissive: In thinking about some time in the past which overlaps a John-unemployed state, we clearly do not talk about, for example, a time where John was not born yet or where he was a baby. Thus, D in (36) is characterized by the presuppositions and implications of the sentence and can be thought of as, for example the period where John is alive and grown up, which properly includes the eventuality time where John is unemployed. Given these considerations, the still-less (34a) can be represented as in (37a):
68 Presuppositions and Informativity with Aspectual still (37b)
Unemployed (j) ———————— - - - - - - - - - - ———— - - -t- - - - - - - - - - - - - - - now
t$ j t# ———— j________________________. . . . ) D In such a case too, then, we can infer both the assertion and the presupposition of the version with still [namely, DRS (38)] from the still-less version [DRS (36)]. Thus, the uninformativity and hence the infelicity of still are accounted for.
(39)
a.
(How’s John?). Well, a month ago he was (still) unemployed. b. At 6.00 p.m., I was (still) in my room. Here, the reference time is anaphoric to the salient time which is a month before the speech time and to 6.00 p.m., respectively. We know that states overlap and do not necessarily surround their reference time (see again the discussion in section 3.2.1). The prior time presupposition, then, rules out the possibility that the state starts only at the point reference time and ensures that it holds before it. Consequently, the use of still, which triggers the ‘prior time’ presupposition, is informative and felicitous. For illustration, consider the DRS for the still-less version of (39b) in (40a) and two possible schemas for it, that is (40b) and (40c): (40) a. [n, t, e: I-in my room (e), t < now, t ¼ 6, t O s(e)] b.
c.
in my room (I) ————————— - - - - - - - - - -6- - - - - - - - - - - - - - - now in my room (I) ———————— - - - - - - - - - -6- - - - - - - - - - - - - - - now
(40a) requires that the ‘being in the room’ eventuality temporally overlaps the reference time, which, in this case, is 6.00. As claimed in section 3.2.1 above, this is compatible with a case like (40b), where I am in my room before 6 as well, and also with a case like (40c), where I being in the room started at 6.00.21 Adding still yields the DRS in (41): 21
It is also compatible, of course, with a situation where being in my room equals 6 or ends at 6.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
5.2.2 Still with point adverbials and punctual reference times In contrast to the infelicitous presence of still in (34c), consider the perfect felicity of this particle in (39a) or (39b):
Yael Greenberg 69
[ n, t, e, tps.: I-in my room (e), t < now, t ¼ 6, t O s(e), tps.
5.2.3 Still with frame adverbials and non-punctual reference times One may hypothesize at this point that still is felicitous in, for example, (39b), (42) and (43) above because the reference times in these sentences are ‘short’ or punctual so, unlike the reference time in the infelicitous past tense (35c) above (‘#John was still unemployed’), they cannot be ‘divided’ into two successive subintervals. But this hypothesis is refuted by the felicity of still with ‘frame’ temporal adverbials, as in examples (6) and (7) above, or as in (44) or
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As explained in section 3.2.1 above, the still-less versions are compatible with scenarios in which the eventuality time starts at the speech time. The still-less version of (42a), for example, can be uttered just as the plane takes off and the still-less version of (42b) can be uttered even when the baby was asleep until now. The addition of still excludes these situations and thus its use is informative and felicitous. There are, of course, other cases of contextually salient times. In a context where we talk about Yitzhak Rabin’s murder, the reference time of a sentence like (43) will be anaphoric with the time of the murder, so still will be informative and felicitous too: (43) John was (still) in London.
70 Presuppositions and Informativity with Aspectual still (45). Such adverbials denote ‘long’, non-punctual times, which can, in principle, also be ‘divided’ into two successive subintervals: (44) (45)
In the summer, Anne was (still) writing her book. Mary was (still) in France in June.
(46) (47)
[n, t, e: Mary-in-France(e), t < n, t 4 June, s(e) O t] [n, t, e, tps.: Mary-in-France(e), t < n, t 4 June, s(e) O t, tps
But if we follow (47), it is not clear why (45), unlike the past tense (35c) above, is felicitous. Following the reasoning described above, here too we should be able to derive the truth of (45) from the truth of its stillless version: Once there is some past interval within June which overlaps e, we can always infer that (a) there is a subinterval of that time, also within June, overlapping with e, and (b) there is a successive subinterval of that time, also within June, which overlaps e as well. The use of still, then, is predicted to be uninformative and hence infelicitous just as in (35c) above. However, unlike (35c), (45) is perfectly felicitous. One possible conclusion can be that our explanation of the infelicity of (35c) above, in terms of the uninformativity of still, is on the wrong track. I will suggest, however, that this explanation is correct, and instead, it is the DRSs in (46) [and consequently also (47)] which are wrong. The main independent reason to suggest this is that a DRS like (46) does not correctly capture the interpretation of sentences like (45): Taking the reference time in this sentence to denote some time which is included in or equal to June, as in (46), is compatible with asserting that Mary was in France from June 17th to June 24th (a time within June) and presupposing that she was also in France before that time, for example from June 13th to June 17th. But the real presupposition of (45) is that Mary was in France before June, that is at least during the last day(s) of May. This is evident from the infelicity of (48):
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
To understand why still is felicitous in these sentences, we first have to examine their temporal structure. There are, in fact, two views in the literature regarding the representation of frame adverbials, and as we shall see now, the interpretation and felicity of still with such adverbials can help us decide between them. According to Pancheva & von Stechow (2004), for example, frame adverbials as in (44) and (45) denote time intervals which include or are equal to the reference time of the sentence. Given this view, the still-less version of, for example, (45) would have the DRS as in (46). Adding still would give us (47), asserting that there is an eventuality of Mary being in France which overlaps some past time t included in (or equal to) June and presupposing that this eventuality overlaps some time prior to that past time t:
Yael Greenberg 71
(48)
#Mary was still in France in June. She arrived there on June 13th.
In the indicated contexts, the reference times of these sentences are t < n, t ¼ the middle ages, and t < n, t ¼ my sabbatical, respectively. Thus, both sentences can be paraphrased by adding the deictic adverbial then to them. We predict that adding still to such sentences will be felicitous, even without the addition of an explicit ‘then’. This prediction is borne out:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
A better representation of (45), then, should take the denotation of the adverbial to be identical to the reference time, as suggested in, for example, Kamp & Reyle (1993) and Artstein (2005). For example, (49) and (50) would be the DRS of (45b) without and with still, respectively: (49) [n, t, e: Mary-in-France(e), t < n, t ¼ June, s(e) O t] (50) [n, t, e, tps.: Mary-in-France(e), t < n, t ¼ June, s(e) O t, tps.
72 Presuppositions and Informativity with Aspectual still (53) (54)
(Context: we are in a history class dealing with the middle ages) The lecturer: Don’t forget that people still believed in ghosts! (Context: we are talking about John’s sabbatical many years ago.) John: I was still an associate professor.
5.2.4 One eventuality, not two, in the definition of still The data with frame adverbials, examined above, support Ippolito’s (2007) claim, adopted 22 This difference between ‘relevant’ periods, which can function as domain restrictions on the default existential quantification, and ‘contextually salient’ ones, which cannot, raises several interesting questions, which are, unfortunately, outside the scope of this paper. First, it is important to check whether this distinction holds with respect to other types of quantification over times as well (e.g. adverbial quantification, see, e.g. Partee 1984). Second, it is interesting to check whether this distinction in the temporal domain carries over the nominal domain. At least on the surface the answer seems to be negative. That is, default existential quantification over individuals can be restricted by contextually salient domains. In the context of talking about my Semantics class, for example, the interpretation of (i) seems to be ‘a student in my class asked me for an extension’:
(i)
Yesterday a student asked me for an extension.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Since in both cases the reference time is anaphoric to the contextually salient period, still is informative and felicitous. It triggers the presupposition that the state holds also before that period, that is that the church believed in ghosts before the middle ages and that I was an associate professor before my sabbatical as well, respectively. It is important to notice, then, that contextually salient periods are different from ‘relevant’ periods, of the sort we looked at in the restricted reading of ‘John was unemployed’. ‘Relevant periods’ (e.g. the period where John is alive and grown up, in the case of considering ‘John was unemployed’) are triggered by the presuppositions and/or implications of the sentence and constitute domain restrictions, D, on the existential quantification over the reference time (i.e. we assume that t D). In contrast, as we saw above, contextually salient periods [as in examples (51) and (52)] do not constitute domain restriction on the existential quantification over the reference time, but rather give the value of the reference time just like the denotation of explicit frame adverbials (and similar to denotation of punctual adverbials). That is, if PC is a contextually salient period and t is the reference time of the sentence, then we take t ¼ PC and not t PC.22 The difference between the two types of periods can be tested by paraphrases with the deictic adverbial then. Unlike (51) and (52), with contextually salient periods, where using then is natural, when we have implicit domain restricting the existential quantification over times which is not contextually salient but ‘relevant’, the use of then is odd. This is seen in (55) (assuming, as before, that A’s question is the only context for B’s answer): (55) A: How’s John? B: Well, he was unemployed (#then).
Yael Greenberg 73
23
(i)
Such a situation, though, will make (i), with again, fine; In June Mary was in France again.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
in the definition in (32) above, that the assertion and presupposition triggered by still make use of one eventuality. To see why this is the case suppose that instead of (32), the definition of the assertion and presupposition of still was (56), where the presupposition of still makes reference to a new P eventuality variable and the temporal continuity between the presupposed and the asserted eventuality (which in Ippolito’s theory are captured by using a single state) is guaranteed by explicitly using Krifka’s (2000) abutting relation: (56) An alternative semantics for still (with no guarantee for one eventuality): Assuming a clause U, with reference time tr and a predicate P with an eventuality e s.t. P(e), (a) still U is defined iff the universe of U has a time interval tps. and an eventuality e#, which meet the following conditions: P(e#), tps. a tr, tps. O s(e#) (b) If defined, then still U is true iff [tr,e: P(e), tr O s(e)] Given this alternative definition of still and assuming, as before, that the DRS of the still-less (57) is (58), the DRS of (57) with still would be now (59) [instead of (50) above]: (57) In June Mary was (still) in France (58) [n, t, e: Mary-in-France(e), t < n, t ¼ June, s(e) O t] (59) [n, t, e, tps. e#: Mary-in-France(e), t < n, t ¼ June, s(e) O t, tps a t, Mary-in France (e#),s(e#) Otps.]: The representation in (59) asserts that some eventuality e where Mary is in France overlaps some past June and presupposes that some such eventuality, e#, overlaps some time prior to and abutting that June. It is important to notice that although e and e# in this DRS are different eventuality variables, they do not necessarily denote different eventualities. Crucially, however, they can denote two such different eventualities. This possibility leads to predicting incorrect truth conditions for (57), since now the DRS in (59) is compatible with a discontinuous stay of Mary in France, for example with a situation where Mary arrived in France from May 24th to the end of May (which is a time prior to and abutting June), left France and came back to France from June 23rd to June 29th. In such a case, it is true that there is a being-in-France eventuality which overlaps June and that there is a beingin-France eventuality which overlaps a time prior to and abutting June, as (59) requires. The problem is that (57) with still would be infelicitous in such a case, since it presupposes that Mary continuously stayed in France from some time before June to at least some time in June.23
74 Presuppositions and Informativity with Aspectual still
(60)
a. Assertion of (57): John is in France (t) ^ t ¼ last June b. Presupposition of (57): dt#,e t# } t [John is in France (t#)]
Since (60a) can hold if John was in France only part of June (e.g. from June 19th to June 26th), the problem is, again, that (57) is wrongly predicted to be felicitous in a case where John was in France from May 24th to May 31st, left France and returned on June 23rd. In contrast, using an event-based semantics with one eventuality in the assertion and presupposition of still, as done in Ippolito (2007) and adopted here, makes the right predictions.24
24 A potential counterexample to the event-based analysis of still, pointed out by an anonymous reviewer, is (i):
(i)
There is still a guard at the gate.
(i) is felicitous even if the guards change regularly, as long as we have temporal continuity. Assuming that events are individualized by their participants, it would be problematic to maintain that we have one eventuality here, which overlaps both the speech time and a time prior to it, as done in this paper. Notice, however, that a similar problem is raised by still-less sentences as in (ii): (ii) Between 3 and 8, there was a guard at the gate (ii) too can be felicitous if there was one guard from 3 to 6, who was replaced by another guard from 6 to 8. But if indeed having different guards necessarily means having different eventualities, we wrongly predict that (iii) would also be felicitous in this situation (twice seems to indicate a plurality of events, see Rothstein 2004): (iii) Between 3 and 8, there was a guard at the gate twice. But (iii) is infelicitous in this scenario. It can become felicitous if the two different eventualities are characterized not only by different participants but also by temporal discontinuity. In other words, despite the changing of the participants in (ii), it seems to talk about one eventuality, namely one where there was some guard or another at the gate. It seems to me that the same thing happens in (i). It is, of course, challenging to understand this pattern, but this is beyond the scope of the present paper. One possible direction is that what we have here is the necessary summing of two eventualities into a single one [what Rothstein (2004) calls S-summing].
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This indicates, indeed, that the use of still requires one and the same eventuality in both the assertion and the presupposition, as required by Ippolito. Assuming that one eventuality of being in France cannot be discontinuous [see, e.g. Rothstein (2004) on temporal continuity of singular eventualities], it is indeed definition (32) above, with one eventuality variable, which can guarantee the required continuous stayof Mary in France. Similar argumentation can be used to show that defining still in temporal terms only [as in Krifka’s original definition in (24) above], without using an event-based semantics, is not enough. Applying this definition to (57) will give us (60), asserting that the sentence ‘John is in France’ is true at a time which is last June and presupposing that there is some time before last June in which this sentence is true as well:
Yael Greenberg 75
5.2.5 ‘Still’ with familiar existentially closed reference times. Based on the comparison between the felicitous and infelicitous sentences above, we may want to suggest that still can only be felicitous when the reference time is not existentially closed. A closer examination, however, shows that this generalization is still not precise enough. Consider (61): (61)
John opened the door. Mary was (still) undressed. (So she told him to wait in the living room)
(62) (63)
Pedro owns a donkey. He beats it. [n, e, e#, t, t#, tps.: john-open-the-door (e), t < n, s(e) 4 t, mary-undressed (e#), t# < n, t# ¼ s(e), s(e#) O t#, tps.
According to (63), the first sentence in (61) asserts that there is some opening-the-door eventuality by John whose time is included in some past time t. This eventuality time supplies the reference time for the second sentence, which says that there is some Mary-being-undressed eventuality whose (past) time overlaps the time where John opened the door. The reference time of the second sentence, then, is indirectly anaphoric with the existentially closed time of the first sentence. A similar thing seems to happen with discourses like (64): (64)
John was still unhappy. His grandmother gave him a candy, but it didn’t help.
Here still felicitously appears in a past tense sentence which is the first in the discourse. Intuitively, though, the reference time of the first sentence is nonetheless anaphoric or more precisely cataphoric: it is supplied by the eventuality time of the second sentence. This is very similar to cataphora in the nominal domain, as in (65), or (66) cited in Geurts (1999): (65) (66)
When he saw me, John was really surprised. The doctor kept warning him but he would never listen, and now Harry’s dead.
Geurts suggests that in cataphoric cases like (66), the reference to the pronoun him is accommodated in a similar way to the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In (61), the past tense sentence with still appears after another past tense sentence. Partee (1984) discusses similar discourses without still [as in (11) above] and argues that the reference time of the second sentence is anaphoric to the existentially closed time of the first one, in a similar manner to the way pronouns [like it in (62)] can be anaphoric with an existentially closed nominal antecedent. Following this approach, (61) would have the DRS in (63):
76 Presuppositions and Informativity with Aspectual still
6 COMPATIBILITY OF THE PRESENT ANALYSIS WITH CURRENT THEORIES OF PRESUPPOSITIONS In the sections above, we developed the idea that still is infelicitous in sentences like (67B#) since in such a case the prior time presupposition can be trivially inferred from the truth of the still-less version in (67B) and hence its use is uninformative: (67)
A: How’s John? B: He was unemployed B#: #He was still unemployed
I would like to finish this paper by examining the compatibility of this claim with current theories of presuppositions. I will focus here on three general suggestions which may be seen as raising problems for the analysis above. First, a general assumption in the literature, for example Stalnaker (1978), is that presuppositions are supposed to be trivial (relative to the common ground). It is unclear, then, why the triviality of the presupposition in (67B#) leads to infelicity. This may be seen as all the more problematic in view of the recent literature that claims that certain presupposition triggers are obligatory whenever they are allowed 25
The difference is that in (64), we accommodate a time, not an individual, and that the coreferring expression is not definite or a proper name, but existentially closed, namely the existentially closed eventuality time of the second sentence.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
accommodation of standard presuppositions. I assume that a similar thing happens with (64).25 The interesting point, then, is that in both (61) and (64), the reference time of the past tense sentence with still is indirectly interpreted as existential but nonetheless still is felicitous. Given the discussion above, the reason should be clear. Since the reference time of the sentence with still is anaphoric, or cataphoric, to another time (albeit an existential one), the use of still is informative and hence felicitous: For example, the still-less version of (61) is compatible with a situation where Mary finished taking off her clothes just as John opened the door. When the presupposition triggered by still is added, this situation is excluded. In general, then, the precise feature which distinguishes infelicitous from felicitous uses of still in past tense sentences is not the existential/ non-existential interpretation of the reference time, but rather the novel/anaphoric one.
Yael Greenberg 77
(68)
Informativenss: ui+1 must be informative, that is ui+1 may not be entailed by uI
In (68), Si is a sentence uttered by a speaker with DRSui, extended by a hearer to a DRSui+1. This is supposed to account for infelicity in cases where existing context entails the information conveyed in a new sentence added to this context. But as we have just seen, this is not what happens in (67B#), whose content is inferable from a potential alternative.26 26
In addition, unlike the kind of infelicity dealt with in van der Sandt’s and Beaver’s constraints, no sub-DRS is entailed by the main DRS in the infelicitous cases with still above (these have no subDRSs to start with).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
[what is called Maximize Presupposition, discussed in, e.g. Sauerland (forthcoming), Percus (2006) and Schlenker (2006)]. On the other hand, theories like van der Sandt (1992), Geurts (1999) or Beaver (2001) posit ‘informativity constraints’ which exclude uninformative accommodation of presuppositions. On the surface, this kind of requirement may be seen as the one violated in the case in (67B#). However, the ‘informativity constraints’ suggested in these theories are based on the Gricean maxim of quantity (according to Geurts 1999) or manner (according to Beaver & Zeevat forthcoming). As is well known, and as Geurts (1999) himself emphasizes, such maxims, and the informativity constraint based on them, are defeasible, and many times breaking them leads to some conversational implicatures. This is not what happens in (67B#): as long as the reference time is novel, the uninformativity of still does not lead to a conversational implicature but to infelicity which seems strikingly stronger than the result of typical violations of Gricean maxims. If a Gricean-based informativity constraint is indeed the one violated in (67B#), then the hopeless infelicity of such sentences is unexplained. A closer look, however, shows that none of these suggestions actually covers cases like (67B#), so these potential problems do not arise in the first place. First, though it is true that presuppositions are trivial given the common ground, this is not what the analysis above suggests with respect to cases like (67B#). Instead, the ‘prior time’ presupposition in such a sentence can be trivially inferred from the truth of the still-less version in (67B), which is not part of the common ground. In fact, this still-less version is not uttered at all. It is a potential alternative to (67B#). For a similar reason, the Gricean-based ‘informativity constraints’ are not applicable to our case. Consider, for example, Geurts’ (1999) version of the constraint, in (68):
78 Presuppositions and Informativity with Aspectual still
27 The fact that the principle does not seem to apply to such alternatives (with and without still) can be seen from the fact that still in the present tense (i) is not obligatory, even if the speaker knows that John was unemployed before now as well:
(i)
A: How’s John? He is (still) unemployed.
28 How exactly to define presuppositional strength is not completely agreed upon. Schlenker (2006) takes F to carry a stronger presupposition than F# just in case: fw 2 W: F is neither truefwg nor falsefwg in wg fw 2 W: F# is neither truefwg nor falsefwg in wg. But see Sauerland (forthcoming) for several other versions, especially meant to ensure that the principle gives the right results also when the relevant lexical items are found in embedded structures. See also Percus (2006) concerning this issue. 29 Ignoring the presupposition that John exists (which is, of course, triggered in the still version as well).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Finally, though ‘Maximize Presupposition’ does deal with potential alternatives, it does not apply to (67B#) either, for two reasons. First, the principle deals with alternatives which are triggered by lexical items, such as know–believe, the–a, both–every, child–children, ran–run [see Sauerland (forthcoming) for more such lexical items]. In contrast, in the cases discussed in this paper, only one of the alternatives contains a lexical item which triggers a presupposition, namely the version with still (e.g. ‘John was still unemployed’), and the second alternative is simply the same sentence without this lexical item (namely ‘John was unemployed’, which itself has no presuppositions).27 In addition, the alternatives covered by ‘Maximize Presupposition’ are characterized as having an equivalent assertive component (in the relevant context), but as forming a presupposition scale, that is one has a stronger presupposition than the other. In such a case, the principle dictates that one chooses the lexical item with the strongest presupposition.28 In contrast, although indeed both the still and the still-less versions in (67) have the same assertive component, they actually do not differ in their presuppositional strength. This can be seen clearly if we adopt Schlenker’s (2006) characterization of presuppositional strength as a subset relation between the sets of worlds where the proposition is neither true nor false. Crucially, in our case the sets of worlds where (67B) (‘John was unemployed’) and (67B#) (‘John was still unemployed’) are neither true nor false is, in fact, equal: The set of worlds where the still-less version is neither true nor false is empty, since this sentence carry no presupposition at all29 (If there is a past interval which overlaps a John-unemployed state, this sentence is true, and if there isn’t such an interval the sentence is false). Crucially, though, the same holds for the version with still: as long as there is a past interval which overlaps a John-unemployed state, both the assertion and the presupposition are true, since, as we have shown above, in such a case it is true that such an interval exists and it is true that there exists an interval prior to it, which also overlaps such a state. If, on the other hand,
Yael Greenberg 79
7 SUMMARY AND DIRECTIONS FOR FURTHER RESEARCH The goal of this paper was to explain so far unnoted and unanalysed infelicitous occurrences of aspectual still. I showed that when still appears in past tense sentences whose reference time is novel and nonanaphoric, the truth of the assertion and that of the ‘prior time’ presupposition this particle triggers can be both inferred from the truth of the minimally contrasting still-less counterpart. Moreover, in such cases the presuppositional status of the ‘prior time’ claim is lost, despite the presence of still. Hence, the use of still in such sentences is uninformative and thus unjustified and infelicitous. Besides the analysis of the novel data, the paper has several more general implications. First, as far as the semantics of still is concerned, the data analysed here support Ippolito’s (2007) claim that the assertion and presupposition of still make use of one eventuality argument, but in contrast to Ippolito’s theory, suggest that this eventuality argument need not be contextually salient. Second, the analysis has implications for the research of the tense and temporal structure of clauses. I claimed that the interaction of still with frame adverbials supports the view that they denote intervals identical to the reference time of the sentence, rather than including it. I also argued that the contrast between felicitous and infelicitous cases of still can only be explained if, contrary to many current analyses, we assume that past tense is not necessarily anaphoric but can be represented as a new variable bound by existential closure. In fact, given the analysis above, the felicity or infelicity of still in past tense sentences can be seen as a diagnostic for determining whether or not the reference time in such sentences is anaphoric or not.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
no past interval overlapping an unemployed state exists, the sentence is simply false. Put in other words, despite the presence of still in the sentence, in such past tense sentences there is no truth–value gap at all. Thus, none of the alternatives here carries a stronger presupposition. This discussion has two important implications. The first is that the Maximize Presupposition principle is not supposed to apply to cases as in (67). Thus, the fact that still is not obligatory in such sentences is not a problem. Second, the fact that the infelicity of still in such sentences is so strong is perhaps due to the fact that still turns out, in fact, to be doubly uninformative in (67B#), namely uninformative both with respect to the content and with respect to the status of the ‘prior time’ claim: Not only the content of the claim can be inferred from the stillless version, still does not even add the information that the status of this claim is presuppositional.
80 Presuppositions and Informativity with Aspectual still
(69)
(How’s John?) a. He wasn’t (#still) unemployed b. He isn’t (#still) unemployed
However, there are some kinds of negative sentences which do allow the presence of still. An example is the extra-clausal negation ‘it isn’t true that’. Crucially, though, this holds for the present tense version only [as in (70b)]. The past tense (70a) seems as infelicitous as its positive counterpart in (67B#) above: (70)
30
(How’s John?) a. Well, It isn’t true that he was (#still) unemployed b. Well, it isn’t true that he is (still) unemployed
Thanks to a reviewer for pointing out this potential problem to me.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The analysis above raises several open questions. For example, above we concentrated on explaining the behaviour of still in matrix positive sentences. An obvious extension of the theory is to try and check whether the contrast between the occurrence of still with novel and familiar reference times, and the informativity-based explanation of this contrast, can be carried over to other, more complex structures. Although this step is beyond the scope of the present analysis, I would like to finish the paper by examining one such construction, namely negative sentences. At least on the surface, the behaviour of still in negative sentences seems to pose a problem for the analysis above.30 Presuppositions are known to project out of the scope of negation. Assuming the informativity-based proposal suggested above, then, one would predict that negative sentences with still will be felicitous even if their reference time is novel, since the problem of uninformativity seen before should not arise here: Unlike the ‘positive’ sentences above, in such sentences the presupposition, which remains ‘positive’, would not be able to be inferred from the truth of the negated still-less assertion. The prediction seems to be, then, that existential past tense negative sentences with still should be felicitous, unlike their positive counterparts. Is this prediction borne out? There is an obvious complication with answering this question: still has been known for a long time to be a positive polarity item, that is to be odd under the scope of negation (see, e.g. Baker 1970; Israel 1996, 1997; Szabolsci 2004). Thus, the fact that still is infelicitous in the past tense (69a) should not be attributed to the novelty of the reference time in this sentence, since it is just as infelicitous in the present tense (69b), with a familiar reference time:
Yael Greenberg 81
(71)
a. The sox have already clinched the pennant b. The sox haven’t already clinched the pennant
Although (70a) differs from (71b) in that it does not deny the preceding utterance (since it is an answer to a question), we can still assume that the use of the extra-clausal ‘it isn’t true that . . .’ indicates denial of a contextually salient positive counterpart as well.31 If this is indeed the reason for the infelicity of still in the past tense (70a), we would expect this kind of contrast in other cases where we get denial of an infelicitous positive counterpart with still. One possible candidate is predicate negation with heavy contrastive stress, as in (72): (72)
(How’s John?) He ISn’t (still) unemployed!
The stress pattern here seems again to express denial of a positive counterpart. A similar claim is made by Horn (1989) and Szabolsci (2004) with respect to the Positive Polarity Items (PPIs) someone. Like still, someone is odd with normal negation [as in (73)], but fine if 31
Horn (1989) takes the cleft construction ‘It’s not that . . .’, which is very similar to the ‘it is not the case that . . .’ construction in (70) to be one of the typical ways to express meta-linguistic or ‘contradiction’ negation.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Put in other words, we find here exactly the same kind of contrast between sentences with familiar and novel reference times as we found in positive sentences. Given that presuppositions are known to scope out of negation, this contrast may be seen as a problem for the informativity-based proposal developed above. There is, however, a potential solution to this problem. A typical conversational use of extra-clausal negation, as in (70), is to express denial or contrast with a contextually salient positive counterpart. If I utter the still version of (70b), for example, I seem to deny a contextually salient alternative saying that John is still unemployed. The infelicity of still in the negated (70a), then, could arise because negation indicates the existence of an infelicitous salient positive counterpart in the context, namely ‘#John was still unemployed’. This kind of solution is supported by an old observation by Baker (1970), who notes that many positive polarity items (like already, would rather or the adverb pretty), which normally cannot occur under the scope of negation, become much better ‘when they represent, word by word, an emphatic denial of a preceding speaker’s assertion’ (p. 169). Thus, for example, whereas (71b) is odd in the null context, it is felicitous when understood as an emphatic denial of (71a):
82 Presuppositions and Informativity with Aspectual still negation ‘is construed as an emphatic denial of a similarly phrased statement’ (p. 413), as in (74): (73) (74)
He didn’t find #something/anything. He found something. Wrong! He DIDn’t/DID NOT find something. [Szabolsci 2004: 413]
We could predict, then, that as long as still appears with contrastive stress on negation, it would be infelicitous in the past tense, just as in positive sentences. This prediction is borne out, (75) seems worse than the present tense (72):32 (How’s John?) He WASn’t (#still) unemployed!
Notice, however, that it is not clear whether and how this kind of explanation can be carried over the behaviour of still in other types of negative sentences, as in, for example, (76): (76)
a. I doubt that John is/#was still unemployed b. None of my friends is/#was still unemployed
32 A reviewer expressed the worry that assuming this kind of argument might lead one to wrongly predict that sentences like (ia) and (ib) should be unacceptable because their positive counterparts, namely (iia) and (iib), are ungrammatical: (i) a. John didn’t see any bird b. John hasn’t ever arrived
(ii) a. *John saw any bird b. *John has ever arrived However, as noted above, regular predicate negation, as in (i), is not (necessarily at least) taken to express denial of a salient positive proposition. It is expressions like ‘It is not true that . . .’ or focused negation which typically have this conversational role. That is, we need instead to check the felicity of sentences as in (iii) and (iv). Indeed, these are not as good as those in (i): (iii) a. #It is not true that John saw any bird b. #/?It is not true that John has ever arrived (iv) a. #John DIDn’t see any bird! b. #/?John HASn’t ever arrived! This contrast is very similar to the well-known observation (e.g. Karttunen & Peters 1979; Horn 1989) that what they sometimes call ‘contradiction negation’ (or meta-linguistic negation) cannot license negative polarity items as in (v), from Karttunen & Peters: (v) Chris didn’t MANAGE to solve some/*any problem. He solved them easily The exact explanation for this pattern is under debate [see, e.g. Horn (1989) for an overview]. Thus, it is not clear that the reason for the oddness of NPIs with ‘contradiction negation’ is the indication of an existing infelicitous positive counterpart [as suggested above for the oddness of still in (70a)] or the fact that this kind of ‘meta-linguistic negation’ cannot count as a licensor for NPIs for semantic or syntactic reasons (see, e.g. Linebarger 1987). But in any case, it seems that the felicity of the NPIs with predicate negation in (i) does not undermine the ‘conversational’ kind of explanation for the infelicity of still with the cases of ‘contradiction negation’ above.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(75)
Yael Greenberg 83
c.
Every boy who isn’t/#wasn’t still sick can join the trip to the mountains
33 One parallel which comes to mind is suggested in Chierchia (2006). Similarly to the kind of argumentation proposed above, Chierchia argues that the ungrammaticality of NPIs like any in positive sentences (as in ‘*I saw any student’) is due to the fact that the main contribution of any is domain widening, which, in positive sentences, is ‘purposeless’ and uninformative. Chierchia, however, is well aware of the fact that this intuitive explanation is not enough since it is not clear ‘why pragmatically driven conditions, which usually can be overridden, give rise . . . to unsavable grammaticality contrasts’ (p. 538). The paper attempts to answer this question by using Chierchia’s recursive pragmatics approach. Further research should check whether this kind of formal framework can be used to explain the infelicitous case of still as well.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In such sentences too, we find the present tense/past tense contrast found with matrix positive sentences, above. It may be possible to use again the ‘conversational’ explanation for the infelicity of the past tense version of (76a), namely to claim that ‘I doubt that John was still unemployed’ is infelicitous since the use of doubt that p indicates that p (in our case the infelicitous ‘John was still unemployed’) is contextually salient. However, this kind of explanation does not seem to be carried over to the infelicity of the past tense versions of (76b) and (76c). I believe, though, that in order to understand better the behaviour of still in such sentences, one first has to understand better the interaction of the focus sensitivity properties of still with different types of ‘negative’ sentences. Szabolsci (2004) shows that the PPI someone interacts in a rather complicated way with various types of ‘negative’ operator [e.g. anti-morphic, anti-additive and monotone decreasing operators, in van der Wouden’s (1994) terminology], both in terms of distribution and interpretation. As far as I know no similar work has been done on the PPI still. It may very well turn out that the polarity sensitivity properties of still, and consequently its distribution and interpretation with various ‘negative’ operators, are even different from that of someone (just as in the realm of Negative Polarity Items (NPIs), not all PPIs are equivalent). Further research, then, should first attempt to characterize these properties with familiar reference times (e.g. in present tense). Only then can one start evaluating whether the incompatibility of still with past tense, as in (76), and in other embedded constructions does or does not constitute a real problem for the informativity-based analysis developed above. In addition, further research should check whether the infelicity of still discussed above is part of a larger pattern of uninformative uses of presupposition triggers (like already or anymore) or particles in general.33 Finally, the interaction of still with other tenses and aspects (e.g. future tense and perfect aspect) should be compared to its interaction with the simple past.
84 Presuppositions and Informativity with Aspectual still Acknowledgements I would like to thank Ariel Cohen, Edit Doron, Manfred Krifka, Angelika Kratzer, Anita Mittwoch and especially Susan Rothstein for their helpful comments concerning this and earlier versions of the paper. Thanks also to the audiences of 4es Rencontres de Semantique et Pragmatique, (Universite d’Orleans, France), the 22nd annual conference of the Israeli Association for Theoretical Linguistics (Jerusalem, Israel) and the 9th Symposium on Logic and Language (Besenyotelek, Hungary) for their comments. Special thanks to three reviewers of JOS, whose constructive comments lead to important improvements of the paper. This research was partially supported by the Israel Science Foundation Grant #951/03.
REFERENCES Abusch, D. (1997), ‘Sequence of tense and temporal de re’. Linguistics and Philosophy 20:1–50. Artstein, R. (2005), ‘Quantificational arguments in temporal adjuncts clauses’. Linguistics and Philosophy 28:541–97. Avrutin, S. & Reuland, E. (2004), Backward anaphora and tense interpretation. In: L. Verbitskaya et al (eds.), Teoreticheskie Problemy Jazykoznanija (‘Theoretical Issues in Language Research’), St. Petersburg University Press. Baker, C. L. (1970), ‘Double negatives’. Linguistic Inquiry 1:169–86. Beaver, D. (2001), Presupposition and Assertion in Dynamic Semantics. CSLI Publications. Stanford, CA. Beaver, D. & Zeevat, H. (2007), ‘Accommodation’. In G. Ramchand and C. Reiss (eds.), Oxford Handbook of Linguistic Interfaces. Oxford University Press. Oxford.
Beck, S. (2006), ‘Focus on again’. Linguistics and Philosophy 29:277–314. Bonomi, A. (1995), ‘Tense, reference and temporal anaphora’. Lingua e Stile 30:483–500. Chierchia, G. (2006), ‘Broaden your views: implicatures of domain widening and the ‘‘logicality’’ of language’. Linguistic Inquiry 37:535–90. Comrie, Bernard. (1985), Tense. Cambridge University Press. Cambridge. Dowty, D. (1986), ‘The effects of aspectual class on the temporal structure of discourse: semantics or pragmatics?’. Linguistics and Philosophy 9:37–61. Geurts, B. (1999), Presuppositions and Pronouns, vol. 3. CRiSPI. Elsevier. Amsterdam. Geurts, B. & van der Sandt, R. (2004), ‘Interpreting focus’. Theoretical Linguistics 30:1–44. Heim, I. (1982), The Semantics of Definite and Indefinite Noun Phrases.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
YAEL GREENBERG The English Department Bar Ilan University Ramat Gan 52900 Israel e-mail:
[email protected]
Yael Greenberg 85 Ko¨nig, E. (1977), ’Temporal and nontemporal uses of ‘noch’ and ‘schon’ in German’. Linguistics and Philosophy 1:173–98. Kratzer, A. (1998), ‘More structural analogies between pronouns and tenses’. SALT VIII. Eds. D. Strolovitch and A. Lawson. GLSA Amherst, 1998. 92–110. Krifka, M. (2000), ‘Alternatives for aspectual particles: semantics of still and already’. In proceedings of the annual meeting of the Barkely Linguistics Society, 26, p. 401–412. Lascarides, A. & Asher, N. (1993), ‘Temporal interpretation, discourse relations, and commonsense entailment’. Linguistics and Philosophy 16:437–93. Linebarger, M. C. (1987), ‘Negative polarity and grammatical representation’. Linguistics and Philosophy 10:325–87. Lo¨bner, S. (1989), ‘German schon–erst– noch: an integrated analysis’. Linguistics and Philosophy 12:167–212. Lo¨bner, S. (1999), ‘Why German schon and noch are still duals: a reply to van der Auwera’. Linguistics and Philosophy 22:45–107. Michaelis, L. (1993), ‘Continuity across scalar models: the polysemy of adverbial still’. Journal of Semantics 10:193– 237. Mittwoch, A. (1993), ‘The relationship between schon/already and noch/still: a reply to Lo¨bner’. Natural Language Semantics 2:71–82. Moens, M. (1987), Tense, Aspect and Temporal Reference. Ph.D. dissertation, University of Edinburgh. Edinburgh, Scotland. Moens, M. & Steedman, M. (1987), ‘Temporal ontology in natural language’. In Proceedings, 25th Annual Meeting of the Association for Computational Linguistics. Stanford University, Stanford, CA. 1–8.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Ph.D. thesis, University of Massachusetts, Amherst, MA. Heim, I. (1994), Comments on Abusch’s Theory of Tense. In Ellipsis, Tense and Questions, ed. Hans Kamp. 143–170. DYANA deliverable R2.2.B, University of Amsterdam. Hinrichs, E. (1986), ‘Temporal anaphora in discourses in English’. Linguistics and Philosophy 9:63–82. Horn, L. (1989), A Natural History of Negation. Chicago University Press. Chicago. Ippolito, M. (2007), ‘On the meaning of some focus sensitive particles’. Natural Language Semantics 15:1–34. Israel, M. (1996), ‘Polarity sensitivity as lexical semantics’. Linguistics and Philosophy 19:619–66. Israel, M. (1997), ‘The scalar model of polarity sensitivity: the case of the aspectual operators’. In D. Forget, P. Hirschbu¨hler, F. Martineau, and M.-L. Rivero (eds.), Syntax and Semantics. John Benjamins. Amsterdam. 209–30. Kamp, H. (1981), ‘A theory of truth and semantic representation’. In Jeroen A. G. Groenendijk, T. M. V. Janssen, and Martin B. J. Stokhof (eds.), Formal Methods in the Study of Language. Mathematical Center Tract 135. Amsterdam. 277–322. Kamp, H. & Reyle, U. (1993), From Discourse to Logic. Kluwer. Dordrecht. Karttunen, L. & Peters, S. (1979), ‘Conventional Implicature’. In C-K. Oh and D. Dineen (eds.), Syntax and Semantics, vol. 11: Presupposition, Academic Press, New York, pp. 1–56. Kehler, A. (2000), ‘Resolving temporal relations using tense meaning and discourse interpretation’. In M. Faller, S. Kaufmann, and M. Pauly (eds.), Formalizing the Dynamics of Information. CSLI Publications. Stanford, CA. 189–205.
86 Presuppositions and Informativity with Aspectual still Schlenker, P. (2006), Maximize Presupposition and Gricean Reasoning. Unpublished Manuscript, UCLA & Institut Jean-Nicod. Stalnaker, R. (1978), ‘Assertion’. Syntax and Semantics 9:315–22. Swart, H. D. & Verkuyl, H. J. (1999), Tense and Aspect in Sentence and Discourse. Utrecht Institute of Linguistics OTS. Utrecht. Szabolsci, A. (2004), ‘Positive polarity– negative polarity’. Natural Language & Linguistic Theory 22:409–52. van der Auwera, J. (1993), ‘Already and still: beyond duality’. Linguistics and Philosophy 16:613–53. van der Sandt, R. (1992), ‘Presupposition projection as anaphora resolution’. Journal of Semantics 9: 333–77. van der Sandt, R. & Huitink, J. (2003), ‘Again’. In P. Dekker and R. van Rooij (eds.), Proceedings of the 14th Amsterdam Colloquium. Amsterdam. ILLC. 181–186. van der Wouden, T. (1994), Negative Contexts. Groningen Dissertations in Linguistics: University of Groningen. von Fintel, K. & Iatridou, S. (2002). Since Since. Unpublished Manuscript, MIT. von Stechow, A. (2002), ‘Temporal prepositional phrases with quantifiers: some additions to Pratt and Francez (2001)’. Linguistics and Philosophy 25:755–800. Zeevat, H. (2004), ‘Particles: presupposition triggers, context markers or speech act markers’. In R. Blutner and H. Zeevat (eds.), Pragmatics and Optimality Theory. Palgrave Macmillan, Basingstoke & New York. 91– 112. First version received: 20.01.2008 Second version received: 15.04.2008 Accepted: 21.09.2008
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Musan, R. (1997), ‘Tense, predicates, and lifetime effects’. Natural Language Semantics 5:271–301. Nelken, R. & Francez, N. (1995), ‘Splitting the reference time: temporal anaphora and quantification in DRT’. ECL 194:261–6. Ogihara, T. (1990), The Semantics of the Progressive and the Perfect in English. DYANA deliverable R2.3.13. ESPRIT Basic Research Action BR3175, University of Edinburgh, Centre for Cognitive Science. Ogihara, T. (1994), ‘Events and states in discourse’. Paper presented at the Annual LSA Meeting, Boston, MA. Pancheva, R. & von Stechow, A. (2004), ‘On the present perfect puzzle’. In K. Moulton and M. Wolf (eds.), Proceedings of NELS 34. University of Massachusetts, Amherst. Partee, B. H. (1973), ‘Some structural analogies between tenses and pronouns’. The Journal of Philosophy 70:601–9. Partee, B. H. (1984), ‘Nominal and temporal anaphora’. Linguistics and Philosophy 7:243–86. Percus, O. (2006), ‘Anti-presuppositions’. In A. Ueyama (ed.), Theoretical and Empirical Studies of Reference and Anaphora: Toward the Establishment of Generative Grammar as an Empirical Science. Report of the Grant-in-Aid for Scientific Research (B), Project No. 15320052. Japan Society for the Promotion of Science. Fukuoka. 52–73. Prior, A. (1967), Past, Present and Future. Oxford University Press. Oxford. Reichenbach, H. (1947), Elements of Symbolic Logic. Macmillan. London. Rothstein, S. (2004), Structuring Events: A Study in the Semantics of Lexical Aspect. Blackwell. Oxford. Sauerland, U. (2007), ‘Implicated presuppositions’. In A. Steube (ed.), Sentence and Context. Mouton de Gruyter, Berlin, Germany.
Journal of Semantics 26: 87–107 doi:10.1093/jos/ffn012
Two nows in Korean EUNHEE LEE AND JEONGMI CHOI University at Buffalo, The State University of New York
Abstract
1 INTRODUCTION This paper is a study of the two lexical adverbs, cikum and icey meaning ‘now’ in Korean narrative discourse in dynamic semantics. The adverb now is often assumed to be an indexical expression referring to the time of utterance or, more accurately, to a time interval including the time of utterance. This assumption does not seem problematic when it comes to a canonical speech situation, where now refers to the moment of utterance for both speaker and hearer. However, unlike the canonical situation of utterance, now also appears in narrative discourse with the past tense, in which now denotes neither the speech time nor the production (or perception) time of the narrative. That is, now has its own independent status in the story world (Rauh 1983; Segal 1990) and is not referenced against the reader’s own real-time experience. In this paper, we discuss how these two nows in Korean, cikum and icey, Ó The Author 2009. 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
This paper presents a study of the two lexical adverbs in Korean, cikum and icey, which are assumed to be synonymous with each other and equivalent in meaning with English now. Because cikum and icey seem to be interchangeable in many instances without significant differences, their distinctive semantic features have been overlooked and not systematically studied. Starting from an overview of previous studies of cikum and icey, which focus on the intra-sentential analysis, this paper claims that, contrary to common assumptions, cikum and icey differ inherently in terms of the viewpoint of perspective taking in narrative discourse. Using examples drawn from a corpus, we argue that cikum simply refers to a time interval that contains the reference point from which the described event is viewed. On the other hand, icey describes a change in situation, showing that the reference point can be perceived as a point that divides the past and the future seen from this vantage point. Subsequently, we show that English now has in fact two functions corresponding to the Korean cikum and icey. The semantic differences between cikum and icey in narrative discourses are represented in discourse representation theory. Cikum preserves the given reference time, elaborating on an event described by the preceding sentence, while icey introduces a new reference time, updating the temporal context with a shifted temporal perspective.
88 Two nows in Korean
(1) a. John is eating breakfast now. b. Finally, at six o’clock, Mr. Johnson started to speak. Now there were only ten people left in the room. The reference points are established by context; often the relevant contextual factors are contained in the antecedent discourse. In the case of (1b), now in the second sentence must be understood as lying in the event time of the previous sentence which is marked as reference point. That is, the perspective shifts to the reference time of the last sentence in the simple past, and the situation is viewed from that temporal location, from the position of someone who could have been attending the meeting (Kamp and Rohrer 1983). In other words, in (1a), the tense is used deictically, that is it refers to the utterance time, which is salient in the speech situation similar to the use of deictic and demonstrative pronouns such as I, you and this, while in (1b), the tense is used anaphorically, that is it refers to a time which is extracted from the previous discourse similar to the use of pronouns and anaphors (see Partee 1983). Another property that has been pointed out about now is that it is only used with a stative predicate when occurring with the past tense in narratives. Thus, (2a) is awkward and not nearly as good as (2b) (Kamp and Reyle 1993: 596). It is observed that the same restriction applies to nows in other languages, for example maintenant in French (Kamp and Rohrer 1983: 265). (2) a. Bill had come home at seven. Now he wrote a letter. b. Bill had come home at seven. Now he was writing a letter. The Korean nows can be viewed in a similar way except that Korean has two lexical adverbs, cikum and icey, which are assumed to be synonymous and equivalent in meaning to English now. For example, in the major Korean dictionaries, cikum and icey are each, in most cases,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
can occur with the past tense in narrative discourse and what are the semantic differences between them. At first sight, it seems a little surprising that the adverb now can occur in the narrative, because we commonly assume that now always refers to utterance time. In most narratives, however, the events of the story are described in the past tense. In order to give a unified semantics for its indexical (deictic) and narrative uses, it is assumed that the adverb now in English, as well as in many other languages, refers to a time interval that overlaps the time of perspective from which the described event is viewed (Kamp and Reyle 1993: 595). The temporal perspective point can be either the speech time, as in (1a), or a past reference time, as in (1b).
Eunhee Lee and Jeongmi Choi 89
(3) Mina-ka cikum/icey hakkyo siktang-eyse pap-ul Mina-NOM now school cafeteria-LOC meal-ACC mek-nun-ta. eat-PRES-DEC1 ‘Mina is eating at the school cafeteria now.’ As we will see shortly, when embedded in a verbal context that provides a reference time, cikum and icey are rather clearly distinguished in terms of whether an expression introduces a new reference time, describing a later situation, or preserves the given reference time, elaborating on the previous situation. Sentential analyses treat reference times as completely undetermined, or as simply given by extralinguistic context. Therefore, the semantic difference between cikum and icey, which depends on a contextually given reference time, does not surface in isolated sentences like (3) above. However, it seems clear that the manner in which the reference time is extracted from the linguistic context may depend on the tenses of the sentence and its preceding sentences. In a context, various temporal relations between events described by the sentences can be observed, and based on this observation, numerous temporal inferences can be made about what happened when. In narrative discourse, as we will see shortly, cikum, like now in many other languages, can only occur with a stative predicate, while icey freely appears with an eventive predicate. Some questions naturally arise: why is it that Korean has two words for now, whereas English has 1
The abbreviations that we will be using in our glosses are the following: TOP: topic, NOM: nominative, ACC: accusative, LOC: locative, POSS: possessive, DR: directional, NEG: negation, PST: past, PRES: present, PRG: progressive, RESL: resultative, CN: connective, CONJ: conjunction, RL: relativizer, DEC: declarative, Q: interrogative, QT: quotative, SUP: suppositive, POL: polite speech level, AH: addressee honorific, SH: subject honorific, IN: indicative mood, CL: numeral classifier, PL: plural.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
defined in terms of each other. According to the Yonsei Korean dictionary, published by the Korean Language Society, cikum and icey are synonyms and both defined as ‘the very moment that one speaks’. Furthermore, as shown in the following example (3), it is very difficult, if not impossible, to exactly pinpoint the differences in meaning between the two in isolated sentences without any context. Regardless of whether cikum or icey is used, (3) describes an action of eating that is ongoing at the utterance time. That is, both cikum and icey can be used to denote an interval that contains the time of utterance. Because cikum and icey seem to be interchangeable in many instances, as in the example below, their own distinctive semantic features have been overlooked and not systematically studied or described.
90 Two nows in Korean
2 REVIEW OF PREVIOUS STUDIES AND THEIR PROBLEMS There are only a few previous studies of the Korean cikum and icey, and most of them are concerned with the properties of cikum as a temporal indexical. For example, Lee (1976) argues that cikum refers to an interval of time containing the present moment. Park (1994) examines the properties of cikum and kuttay, which correspond to now and then in English, and argues that, unlike now in English, cikum refers to a time interval which includes the past and the future.2 There are very few studies of icey, let alone comparative studies of cikum and icey. The first scholar who made a brief comparative study of cikum and icey is Park (2003). Park analyses sentences with cikum and icey, replacing one with the other and trying to find similarities and differences between them. Consider his examples: (4) Ku-nun cikum/icey ttena-n-ta. he-TOP now leave-PRES-DEC ‘He is now leaving.’ 2
This claim is based on the fact that cikum can sometimes occur with the future and past tenses. However, Park’s claim that cikum and now are semantically different is unconvincing given that English now can occur with the futurate progressive and the present perfect, just like cikum.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
only one? Why does icey behave differently than the now word in many other languages? What does the study of cikum and icey tell us about the property of now as a temporal (indexical and anaphoric) adverbial in natural languages? In this paper, we will show that cikum and icey differ inherently from the viewpoint of perspective taking in discourse and provide a systematic semantic account based on narrative discourse data using the framework of discourse representation theory (DRT; Kamp and Reyle 1993; van Eijck and Kamp 1997). Based on our analysis, we will argue that English now, and possibly now in other languages, in fact, has two functions corresponding to the two lexical nows in Korean. This paper is organized as follows. Section 2 reviews previous studies and their problems. Section 3 presents an analysis of cikum and icey in narrative discourse, describing various lexical and grammatical aspects they can combine with when the sentence is in the past tense. We also discuss English now and argue that its function is similar to the Korean icey. In section 4, the semantic differences between cikum and icey are represented in DRT. Section 5 concludes the paper by summarizing the main points and suggesting directions for future research.
Eunhee Lee and Jeongmi Choi 91
(6) Mina-ka cha-sako-lul tangha-yss-ta. Mina-NOM car-accident-ACC suffer-PST-DEC ‘Mina has had a car accident.’ (7) Mina-ka ithul-cen-pwuthe aph-ass-ta. Mina-NOM two.day-before-from sick-PST-DEC ‘Mina has been sick for two days now.’ Therefore, the implicature arising in (5) is not caused by icey. Even without icey or cikum, (5) has the implicature that he is not here, as illustrated in (8). (8) Ku-nun ttena-ss-ta. he-TOP leave-PST-DEC ‘He has left.’ Contrary to Park’s observation, cikum with the past tense also implicates the current relevance of the described situations, which is clearly shown in example (9). In (9), what the speaker B means by cikum is not the exact departure time of the train, but the fact that the described event has just ended, with this instantaneous event resulting in a change of state, that is the train is not in the station anymore.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(5) Ku-nun cikum/icey ttena-ss-ta. he-TOP now leave-PST-DEC ‘He has just left’ According to Park, in (4) the sentence with icey should be paraphrased as ‘he has been here all along and now he is leaving’ or as ‘he has been here and is supposed to leave soon’, whereas what the sentence with cikum indicates is that ‘the time to leave is just now, not another time’ or ‘what he is trying to do now is leaving not staying’. However, (4) with cikum also entails that he has been here. Let us consider the same example in the past tense. He argues that in (5), icey implies that he is not here now, along with the fact that he has left, but cikum only indicates that the time he left is just now and it cannot be seen as implying his not being here. However, the Korean past tense pragmatically implies that the result state of an event still holds at the utterance time (Lee 1987). Without any overt time adverbials, sentences with the past suffix -ess implicate current relevance of the described situation. For example, (6) indicates that the accident has occurred recently and that possibly Mina is in the hospital now. In English, this is often expressed with the present perfect. Also, sentences with durative adverbials show similar effects. Example (7) indicates that Mina is still sick.
92 Two nows in Korean (9) A: Sillyey-ha-p-ni-ta. Yeset-si Seoul-hayng impoliteness-do-AH-IN-DC six-o’clock Seoul-for kicha-lul ta-lyeko ha-nuntey-yo. train-ACC take-intend.to do-CN-POL ‘Excuse me, I’m leaving on the 6:00 for Seoul.’ B: Seoul-hayng kicha-nun cikum ttena-ss-supnita. Seoul-for train-TOP now leave-PST-DEC ‘The train for Seoul has left now.’
3 AN ANALYSIS OF CIKUM AND ICEY IN NARRATIVE DISCOURSE We examined two hundred actual discourse examples of cikum and icey from the Seyjong written corpus, which is an abridged version of the Seyjong Corpus jointly published by the National Korean Language Institute and the Ministry of Culture and Tourism in Korea. The compilation is based on ten different genres and topics, including novels, science, humanities (e.g. anthropology and philosophy), newspapers, art and life, essays, education and society. In the majority of cases in which cikum and icey are used, they refer to the speech time (82% for cikum and 71.5% for icey). In the remaining cases, they occur with the past tense, changing the temporal
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
None of the previous analyses include naturally occurring discourse examples and they only deal with constructed sentences isolated from the discourse context. However, on the basis of isolated sentences, it is not easy to show their differences, because in isolation most sentences sound perfectly plausible regardless of whether cikum or icey is used. Furthermore, previous analyses cannot deal with the use of cikum in past tense clause in narratives; they only account for the indexical use of cikum, referring either to a present ongoing event or to a present resultant state (when used with the past tense in isolated sentences). Their main interest lies in explaining cikum’s dependency either on the tense of the sentence or the type of verb that is used, mainly focusing on an intra-sentential analysis. Furthermore, although some scholars have tried to find differences between the Korean cikum (and icey) and English now, no previous analysis raised questions such as ‘Why are there two words for now in Korean? What are their distinct characteristics, and how are they similar or different from now in other languages?’ In the following section, a discourse-based account is presented, which can provide answers to these questions.
Eunhee Lee and Jeongmi Choi 93
Cikum Telic verbs Stative predicates Modal expressions Negative (-ci ahn) Result state (-a iss) Progressive (-ko iss) Inchoative (-toy, -ci) Total Table 1
2 10 5 6 2 11 0 36
5.5% 27.8% 13.9% 16.7% 5.5% 30.6% 0% 100%
Icey 17 4 0 10 0 2 24 57
29.8% 7% 0% 17.5% 0% 3.5% 42.2% 100%
Various lexical and grammatical aspects co-occurring with cikum and icey
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
perspective, in a way similar to the English example (1b). In this paper, we focus on how cikum and icey combine with the past tense, since our main interest lies in their ability to shift the temporal perspective, and the semantic difference between the two becomes clearly visible only when they are embedded in discourse contexts. In other words, the current article only provides a detailed examination and analysis of the adverbs combining with the past in narratives, and the result of the investigation will have to be extended to cover the (majority) present tense cases. Among the two hundred discourse examples we looked at, 36 examples of cikum and 57 examples of icey occur with the past tense (18% for cikum and 28.5% for icey). Table 1 shows various lexical and grammatical aspects with which cikum and icey can combine in the past tense. The most striking aspects of these data are the following: the large percentage of cases where cikum is used with (i) the progressive -ko iss ‘be -ing’ and (ii) stative predicates, and also the large percentage of cases where icey is used with (iii) the inchoative -toy or -ci ‘get, become’ as well as (iv) telic verbs. In the remaining cases, cikum occurs with modal expressions and the result state form and both cikum and icey appear in negative sentences. Based on this observation, we claim that cikum occurs with state descriptions whereas icey combines with event descriptions; (10) below is an example in which cikum occurs with modal expression and negation. Since modals and result states are state descriptions, the fact that icey does not occur with them also supports the claim that icey only occurs with dynamic event descriptions.
94 Two nows in Korean (10) Manswu-nun pang-ulo tuleka-ass-supnita. Mansu-TOP room-LOC enter-PST-DEC ‘Mansu entered the room.’ cikum i kipwun-ulo-nun amwukes-to mos now this feeling-with-TOP nothing-even NEG ha-l kes kath-ass-supnita. do-seems-PST-DEC ‘It seemed that he couldn’t do anything now feeling like this.’
(11) Swuhye-nun amwu soli-to tuli-ci anh-ass-ko Swuhye-TOP no sound-also be heard-NEG-PST-and casin-i cikum ttuy-ko iss-ta-nun nukkim-cocha self-NON now run-PRG-DEC-RC feeling-even tul-ci ahn-ass-ta. enter-NEG-PST-DEC ‘Swuhye could not hear any sound and could not even feel that she was now running.’ (12) Taylicem-un ilpon ponsa-lo palsonghay-ss-ko agency-TOP Japan head.office-LOC dispatch-PST-and ponsa-ese -nun cikum memwulu-ko iss-nun head.office-from -TOP now stay-PRG-RC
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Although both adverbs involve a shift of temporal perspective, the properties of cikum and icey, as revealed by these facts, are quite different. We will argue that, in a sentence where cikum is used, the temporal perspective point is the very time at which events are being viewed or narrated. By contrast, the temporal perspective point of icey seems to be the point that divides two opposite states. In the following we will back up this claim. Let us examine the uses of cikum first. For cikum, the shifted reference point is perceived simply as a temporal vantage point from which the described event is viewed. As a result, like English now, cikum almost invariably takes an internal viewpoint, that is it describes something as going on at the reference time. As shown in Table 1, the majority of cases in which cikum is used are stative descriptions and there were only two cases where cikum was used with a telic eventive verb. First, let us look at the cases where cikum is used with the progressive -ko iss ‘be ing’. In (11) and (12), the events are viewed from a past perspective point as ongoing, adding presentness and vividness to the situation.
Eunhee Lee and Jeongmi Choi 95
ilakh-e ponay-o-ass-ta. Iraq-LOC send-come-PST-DEC ‘The agency dispatched it to the headquarters, and the headquarters forwarded it to the local branch in Iraq I was staying at now.’
(13) Nay kiek-ulo-nun han cip-to ppattuli-n my memory-in-TOP one house-even miss-RL kes kath-ci ahn-untey. . . Kulena cikum/*icey seem like-NEG-CN but now nay son aney-nun pwunmyeng sinmwun-i han cang my hand in-TOP certainly newspaper-NOM one nam-a iss-nun kes-i-ess-ta. remain-RESL-RL thing-be-PST-DEC ‘As far as I remembered, it seemed I had not missed a single house. However, without a doubt, now I had one newspaper left in my hand.’ In (13), cikum indicates a current state of the narrator’s (a newspaper delivery person) having one newspaper left in his hand. Furthermore, the fact that the narrator has a newspaper that has not been delivered overlaps the time when he tried to remember if he missed any houses. If cikum is replaced with icey, which requires a change of state, the discourse sounds unnatural. Let us now turn to icey. Unlike cikum, icey is used with inchoatives or resultatives most often, implying that there must be a qualitative change of state, as shown in (14). 3
In order to observe a context change, the first sentence must describe a telic event. If the first sentence gives an atelic description, the following sentence does not generally describe a later event, but would express an overlapping one.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
According to Kamp and Reyle (1993), Kamp and Rohrer (1983), Hinrichs (1986) and Partee (1983), stative predicates preserve the current temporal reference and give a more detailed description of the event described in the previous sentence, whereas an event sentence introduces a new reference time, updating the context with a further episode. Since the stative predicate overlaps an event described in the previous sentence, a sentence in which cikum is used expresses an overlapping temporal relation with the event described in the preceding sentence.3 Consider (13), in which cikum modifies a sentence in the result state form.
96 Two nows in Korean (14) Minswu-nun say os-ulo kalaip-ess-supnita. Minswu-TOP new clothes-into change-PST-DEC Ku-nun icey/*cikum wancenhi mosup-i he-TOP now completely appearance-NOM pakkuy-ess-supnita. be changed-PST-Dec ‘Minswu changed into new clothes. Now his appearance had changed completely.’
(15) Talkyal hana-ka elmana pissata-ko, tayswulopta-ko, egg one-NOM how expensive-Q important-Q kuluhkey sai coh-ten iwus sachonkkili like that relation good-used to neighbour cousin meli-ka theci-tolok ssawutani. . . icey/?*cikum head-NOM break-until fight now twu cip-un kuman sai-ka two house-TOP unfortunately relation-NOM nappa-ci-ess-supnita. bad-become-PST-DEC ‘How expensive is an egg? How important is it? How could the two neighbours who used to have such a good relationship fight so furiously over an egg? Now the two neighbours have become estranged.’ In (15), the final sentence with icey describes an event of two neighbours’ becoming estranged, which follows the event of their fighting over an egg. Here, the event modified by icey is a dividing point between two opposite states: the state of getting along well and the state of having a bad relationship. The event of becoming estranged follows the event of fighting in the previous sentences, moving the narrative time forward. The existence of icey in Korean shows that the shifted temporal perspective can be perceived not only as a temporal vantage
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In (14), the protagonist Minswu changed into new clothes, which made him look like a different person. In order to indicate a change of state from his appearance before changing into new clothes to his resulting appearance, icey is used. Since cikum does not imply a change of state but refers to an ongoing state, it is deviant in this context. Let us observe another discourse example (15).
Eunhee Lee and Jeongmi Choi 97
point from which the described event is viewed but also as a point between two different states seen from the temporal vantage point. Another interesting result of our corpus research is that, when used with the past tense, icey appears in more diverse genres such as novels, newspaper articles, and diaries. In the corpus, we found that all the examples of cikum in past tense sentences were drawn from novels, while 20 examples of icey (35%) in past tense sentences came from non-fiction sources such as newspaper articles and diaries. Therefore, icey’s function is not restricted to changing the temporal perspective in a narrative. Example (16) contains an example of icey in a newspaper article.
In (16), icey refers to the change of situation from the building having a lot of foreign company offices to the current situation where more than half of them have left. Cikum does not normally occur with past tense in newspaper articles which deliver a sequence of events in a factual manner.4 Let us look at another example of icey, taken from a diary. (17) Icey an ip-nun-ta-ko nehetu-n kin now NEG wear-PRES-DEC-QT put-RL long os-i tasi nao-ko. . . clothes-NOM again come-CN ‘I took out the long clothes again, which I had put into the closet, thinking that I would never wear them now. And . . .’ In (17), the writer took her clothes out which had been originally shoved away in the closet because at that time she thought she would never wear them again. In order to indicate a change of state from the state of her wearing the clothes to the state of not wearing the same 4 In Kamp and Rohrer (1983: 265), it is noted that French maintenant ‘now’ would never be found ‘in a newspaper article which reports a sequence of events in a purely factual, detached manner. It is only when the author takes the temporal perspective of the past that sentences with maintenant, dans une heure, jusqu’ici show up’.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(16) Han ttay i kos-ey ipcwuhan oykwuk kiep one time this place-at move in foreign company swusca-ka ilpayk sipkay-lul nem-ess-una number-NOM hundred ten-ACC over-PST-but icey celpan isang-i ttena-ss-ko. . . now half more than-NOM leave-PST-CN ‘At one time, there were more than a hundred and ten foreign company offices in this building, but now more than half of them have left. . .’
98 Two nows in Korean
(18) Ku-nun cikum han yein-kwa salang-ey ppaci-ess-nuntey he-TOP now one woman-with love-in fall-PST-CN eccihal ci molukey-ss-ta-ko hay-ss-ta. what to do do not know-PST-DEC-Q do-PST-DEC ‘He said that he had now fallen in love with some woman, and he didn’t know what to do.’ In (18), the main verb which describes an event of him falling in love implies a current state of him being in love at the time of his statement, and cikum seems to modify this result state, rather than the past event of falling in love7; (19) is the other example. (19) Cikum myeth ttay-maney wusen hekicung-iya Now several time-within for now hunger-as for myenha-yss-ciman maksang eti-lo ka-ya ha-l ci-lul escape-PST-CN when it comes to where-LOC go-must-CN 5 In Korean, activities with the present tense denote both habitual and current events, as in (i). But with cikum, (i) can only mean ‘Mina is playing the piano’.
(i)
Mina-ka piano-lul chi-n-ta. Mina-NOM piano-ACC play-PRES-DEC ‘Mina plays the piano’ or ‘Mina is playing the piano’
6 The constraint is particularly clear in a language such as French, where there are two past tenses, the Passe´ Simple and the Imparfait. The word maintenant ‘now’ goes with Imparfait but not with Passe´ Simple’ (Kamp and Reyle 1993: 595). 7 See the discussion under example (8) above for the implicature arising from the use of a past tense. The Korean past tense implicates the current relevance of the described situation.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
clothes, icey is used. Since cikum coerces an ongoing process interpretation,5 the sentence does not make sense if icey is replaced with cikum. As mentioned above, in the corpus that we used, there were only two examples in which cikum was used with telic eventive sentences. These two examples seem like counterexamples to the claim that cikum simply refers to a shifted temporal perspective point from which the event is viewed as ongoing; we have already seen that cikum is naturally used with the progressive -ko iss and stative predicates most often in past tense sentences. For example, it has been argued that in English and French6 it is almost impossible for now or maintenant to refer to a past time if the clause in which it occurs describes an event rather than a state (Kamp and Reyle 1993: 595; Kamp and Rohrer, 1983: 265). Then, how can we explain the exceptional cases in which cikum occurs with an event predicate? Consider (18).
Eunhee Lee and Jeongmi Choi 99
sayngkakha-myen chamulo atulhaki-man hay-ss-ta. think-when indeed vague-only do-PST-DEC ‘I now barely escaped from hunger after skipping several meals, but when I thought about where to go, I only had the vaguest ideas.’
(20) Kulena nay-ka nolay-lul pwulu-nun swunkan-mada but I-NOM song-ACC sing-PRS moment-each kulen kes-ey tayhay sayngkakha-ko iss-ci nun anh-keyssci. such thing-about think-PRG-NEG-PRS-SUP Na-nun icey haypalaki 2 cip-uy ‘‘maum I-TOP now sunflower second album-POSS heart kiph-un kot-ey kutaylo’’-lul pulu-ko iss-ess-ta. deep-RL place as it is-ACC sing-PRG-PST-DEC ‘However, I did not think I would ponder upon such things every moment I sing a song. Now I was singing a song, titled ‘From the bottom of my heart’ from the second album of the Sunflower.’ If we examine the interpretation more closely, we soon find that (20) is not necessarily a counterexample to the claim that icey is not likely to be used with the progressive since it indicates a change of state. In the second sentence, the narrator suddenly notices that he has already finished a song he started humming a while ago and now he is humming a different song to himself. The progressive seems to be employed to have an effect of ‘the next thing I remember, I was singing. . .’. Note that the first sentence describes the narrator’s inner thoughts about something else. Dowty (1986: 56) discusses a similar phenomenon in English progressives. According to him, the cases where the progressive sentence refers to a later situation than its preceding sentence indicate a very particular psychological effect of the protagonist of the narrative: an event begins to happen, but it is only when it is already in progress that the protagonist realizes what is going on. Icey’s tendency not to blend
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Similarly, in (19), the sentence describing an event of the narrator escaping hunger implies her current state of being fed for now. Note that there were only two exceptional cases and, in both cases, the verb meaning implicates a current state. More data are needed to see when exactly cikum can occur with event descriptions. In the corpus that we used, we have found two examples where icey occurs with the progressive -ko iss. If icey indicates a change of state, it is not expected to co-occur with the progressive, which is always stative. Consider (20).
100 Two nows in Korean well with the progressive is also observed in the dialogue (21) below. In an isolated sentence, both the progressive with cikum and icey sound plausible, but when relevant and concrete contexts are provided, as in a narrative discourse, the difference between cikum and icey becomes visible. In (21B), icey can be used with the progressive only when it indicates a change of state in the given context; that is when speaker B implies a change of state from Mina’s doing something other than playing the piano to the current state of her playing the piano. Since speaker A asks what Mina has been doing all his time, rather than a new activity that she has just started, icey is awkward.
In our corpus, we also found four instances where icey is used with stative predicates; (22) contains an example. (22) Icey i haknyen-i-n minswu-nun ssulssulhaki-man now second grade-be-NM Minsu-TOP feel lonely-only hay-ss-supnita. do-PST-DEC ‘Minsu, who was now in second grade, was very lonely.’ Although icey is used with the copular be ‘be in the second grade’, (22) still implies a change of state, namely, going up to the second grade. Therefore, we can freely replace the phrase with i haknyen-i toy-n ‘who has become second grade’. The other three cases were similar. Based on these observations, we claim that cikum has the same meaning as now in English and other languages in that it signals that the reference time (Rpt) overlaps the described eventuality. In other words, the eventuality modified by cikum simply refers to an open time interval, whether long or short, which contains the time of perspective. Due to this overlapping relation between the Rpt and the described eventuality, cikum occurs with state descriptions. On the other hand, icey is compatible with inchoative and resultative event descriptions. The event has an open time span with only one end of its boundaries
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(21) A: Mina-nun totaychay pang-eyse yethay Mina-TOP on earth room-LOC so far mwe ha-ko iss-ni? what do-PRG-Q ‘What on earth has Mina been doing in her room all this time?’ B: Cikum/*icey piano chi-ko iss-eyo. now piano play-PRG-DEC ‘She is playing the piano.’
Eunhee Lee and Jeongmi Choi 101
(24) Nor did the boy address him or Peter. He stood a little apart from them, kicking at the screen, and then he walked off across Goughdale between the crumbling towers. Stephen could remember how hot it had been, the sky a dazzling white-blue, the heat making the air wave and shiver above the dry yellowed turf. Dusk now brought a stillness and its own grey translucent light. (25) Complexity was neither necessary nor desirable; it was of no practical use and it solved nothing in the end. Someone touched his elbow so timidly that he thought it had been accidental, until the gesture was repeated with more insistence. Now he turned and saw Nebamun walking beside him. Based on these empirical facts, we claim that English now in fact has two functions corresponding to Korean cikum and icey. In other words, 8 We have observed that, in a majority of cases, cikum and icey are used with the present tense, referring to the utterance time. Given the semantic differences between the two that we have found through the examination of narratives, we speculate that cikum simply refers to the utterance time when it is used with the present tense, while icey indicates that the utterance time is a dividing point between the past and the future, implying a change of state. This speculation needs to be justified by the examination of the present tense cases, which is outside the scope of the current paper and is left to future research.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
fixed, either the starting point or the resulting point. Icey shows that the Rpt can be perceived as a dividing point between two different states seen from the temporal vantage point.8 So far, we have assumed that now only occurs with stative predicates, as argued by Kamp and Reyle and Kamp and Rohrer, and therefore the Korean icey is an idiosyncratic feature in this language. Given this assumption, it is puzzling why Korean has two words for now while other languages have only one. However, contrary to the claim of Kamp and Reyle and Kamp and Rohrer, we found examples where now is used with telic verbs in the past tense in the British National Corpus. In the examples below, it is clear that the sentence modified by now indicates a change of state, moving the reference time forward. This is exactly what icey does in narratives. (23) The early life of their son, Henry II, was marked by the wrangle between Stephen and Matilda for the sovereignty of England. His succession to the English crown in 1154 at the age of 21 was hailed with relief by those worn down by insurrection and counter-bargaining. Anarchy was over and there now began the reign of one of the strongest kings ever to rule England, one whose territories and titles gave him authority over a great part of north-western Europe.
102 Two nows in Korean even English now performs the function of shifting to a new stage in narratives, which is accomplished by icey in Korean. This is expected if we assume that the reference point (Rpt) can be a temporal vantage point from which the present situation is viewed (imperfective perspective) or as a point which divides the past and the future seen from this vantage point (perfective perspective). In Korean, perspective taking is lexicalized, while in English it perhaps derives from the rhetorical structure of the narrative. 4 SEMANTIC REPRESENTATION
(26) John entered the bar. He was wearing a black jacket. He sat down. It is one of the principles governing past tense narrative discourse that the eventuality described by a non-initial sentence is interpreted as standing in some specific relation to some other event e# introduced by an earlier sentence. The temporal relation between e and e# becomes especially important when the new sentence does not contain a temporal adverbial, so that the new eventuality e can be located only in relation to the antecedent context. How e and e# are related depends on a number of factors. One of these is whether e# is an event or a state. If e# is a state, the relation is almost invariably that of overlap. This is the case for the second sentence of (26), for sentences in the progressive always describe states. If e# is an event, then it is typically understood as following the event e. This case is illustrated by the last sentence of (26). When the last sentence gets to be interpreted, the event of John entering the bar is still the last mentioned event. And the new event, that of him sitting down, is naturally seen as following the event of his coming in. Kamp and Reyle adopt Reichenbach#s notion of reference time by introducing into the DRS a condition of the form Rpt ¼ a, where a is some discourse referent which represents a time or an event and which is already present in the DRS. Example (27) below is the DRS for (26).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In this section, the semantic differences between cikum and icey are represented in DRT (Kamp and Reyle 1993; van Eijck and Kamp 1997). In DRT, it is the discourse representation structure (DRS), not the individual sentence as such, which is semantically interpreted. Hence, the theory is useful for analysing narrative discourses and the anaphoric relations among sentences occurring in them. A DRS for tensed discourse includes events, states and times as objects in the universe of discourse, and it specifies relations of precedence and inclusion among them. Observe how DRT analyses a sequence of sentences as in (26).
Eunhee Lee and Jeongmi Choi 103
(27)
(n is the utterance time, x < y means x temporally precedes y, x 4 y means x is temporally included in y, and O means temporal overlap). The first sentence of (26) introduces an event discourse referent e of John entering the bar, which is included in its location time t. The location time precedes the utterance time. The past tense of the second sentence contributes, just as that of the first sentence, the condition that the described eventuality precedes n. The first step needed to process the second sentence of (26) involves adding the condition Rpt ¼ e, meaning that e is to act as reference point for the second sentence. Since the eventuality denoted by the second sentence is a state, the processing principle entails that it includes the reference point. The interpretation of the third sentence will again require the choice of a reference time. In other words, the value of Rpt will have to be reset. In the present case, e remains the reference point, so Rpt is reset to the same value. To allow for the resetting of Rpt, when a sentence whose interpretation required the choice of an Rpt has been fully processed, the equation expressing this choice is eliminated from the resulting DRS. The third sentence describes an event and so the relation between this event and the reference point is succession. The DRS thus
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
e t n x y s t# u w e# t$ v e4t t
104 Two nows in Korean constructed is true iff there is an embedding function that verifies the DRS in the given model. Let us see first how DRT analyses the sentence that contains cikum and a past tense. Example (29) is the DRS for example (13), which is repeated as (28).
(29) n e t s t# x y z e4t t
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(28) Nay kiek-ulo-nun han cip-to ppattuli-n my memory-in-TOP one house-even miss-RL kes kath-ci ahn-untey. . . Kulena cikum/*icey seem like-NEG-CN but now nay son aney-nun pwunmyeng sinmwun-i han cang my hand in-TOP certainly newspaper-NOM one nam-a iss-nun kes-i-ess-ta. remain-RESL-RL thing-be-PST-DEC ‘As far as I remembered, it seemed I had not missed a single house. However, without a doubt, now I had one newspaper left in my hand.’
Eunhee Lee and Jeongmi Choi 105
person had one newspaper left while he remembered that he had not missed any houses. A DRS construction rule for cikum is provided in (30). Cikum with the past tense introduces into the universe of DRS a new state discourse referent s, referring to the state description modified by cikum. It also contributes a DRS condition such that the state includes the given reference time, which is introduced by its preceding sentence. (30) Triggering configuration: cikum VP(s)-ess Choose Rpt: Rpt ¼ e Introduce into UK: a new state discourse referent s Introduce into ConK: e 4 s
(31)
Iwus sachonkkili meli-ka theci-tolok ssawutani.. neighbour cousin head-NOM break-until fight icey twu cip-un kuman sai-ka now two house-TOP unfortunately relation-NOM nappa-ci-ess-supnita. bad-become-PST-DEC ‘How could the two neighbours fight so furiously? Now the two neighbours became estranged.’
(32) n e t e# t# x e4t t
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us now see how DRT analyses examples that contain icey. The DRS for example (15) is (13), which is repeated as (31). Here, e# is the beginning point of a result state of neighbours’ being estranged.
106 Two nows in Korean (33). The event modified by icey introduces its own event discourse referent, which follows the given Rpt. (33) DRS construction rule for icey Triggering configuration: icey VP(e)-ess Choose Rpt: Rpt ¼ e Introduce into UK: a new event discourse referent e# Introduce into ConK: e < e#
5 CONCLUSION In this paper, we have argued for a discourse-based account of the two Korean nows, which are cikum and icey, showed their semantic differences in narrative discourse and modelled this difference in DRT. Based on the corpus data, we have argued that, in a sentence where cikum is used, the shifted reference point is perceived simply as a temporal vantage point from which the described event is viewed or narrated. As a result, cikum almost invariably takes an internal viewpoint, that is it describes something as going on at the time of description. On the other hand, icey is compatible with inchoative and resultative event descriptions. Therefore, we argued that the event icey modifies is the dividing point between two different states. In DRT, when now does not introduce its own reference time, but is anchored to the previous reference time, it does not forward the narrative time, but provides elaboration. This is the function of cikum. When now introduces its own
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Note that the construction rules (30) and (33) represent a way to construct the correct DRSs in past tense cases. This is not a final analysis of cikum and icey since it does not extend to the present tense cases yet. We cannot provide such a general theory which covers all cases in this paper since we have not studied the present tense cases in detail. Furthermore, the construction rules seem to make the difference between the two adverbs the result of a lexical choice. There is an alternative way of looking at the difference: cikum only combines with states, and that the overlap relation which is associated with it follows from the general principle that state sentences overlap with the current reference time. On the other hand, icey only combines with events, and the sequential interpretation follows from the fact that event descriptions move the narrative time forward. Our rules (30) and (33) put into a single rule both the lexical meanings of the adverbs and the effects of general principles of narration involving aspectual class. In the future, as we build a theory which works for both past tense narration and present tense uses, we would need to separate out these two contributions.
Eunhee Lee and Jeongmi Choi 107
new reference time, shifting the temporal perspective, it indicates a change of state and moves the narrative time forward. This is the function of icey. In English, the same word now has both functions. We observed that, contrary to the common assumption, English now in fact is used with punctual eventive verbs and serves a similar function as icey does. A more detailed study of how the two perspectives in discourse are derived in English and other languages is left for future research, as is the construction of a general theory that covers both past tense narratives and the present tense cases.
REFERENCES Dowty, D. 1986. The effect of aspectual class on the temporal structure of discourse: semantics or pragmatics? Linguistics and Philosophy 9:37–61. Hinrichs, E. 1986. ‘Temporal Anaphora in Discourse in English’. Linguistics and Philosophy 9:63–82. Kamp, H. & U. Reyle. 1993. From Discourse to Logic. Kluwer. Dordrecht. Kamp, H. & C. Rohrer. 1983. ‘Tense in Texts’. In S. Bauerle, A. von Stechow (eds.), Meaning, Use, and Interpretation of Language. De Gruyer. Bauerle. 250–269. Lee, C. 1987. ‘Temporal Expressions in Korean’. In J. Verschueren & M. Bertuccelli-Papi (eds.), The Pragmatic Perspective. John Benjamins. Amsterdam. 405–447. Lee, K. 1976. ‘The Meaning of Cikum. Ehakyenkuw 12-2’. In Y.-K. Ko & K.-Sh. Nam (eds.), Korean Syntax and Semantics. 1982. Top Publishing Company. Seoul. 263–282. Park, C. 1994. ‘Temporal Indexicals in Korean: A DRT Approach’. Unpub-
lished MA thesis, Korea University, Seoul, Korea. Park, S. 2003. ‘Icey and Cikum’. Hangulhakhoey 260:183–197. Partee, B. 1983. ‘Nominal and Temporal Anaphora’. Linguistics and Philosophy 7:243–286. Rauh, G (ed.). 1983. Essays on Deixis. Narr, Tu¨bingen. Reichenbach, H. 1947. Elements of Symbolic Logic. Academic Press. New York. Segal, E. M. 1990. ‘Fictional Narrative Comprehension: Structuring the Deictic Center’. The Fifth Annual Conference of the Cognitive Science Society. 526–533. van Eijck, J. H. Kamp. 1997. Representing discourse in context. In J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language. Elsevier. Amsterdam. 179–237.
First version received: 26.09.2007 Second version received: 25.01.2008 Accepted: 16.05.2008
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
EUNHEE LEE AND JEONGMI CHOI Assistant professor Department of Linguistics 623 Baldy University of Buffalo, The State University of New York e-mail:
[email protected]
