Presuppositions and Discourse: Essays Offered to Hans Kamp 1849507821, 9781849507820

Citation preview

Presuppositions and Discourse: Essays offered to Hans Kamp

Current Research in the Semantics/Pragmatics Interface

Series Editors: Klaus von Heusinger, University of Stuttgart, Germany Ken Turner, University of Brighton, UK Other titles in this series: De Brabanter & Kissine (eds.) Mosegaard Hansen Pietarinen (ed.) Aloni et al. (eds.) Von Heusinger & Turner (eds.) Bultinck Wedgwood Dobrovol’skij & Piirainen Peregrin (ed.) Kamp & Partee (eds.) Gutie´rrez-Rexach (ed.) Bras & Vieu (eds.) Ne´meth & Bibok (eds.) Leezenberg Papafragou Peeters (ed.) Jaszczolt (ed.) Geurts Jaszczolt Turner (ed.)

Volume 20, Utterance Interpretation and Cognitive Models Volume 19, Particles at the Semantics/Pragmatics Interface Volume 18, Game Theory and Linguistic Meaning Volume 17, Questions in Dynamic Semantics Volume 16, Where Semantics Meets Pragmatics Volume 15, Numerous Meanings: The Meaning of English Cardinals and the Legacy of Paul Grice Volume 14, Shifting the Focus: From Static Structures to the Dynamics of Interpretation Volume 13, Figurative Language: Cross-cultural and Cross-linguistic Perspectives Volume 12, Meaning: The Dynamic Turn Volume 11, Context-Dependence in the Analysis of Linguistic Meaning Volume 10, From Words to Discourse: Trends in Spanish Semantics and Pragmatics Volume 9, Semantic and Pragmatic Issues in Discourse and Dialogue Volume 8, Pragmatics and the Flexibility of Word Meaning Volume 7, Contexts of Metaphor Volume 6, Modality: Issues in the Semantics-Pragmatics Interface Volume 5, The Lexicon-Encyclopedia Interface Volume 4, The Pragmatics of Propositional Attitude Reports Volume 3, Presuppositions and Pronouns Volume 2, Discourse, Beliefs and Intentions: Semantic Defaults and Propositional Attitude Ascription Volume 1, The Semantics/Pragmatics Interface from Different Points of View

Presuppositions and Discourse: Essays offered to Hans Kamp EDITED BY Rainer Ba¨uerle Institut fu¨r Maschinelle Sprachverarbeitung Universita¨t Stuttgart Uwe Reyle Institut fu¨r Maschinelle Sprachverarbeitung Universita¨t Stuttgart Thomas Ede Zimmermann Institut fu¨r Kognitive Linguistik Johann Wolfgang Goethe-Universita¨t

United Kingdom – North America – Japan India – Malaysia – China

Emerald Group Publishing Limited Howard House, Wagon Lane, Bingley BD16 1WA, UK First edition 2010 Copyright r 2010 Emerald Group Publishing Limited Reprints and permission service Contact: [email protected] No part of this book may be reproduced, stored in a retrieval system, transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright Clearance Center. No responsibility is accepted for the accuracy of information contained in the text, illustrations or advertisements. The opinions expressed in these chapters are not necessarily those of the Editor or the publisher. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-1-84950-782-0 ISSN: 1472-7870 (Series)

Awarded in recognition of Emerald’s production department’s adherence to quality systems and processes when preparing scholarly journals for print

Current Research in the Semantics/Pragmatics Interface (CRISPI)

Series Editors: Klaus von Heusinger, University of Stuttgart, Germany and Ken Turner, University of Brighton, UK Editorial Advisory Board: N. Asher, USA B. Birner, USA C. Casadio, Italy M. Dascal, Israel B. Fraser, USA T. Fretheim, Norway B. Gillon, Canada P. Gochet, Belgium J. Groenendijk, The Netherlands Y. Gu, PRC L. Horn, USA Y. Huang, New Zealand A. Kasher, Israel M. Krifka, Germany S. Kubo, Japan C. Lee, Korea S. Levinson, The Netherlands T. McEnery, UK F. Nemo, France P. Pelyvas, Hungary J. Peregrin, Czech Republic A. Ramsay, UK R. Stalnaker, USA M. Stokhof, The Netherlands J. van der Auwera, Belgium R. van der Sandt, The Netherlands G. Ward, USA H. Zeevat, The Netherlands The aim of this series is to focus upon the relationship between semantic and pragmatic theories for a variety of natural language constructions. The boundary between semantics and pragmatics can be drawn in many various ways; the relative benefits of each gave rise to a vivid theoretical dispute in the literature in the last two decades. As a side effect, this variety has given rise to a certain amount of confusion and lack of purpose in the extant publications on the topic. This series provides a forum where the confusion within existing literature can be removed and the issues raised by different positions can be discussed with a renewed sense of purpose. The editors intend the contributions to this series to take further strides towards clarity and cautious consensus.

Contributors List Nicholas Asher E´quipe Lilac, Institut en Informatique de Toulouse (IRIT) Universite´ Paul Sabatier 118 Route de Narbonne 31062 Toulouse Cedex 9 France David Beaver Department of Linguistics The University of Texas at Austin Calhoun Hall 405 1 University Station B5100 Austin, TX 78712-0198 USA Manfred Bierwisch Zentrum fu¨r Allgemeine Sprachwissenschaft Schu¨tzenstr. 18 10117 Berlin Germany Robin Cooper Department of Philosophy, Linguistics and Theory of Science University Go¨teborg Box 200 SE 405 30 Go¨teborg Sweden Paul Dekker Faculteit der Geesteswetenschappen ILLC Universiteit van Amsterdam P.O. Box 94242 1090 GE Amsterdam The Netherlands vii

viii PRESUPPOSITIONS

AND

DISCOURSE: ESSAYS

OFFERED TO

HANS KAMP

Bart Geurts Department of Philosophy University of Nijmegen Postbox 9103 6500 HD Nijmegen The Netherlands Staffan Larsson Department of Philosophy, Linguistics and Theory of Science University Go¨teborg Box 200 SE 405 30 Go¨teborg Sweden Michael Morreau Department of Philosophy University of Maryland Skinner Building College Park, MD 20742 USA Barbara H. Partee Department of Linguistics University of Massachusetts, Amherst Amherst, MA 01003-9274 USA Robert van Rooij Faculteit der Geesteswetenschappen ILLC Universiteit van Amsterdam P.O. Box 94242 1090 GE Amsterdam The Netherlands Rob van der Sandt Department of Philosophy Radboud University Nijmegen Erasmusplein 1 6525 HT Nijmegen The Netherlands

Contents Contributors List

vii

Preface

xi

1

Presupposition: An (un)Common Attitude? ROBERT VAN ROOIJ

2

Pragmatic Strategies ROB VAN DER SANDT

3

Have You Noticed that Your Belly Button Lint Color is Related to the Color of Your Clothing? DAVID BEAVER

1 29

65

4

Accommodation and Reaccommodation in Dialogue ROBIN COOPER AND STAFFAN LARSSON

101

5

Specific Indefinites, Presupposition and Scope BART GEURTS

125

6

Deixis, Binding and Presupposition NICHOLAS ASHER

159

7

BECOME and its Presuppositions MANFRED BIERWISCH

189

8

Only If an Indefinite is Topical, the Pronoun Picks it Up PAUL DEKKER

235

9

Good and Bad Transitive Arguments MICHAEL MORREAU

261

10

Privative Adjectives: Subsective Plus Coercion BARBARA H. PARTEE

273

ix

Preface In hindsight, the debate about presupposition following Frege’s discovery that the referential function of names and definite descriptions depended on the fulfillment of an existence and a uniqueness condition was curiously limited for a very long time. On the one hand, it was only in the 1960s that linguists began to take an interest and showed that presupposition was an all-pervasive phenomenon far beyond the philosophers’ pet definite descriptions. And on the other hand, and this is our real concern, it is now only too obvious that the uniqueness condition is too restrictive to be applicable to the general case. An utterance of ‘‘The cat is on the mat’’ should not imply that there is only one cat and one mat in the whole world. The obvious move is to limit the uniqueness condition to some notion of utterance context. Theories of context-dependent propositions began to sprout in the 1970s (with Gazdar, Karttunen, and Stalnaker as the main protagonists), but a full-fledged notion of a context which evolves not only as we go through a sequence of sentences, but also while we analyze the internal structure of sentences, was only available with the advent of dynamic semantics. Dynamic semantics came into being through the need to account for anaphoric reference. But before long, van der Sandt showed that the problem of linking the pronoun to a suitable antecedent was but a variant of the all-embracing problem of presupposition projection. So it is hardly a decade ago that presupposition theory has been cast into a promising theoretical form. And as this provided an inspiring starting point for further research, we thought that, at the turn of the century, the time had come to discuss what the new theory had taught us and which promising further perspectives had been opened up. This was the motivation behind the conference on ‘‘Presupposition’’ which we convened in Stuttgart in October 2000, and which gave rise to the papers xi

xii

PRESUPPOSITIONS

AND

DISCOURSE: ESSAYS

OFFERED TO

HANS KAMP

collected in this volume. The conference was funded by the Deutsche Forschungsgemeinschaft, which we gratefully acknowledge. Undoubtedly, presupposition theory is a major chapter in the success story of dynamic semantics. A conference on the topic thus also seemed to us the ideal birthday present for one of the founding fathers of dynamic semantics, our teacher and friend Hans Kamp, on his 60th birthday. To him we dedicate the volume as an expression of our gratitude for his untiring effort to make us understand. Finally, we want to express our gratitude to Ralf Jankowitsch for translating the manuscript into LaTeX, and to Ken Turner for the time and meticulous effort he invested into the making of this book. Rainer Ba¨uerle Uwe Reyle Thomas Ede Zimmermann Editors Stuttgart, September 2009

1

Presupposition: An (un)Common Attitude? ROBERT

VAN

ROOIJ

Abstract In this paper I argue that presupposition should be thought of as a propositional attitude. I will separate questions on truth from questions of presupposition satisfaction by making use of a twodimensional analysis. The update of what is presupposed will be accounted for by eliminating arrows, which also makes possible an appealing analysis of modal subordination.

Introduction In traditional pragmatic theories the notion of context plays two roles. First, it should contain enough information about the conversational situation to determine what is expressed by a sentence. Second, it should contain enough information about what the participants of the conversation commonly assume about the subject matter of the conversation to determine whether what is said by a speaker is appropriate or not. The central idea behind Stalnakerian pragmatics is that there is a single notion of context that plays both of these two roles, and that both kinds of information modeled by this single context change during a conversation in an interactive way. A context, modeled

Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

1

2

ROBERT

VAN

ROOIJ

by a set of possibilities, represents that what is presupposed by the participants in a conversation. Despite the fact that Discourse Representation Theory (Kamp, 1981; Heim, 1982) and dynamic semantics (Groenendijk & Stokhof, 1991; Veltman, 1996) can be looked upon as attempts to incorporate Stalnaker’s ideas into a rigorous theoretical model, the resulting dynamic theories differ on some essential points from Stalnaker’s suggestions. First, where Stalnaker (1978) argued that the possibilities that are used to represent contexts should be possible worlds, proponents of these dynamic theories account for the antecedentpronoun relation in terms of possibilities that are finer-grained than worlds. Second, although Stalnaker always argued for a dynamic view of language use, he didn’t give up the traditional distinction between content (truth conditions) and force (the way a sentence changes a context) of an assertion, while in dynamic semantics the meaning of a sentence is equated with its context-change potential. Third, where Stalnaker tried to explain linguistic presupposition in terms of what speakers normally presuppose by their use of these sentences, and thus taking presupposition to be primarily a propositional attitude, dynamic semantics either accounts for presuppositions in a way equivalent to Peters’ (1975) three-valued logical account (Beaver, 1995), or (partly) in terms of a syntactic underspecification analysis (van der Sandt, 1992). The three ways in which standard dynamic semantics differs from Stalnaker’s original suggestions are closely related to each other. First, what is presupposed by the participants in a conversation is according to all a crucial contextual parameter to determine content and appropriateness of sentences. Stalnaker argues that it is an attitude playing a role in action very similar to that of belief. As a result it should be modeled in terms of possibilities whose fine-grainedness is relevant for the analysis of deliberation: possible worlds. Dynamic semanticists – following Lewis (1979) for the analysis of belief – argue that contexts should consist of possibilities much finer-grained than worlds, i.e. world-assignment pairs. In distinction with Lewis (1979), however, no one has ever explicitly argued how this fine-grainedness could be relevant for action. Perhaps because proponents of dynamic semantics have given up the idea that contexts represent that what is presupposed, i.e. a propositional attitude of participants in a conversation. Now, Stalnaker (1998) argues for a partly referential analysis of anaphoric pronouns, and one of the main reasons for this – no doubt – is that in this way the fine-grainedness of possibilities could, and should, be that of possible worlds. The reason being that on a

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

3

referential analysis of anaphoric pronouns the use of a singular pronoun comes with a uniqueness assumption. Second, a three-valued logic accounts for linguistic presupposition in terms of entailment. One of Stalnaker’s reasons to account for the behavior of linguistic presuppositions in terms of what speakers presuppose is to be able to account for the intuition many people have that the truth of the linguistic presupposition of a sentence can be irrelevant to the truth or falsity of the sentence, or its content. All that matters is whether the linguistic presupposition is already satisfied by the context. As a result, we should be able to determine the truth-value of sentences in worlds/possibilities outside the context; something that is impossible in standard dynamic semantics because no distinction is made between content and force. Third, if we want to respect the distinction between content and force, we have to be able to determine the truth-value of a sentence containing an anaphoric pronoun in possibilities outside the context. It is not at all clear how to do this when no uniqueness requirement is made on the use of singular pronouns. When such a requirement is made, however, it is easy to see that the content-force distinction can be maintained. In a related paper (van Rooij, 2001) I argue to account for the antecedent-pronoun relation in a way that respects the distinction between content and force, and – by adopting a uniqueness requirement for singular pronouns – model possibilities (essentially) as fine-grained as worlds. In this paper I will deal with presuppositions. I will think of presupposition more explicitly as a propositional attitude, account for this attitude in possible world semantics, explain some presuppositional phenomena in terms of it, while respecting the distinction between content and force. Before I will do that, however, I will first state the way in which presuppositions are standardly accounted for within dynamic semantics.

Standard Implementation According to dynamic semantics, the meaning of a sentence is its context-change potential, where contexts are identified with information states that represent what is commonly assumed in a conversation. The meaning of a sentence is modeled as an update function that takes a context as its argument and has the updated context where the sentence is accepted as its value. Assuming that a sentence cannot be used appropriately in a context that does not entail, or satisfy, its triggered presupposition, this function will be partial.

4

ROBERT

VAN

ROOIJ

I limit myself here, and in the rest of this paper, to the propositional case and represent a context, C, by a set of possible worlds. A possible world is a function from atomic formulae to the two classical truthvalues. Just like Veltman (1996), I treat might, B, as a test-operator. I will follow Beaver (1995) in using a special presuppositional connective ‘q’. We might treat disjunction and implication syncategorematically, by having ‘AWB’ and ‘A-B’ stand for ‘:(:AX:B)’, and ‘:(AX:B)’ respectively. The update function is defined as follows: [A](C) ¼ {wAC |w(A) ¼ 1}, if A is atomic [:A](C) ¼ C[A](C) [AXB](C) ¼ [B]([A](C))  [BA](C) ¼ C, if [A](C) 6¼ |, | otherwise  [qA](C) ¼ C, if [A](C) ¼ C, undefined otherwise   

The appealing feature of this analysis of presuppositions within dynamic semantics (as stressed by Heim (1983)) is that it seems to solve the projection problem simply by means of rules of interpretation. Assuming that context C satisfies presupposition P iff [P] (C) ¼ C, we can say that sentence A presupposes P iff for all contexts C, [A](C) is defined only if C satisfies P. As a result, it follows for instance that if A presupposes P, sentences of the form :A, BA and AXB do so too, but that BXA need not.

Presupposition as a Propositional Attitude Motivation An important insight shared by Stalnaker and Kamp is that we presuppose not only something about the subject matter of conversation, but also about the conversational situation itself. Perhaps the most important kind of information about the conversational situation that agents have presuppositions about is the information that (other) agents presuppose (about the conversational situation). For reasons like this, Stalnaker (1970, 1973, 1974, 1978, 1998, 2002) argues that presupposition should be thought of as a propositional attitude. According to the functional analysis of attitudes, an agent stands in a certain attitude relation to a proposition, if by means of this relation, together with the assumption that the agent is rational, we can explain the behavior of the agent. Attitudes are seen as dispositional, or functional, states of a rational agent, and these states are individuated by the role they play in determining the behavior of the agent who is in such a state. This picture suggests that contexts

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

5

represent presuppositions and should also be thought of as propositional attitudes: we have to know what the speaker is presupposing in order to explain his behavior when he is engaged in a conversation. According to Stalnaker (1970), we should explain the appropriateness of what someone says not only in terms of what he believes and desires, but also partly in terms of what he presupposes. To be able to explain the actions of rational agents, we must assume that the believers know their own minds, i.e. have introspective access to their own minds. But if speech is action, and if the appropriateness of the speech acts of agents is to be explained partly in terms of what agents presuppose, then we also have to assume that the attitude of presupposition similarly allows ‘introspective’ access to its content: The world, as Wittgenstein said, is all that is the case, and a possible world is all that would be the case if that world were actual. Every proposition, relevant or not, that is taken for granted by the participants in a conversation will be true in all of the possible worlds that define the context. [ . . . ] Facts about what is presupposed in a context are not only facts about the actual world in which the discourse is taking place (which may or may not be a member of the context set), but also facts about the worlds that define the context. (Stalnaker, 1998: 6)

The concept of presupposition is closely related with that of common knowledge. Presupposition and common knowledge are closely related because in the ideal case the two coincide. The worlds compatible with what is commonly known are in the ideal case the relevant alternatives with respect to which we have to judge the informativity and acceptability of that what is asserted by speakers. Suppose that in a Kripke model Ki and Kj are the reflexive, transitive and symmetric epistemic accessibility relations modeling the knowledge of i and j, the two participants of the conversation.1 What is commonly known by them can then be represented by K fi;jg , the reflexive transitive closure of Ki [ Kj.2 What is commonly known by the agents in world w can be represented by fv 2 W : wK fi;jg vg, abbreviated by K fi;jg ðwÞ. Although in the ideal case the context represents what is common knowledge for the participants in the conversation, it is clear that ideal conditions do not always obtain. For one thing, what is presupposed 1 A relation R is reflexive if ’x : xRx ; transitive if ’x, y, z : (xRy & yRz)-xRz; and symmetric if ’x, y : xRy-yRx. 2 If /x, yS and /y, zS are in the union of relations R and Ru, R [ Ru, then both /x, xS and /x, zS are also in the reflexive transitive closure of R [ Ru.

6

ROBERT

VAN

ROOIJ

need not be true: discourse can be based on an assumption that later turns out to be false. So, it seems that presupposition should be associated with common belief rather than with common knowledge. The attitude of common belief should be based on the doxastic accessibility relations of the agents, relations that need not be reflexive. This non-reflexivity represents the fact that agents can have false beliefs. If presupposition is analyzed as common belief, we correctly predict that also presupposition is a non-veridical propositional attitude. An important motivation for treating presupposition as a nonveridical propositional attitude is that we can then respect the traditional distinction between the content and the force of a speech act (of assertion), and can separate questions of entailment from questions of presupposition satisfaction. The views that we should separate content from force, and valuation of truth from presupposition satisfaction are closely related, and they have been defended consistently over the years by Stalnaker. I suggested that an assertion should be understood as a proposal to change the context by adding the content to the information presupposed. [ . . . ] Meaning determines the content of an assertion as a function of the context, and the assertion rule takes the prior context set to a posterior context set [ . . . ] Some of the dynamic semantic theories subsequently developed by linguists have blurred the distinction between content and force by combining the two steps (meaning plus prior context to content, and prior context plus content to posterior context) into one. [ . . . ] I think this streamlined representation captures much of what is important about the dynamic process of speech, but what it leaves out is the possibility of evaluating the truth or falsity of what is said relative to possible situations that are not compatible with the prior context. Sometimes when a statement rests on false presuppositions, the question of the actual truth of the statement does not arise, but other times a speaker may succeed in making a claim that is actually true or false, even when taking for granted, in making the claim, something that is in fact false. In such cases, our semantic theory should tell us what is said, and not just how what is said changes the context. Sentences that say different things in some contexts may nevertheless change contexts in the same way. (Stalnaker, 1999: 11)

One example he discusses for which the separation of content from force seems crucial is Donnellan’s (1966) case of the referential use of definite descriptions. When the description in the sentence The man drinking a martini is a philosopher is used referentially, the proposition

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

7

intended to be communicated/expressed might be true, although it presupposes a falsehood. Stalnaker (1973, 1974), followed by linguists like Karttunen and Peters (1979), argues that in general the proposition expressed by a sentence might be true independent of the truth of the presupposition: a sentence like Even Bill likes Mary can be true without it being unlikely that Bill likes Mary. So, although the use of the sentence gives rise to this presupposition, the sentence by itself doesn’t entail it. This suggests that the truth-value of, or proposition expressed by, a sentence should be determinable independently of the truth of the presupposition. Note, however, that this is impossible in standard dynamic semantics, where all of the attention is given to the update of what is presupposed, and where truth is treated as at best a derived notion. But if in semantics truth and truth conditions are of primary importance, we should be able to say when a sentence is true, even if it is interpreted with respect to a presupposition state that is non-veridical. Thus, truth and presupposition satisfaction should be accounted for on different dimensions. To be able to do this, we have to define the truth conditions of sentences in a manner different from that of standard update semantics – namely, one that is more independent of the update function of what is presupposed. The non-veridicality of contexts suggests that we should treat the valuation of truth separately from context change – distinguish content from force. In the remainder of this paper I will show how we can systematically account for presupposition satisfaction without giving up the possibility of determining the content of a sentence separately from the way it changes the context. For context change, I will rely mainly on work in dynamic epistemic semantics, where updates are defined in terms of eliminating arrows instead of eliminating worlds.3 Formalization When a speaker presupposes something, he presupposes it in a world or a possibility. A possibility will be represented by a pointed model, /M, wS, where M ¼ /W, Ba, BbS and w a distinguished element of W representing the actual world and should be thought of as a valuation function from atomic propositions to truth-values. The Bi s in the 3 Updating through the elimination of arrows instead of worlds has been used, among others, by Landman (1986a) and Veltman (1996). Its limitations for multi-agent settings are discussed in Gerbrandy (1999). Note that elimination of arrows can (as far as this section is concerned) be reduced to elimination of words if (as in ordinary Kripke models) the latter is defined as determining truth-value assignments without necessarily being such assignments themselves.

8

ROBERT

VAN

ROOIJ

model are accessibility relations that are serial, transitive and Euclidean and represent what is believed by i.4 I will assume with Stalnaker (2002) that what is presupposed by agent a in a conversation with agent b is what the first agent thinks is commonly believed between a and b. Assuming that Ba represents the doxastic accessibility relation of a, what is presupposed by a in w, abbreviated by Ra(w), is the following set (where * is the transitive closure operator):5    Ra ðwÞ ¼ [v2B aðwÞ B fa;bg ðvÞ Presupposition satisfaction suggests that conjunction should be treated in an asymmetric way. Most authors agree that one must account for this either distributively in terms of Peters’ (1975) threevalued logic, or globally in terms of context satisfaction in dynamic semantics.6 In this paper I will make use of insights of manydimensional logics and of dynamic semantics as well. However, I will do this in an unusual way. First I will assume the two-dimensional (or four-valued) analysis of Herzberger (1973) and Karttunen and Peters (1979) according to which the logic of truth and that of presupposition should be treated at separate dimensions. Karttunen and Peters assume that one then has to represent presupposition and assertion at different dimensions, but this is not needed. Making use of Beaver’s (1995) presupposition operator, we will represent an atomic sentence A that presupposes P asserted by speaker i as follows: qiPXA. To determine in possibility /M, wS whether P is presupposed by i, we have to check what is presupposed in this possibility by i, Ri(w). Thinking of presupposition as a non-veridical propositional attitude, we can account for the dynamic aspects of presupposition satisfaction without giving up the idea behind a two-dimensional analysis of presupposition satisfaction. That is, although we will predict that conjunction behaves asymmetrically with respect to presupposition satisfaction, ‘and’ will still be treated in a symmetric way. The reason is that truth and presupposition satisfaction are defined separately from the update A relation R is serial on W if ’xAW: (y : xRy and Euclidean if ’x, y, z : (xRy & xRz)-yRz. 5 The transitive closure of the two Euclidean relations Ba and Bb need not be Euclidean itself. Thus, although negative introspection holds for what is believed, it doesn’t hold for what is commonly believed. This only holds in a world w in which B a ðwÞ \ B b ðwÞ a ;. Of course, once we assume that only those things are presupposed that are explicitly asserted in a conversation, it is very natural to assume that the presuppositional relation respects negative introspection. This will be important later. 6 One can also easily prove that the two come down to the same thing. 4

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

9

function (although they will be defined simultaneously). For the time being I will concentrate only on the truth-conditional connectives. We will assume that a sentence has two values: (i) a sentence is true or false, i.e. 1 or 0; (ii) a sentence has no presupposition failure or it has one, i.e. þ or . The combined truth and presupposition satisfaction conditions of sentences are given below: [[A]]M,w ¼ /w(A), þS, iff A is atomic,  [[:A]]M,w ¼ /1, þS iff [[A]]R,w ¼ /0, þS  [[AXB]]M,w ¼ /1, þS iff [[A]]M,w ¼ /1, þS and [[B]]Upd(A,M),w ¼ /1, þS  [[qiA]]M,w ¼ /1, þS iff ’vARi(w) : [[A]]M,v ¼ /1, þS 

Observe again that the presupposition value of a conjunction is determined in a symmetric way. That is, if either A or B has a presupposition failure, the conjunction AXB will have a presupposition failure as well. However, to determine the presupposition value of a conjunction of the form AXB in possibility /M, wS, we look at the presupposition value of B in possibility /Upd(A, M), wS – the update function is being relevant here. This is the point at which we take over the insights of dynamic semantics. The update of M with A, Upd(A, M) should really be thought of as the change from one model to another. So, from updating model /W, Ba, BbS with a proposition we go to another unique model hW 0 ; B 0a ; B 0b i. Assuming that W u ¼ W, we take it that only the accessibility relations of the agents involved in the conversation will change. The update Upd(A, M), is defined as follows (assuming that a and b are participants of the discourse and that b believes what a tells him): 

Upd(A, M) ¼ /W, Upd(A, Ba), Upd(A, Bb)S

Notice that although speaker a will typically already believe what he asserts, the update of accessibility relation Ba with A, Upd(A, Ba), is still required, because the other participants did not yet know that he believes it. The update of an accessibility relation is then defined as below: 

Upd(A, R) ¼ {/u, vSAR| [[A]]R,v ¼ /1, þS}

Notice that this update function is eliminative, but instead of eliminating worlds in R(w) it eliminate tuples, or arrows, in R. It eliminates all arrows in R that point to a non-A-world. This has the effect that after the update of R with A, not only all accessible worlds v verify A, but also all worlds u accessible from v make A true. Thus,

10

ROBERT

VAN

ROOIJ

after the update with A it is not only presupposed that A, but it is also presupposed to be presupposed that A. According to the above update function we can account for updates of presupposition states by means of change of accessibility relations. But if a world represents everything that is the case, wouldn’t it be more natural to claim that after update the accessibility relation stays the same but the world changes? Indeed, that seems to be the case. Fortunately, however, this is exactly what we have done. Let me explain. According to dynamic logic (Harel, 1984) the meaning of a sentence is seen as a relation between states. Normally states are just seen as states in one model, but to account for changes in belief/ presupposition we have thought of updates as relations between pointed models. But then, what does a pointed model like /M, wS represent? Well, we have assumed that a ‘naked’ world represents only information about propositional variables, and not about what agents believe and/or presuppose in that world. To also represent the latter kind of information we need accessibility relations and other worlds to be given in the rest of the model. Now, if a world is everything that is the case, this means that a world should not be represented by a function from propositional variables to truth-values, but rather by a pointed model, /M, wS. So we see that the assumption that an update is defined as a function from one pointed model to another is in accordance with the basic intuition that the world changes after something new is presupposed, if we think of a world as everything that is true in this world according to the model.7 If we assume that sentence A presupposes P iff ’i : ’/M, wS: if [A]M,w ¼ þ, then ’vARi(w) : [[P]]M,v ¼ 1, the above implementation gives rise to the same presuppositional predictions as the standard implementation of the satisfaction account. In particular, on the assumption that John stopped smoking gives rise to the presupposition that John used to smoke, this implementation predicts that sentences like John didn’t stop smoking and John stopped smoking and Mary is sick will also give rise to this presupposition, but John used to smoke and he stopped doing so will never give rise to presupposition failure. That the predictions of our implementation considered above are the same as according to the standard implementation should not surprise you: they are based on the fact (cf. Gerbrandy, 1999) that 7 If we represented updates (i.e. actions) as formulas too, this would mean, formally, that a model (or supermodel ) consists of a collection of pointed models and that the meaning of an action can be thought of as an accessibility relation between pointed models. See Baltag (1999) for the general picture.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

11

updating by eliminating arrows corresponds closely with update in standard update semantics. That is, if we think of worlds just as functions from propositional variables to truth-values, it holds for any sentence A, context C and world w: If C ¼ R(w), then [A](C) ¼ Upd(A, R)(w). Although the predictions of the above implementation of the satisfaction approach are similar to the predictions on the standard approach, there are still some important differences. First, note that by treating presupposition as a propositional attitude, we can evaluate in a distributive way whether a presupposition associated with a sentence is satisfied by what the speaker presupposes. This is possible, of course, because we have represented in a single possibility all the information that is normally represented only in a whole context/information state. Second, and related, we can now account for the dominant view in the seventies that presupposition satisfaction and truth should be evaluated at different dimensions. [ . . . ] if presupposition is defined independently of truth-conditions, then we can separate the question of entailment relations from the question of presupposition. [ . . . ] one may say that sometimes when a presupposition is required by the making of the statement, what is presupposed is also entailed, and sometimes it is not. One can say that ‘‘Sam realizes that P’’ entails that P – the claim is false unless P is true. ‘‘Sam does not realize that P,’’ however, does not entail that P. That proposition may be true even when P is false. All this is compatible with the claim that one is required to presuppose that P whenever one asserts or denies that Sam realizes it. (Stalnaker, 1974: 54)

We have already seen that according to Karttunen and Peters (1979) and others a sentence like Even Bill likes Mary presupposes something that it does not entail. Thus, the sentence can be true without it actually being unlikely that Bill likes Mary, because what is presupposed need not be true. Notice that we can now account for this intuition without assuming with Karttunen and Peters (1979) that we should thus represent presuppositions separately from assertions.8 On the other hand, we can also account for the intuition that a factive verb both presupposes and entails that its complement is true.9 To analyze Sam realizes that P we add the following construction 8 Which has interesting consequences with respect to the traditional binding-problem. I will limit myself in this paper to the propositional level, however, and the bindingproblem will thus not bother us. 9 Throughout the paper I will assume the same for an aspectual verb like stop.

12

ROBERT

VAN

ROOIJ

to the language: if P is a sentence, Real(s, P) is a sentence too. To interpret the formula we add a primitive reflexive accessibility relation to the model, Ks, modeling what Sam realizes.10 The formula is then interpreted as follows: 

[[Real(s, P)]]R,w ¼ 1 iff ’vAKs(w) : [[P]]R,v ¼ 1

Notice that because Ks is reflexive, according to this analysis the formula entails, but does not presuppose, that P. To account for the presupposition, we represent the sentence Sam realizes that P by the formula qPX Real(s, P), which both presupposes and entails that P. If we now represent Sam does not realize that P by :(qPXReal(s, P)), this sentence presupposes that P, but can still be true in case P is false (in case weR(w)).

No Cancellation or Local Accommodation Consider the following well-known problematic examples for the traditional satisfaction theory: (1) If I realize later that I have not told the truth, I will confess it to everyone. (2) Frank doesn’t know that the earth is flat, because the earth isn’t flat. (3) a. Either John never used to smoke, or he stopped smoking, b. Either John stopped smoking, or he never used to smoke. These examples are problematic for the standard satisfaction approach because this account wrongly predicts in all these cases that the sentences give rise to presuppositional readings that intuitively are not available. Sentence (1) is wrongly predicted to presuppose that the speaker did not tell the truth; sentence (2) is predicted to presuppose that the earth is flat, which is in conflict with what is asserted; and sentences (3a) and (3b) are predicted to give rise to a presupposition incompatible with what is asserted in the other disjunct, which is absurd. Traditionally, these examples gave rise to the hypothesis that presuppositions can sometimes be cancelled for reasons of 10 Our simple update function for models has limitations here: if we attributed to Sam attitudes about what the discourse participants believe, things would go wrong. I will ignore such attributions in this paper. See Baltag (1999) and Gerbrandy (1999) for analyses where this can be accounted for too.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

13

informativity,11 and in more recent discussions, Heim (1983) and van der Sandt (1992) argue that presuppositions should sometimes be locally accommodated. But there are problems with both proposals, both formally and conceptually. The formal problem is that it is not at all clear how to account for cancellation and/or local accommodation in the framework of the satisfaction approach. The conceptual problem for cancellation is that it becomes unclear why the presupposition trigger was used in the first place, and for local accommodation how to explain what is supposed to be going on when we locally accommodate a presupposition. In the rest of this paper I will suggest how to account for these apparent counterexamples of the satisfaction analysis by assuming that there might be more than one information state around that could satisfy the triggered presupposition. I will do this all in terms of the above stated possible world analysis. Conditionals Let’s first discuss conditional sentences. I will follow Stalnaker (1968) in assuming that both indicative and subjunctive conditionals should be analyzed in terms of selection functions/similarity relations. Let us assume with Stalnaker and Lewis that there exists a similarity relation rw between worlds, urw v meaning that u is at least as close to w as v is. To let the selection explicitly depend on what is presupposed, I will assume that the relation rw ignores what agents presuppose in the worlds. Instead, I will define a new relation oR,w between possibilities that is dependent both on rw and on what is presupposed in /R, wS (where ow and Ew are defined in terms of rw in the usual way, and where R is the presuppositional accessibility relation of the speaker):12 

hR0 ; vioR;w hR00 ; ui

iff ðiÞ vow u; or ðiiÞ v w u and R00 ðuÞ  R0 ðvÞ  RðwÞ; or ðiiiÞ v w u and R00 ðuÞ  R0 ðvÞ  R00 ðwÞ

Thus, /Ru, vS is closer to /R, wS than /Rv, uS iff either v is closer to w than u, or they are equally close, but what is presupposed in /Ru, vS is more similar to what is presupposed in /R, wS than what 11 We might interpret the proposals of Gazdar (1979), Soames (1982), van der Sandt (1988) and some remarks of Stalnaker (1974) in this way. 12 In this subsection I will ignore the possibilities that more than one presupposition state might be around.

14

ROBERT

VAN

ROOIJ

is presupposed in /Rv, uS. The set of closest A-possibilities to /R, wS is the following set: 

0

f hR;wi ðAÞ ¼ fhR0 ; vij ½½AR ;v ¼ h1; þi & :9hS; ui : ½½AS;u ¼ h1; þi & hS; uiohR;wi hR0 ; vig

A conditional sentence of the form if A then B is then counted as true in /R, wS iff all the closest A-possibilities to /R, wS updated with A are B-possibilities: 

0

½½A4BR;w ¼ h1; þi iff f hR;wi ðAÞ  fhR0 ; vij ½½BUpdðA;R Þ;v ¼ h1; þig

Following Stalnaker’s (1975) suggestion that the antecedent of an indicative conditional selects, if possible, worlds compatible with what is presupposed, I will assume that for such conditionals the selection function has to satisfy the following condition:  f/R,wS(A)D{/Ru,

vS|vAR(w) & RuDR}

Now we are ready to discuss a traditional problem for the satisfaction approach.13 Karttunen (1974) and Stalnaker (1974) noted that although the following sentences intuitively express the same proposition, it is clear that the speaker of (4b) will not presuppose

13 According to Geurts (1995) conditional sentences are a major problem for the satisfaction approach toward presuppositions. This is already the case, according to Geurts, when conditional sentences are analyzed as material implications, but problems even increase, still according to Geurts, when the sentences are analyzed in terms of similarity relations. Assuming that c{w} means that sentence c presupposes w, and that conditionals are analyzed in terms of similarity, Geurts claims the following:

[ . . . ] the satisfaction theory implies that even if w is contextually given, the presupposition in fWc{w} isn’t automatically satisfied – a consequence which strikes me as paradoxical if not absurd. (Geurts, 1995: 282) Paradoxical or not, it is exactly what Heim’s (1992) analysis of conditional sentences in her context-change analysis of presuppositions predicts. She analyzes presuppositions as we did in the Section on Standard Implementation, and proposes the following analysis for conditionals (where f is a function which assigns to a world and a set of worlds a subset of the latter): 

½A4BðC Þ ¼ fo 2 C jf o ð½Að[fK Cj½AðKÞ is definedgÞÞ

¼ ½Bf o ð½Að[fK Cj½AðKÞ is definedgÞÞg But this has the unwanted prediction Geurts points to. Notice, however, that this wrong prediction has nothing to do with the satisfaction analysis per se, but only follows from Heim’s particular implementation of things. In fact, the analysis I propose in this section doesn’t give rise to this unwanted prediction.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

15

that Harry has not told the truth, while the speaker of (4a) typically does.14 (4) a. b.

If Harry realizes later that he has not told the truth, he will confess it to everyone. If I realize later that I have not told the truth, I will confess it to everyone (said by Harry).

Both antecedents presuppose that Harry has not told the truth. Why does the first conditional also have this presupposition, but not the second? Notice, first, that everything else being equal, if Ru(v) ¼ R(w) 6¼ Rv(u), /Ru, vS is considered to be closer to /R, wS than /Rv, uS: possibilities where the same is presupposed as in /R, wS are preferred to possibilities where something different is presupposed. This explains why in normal cases like (4a) the presupposition of the antecedent is also a presupposition of the whole conditional. To explain why this is not the case for (4b), let’s see again how we represent the antecedent of this sentence (where I assume that the speaker is h): qhPXReal(h, P). Notice that for possibility /Ru, vS to be among the selected ones, it has to be the case that the speaker, i.e. h himself, presupposes in this possibility that P is true, otherwise the formula cannot be true and appropriate in /Ru, vS. But this means that what is presupposed in /Ru, vS cannot be the same as what is presupposed in /R, wS, otherwise the antecedent would not be informative anymore with respect to what the speaker presupposes. [ . . . ] If a speaker explicitly supposes something, he thereby indicates that he is not presupposing it, or taking it for granted. So when the speaker says ‘‘if I realize later that P,’’ he indicates that he is not presupposing that he will realize later that P. But if it is an open question for a speaker whether or not he will at some future time have come to realize that P, he can’t be assuming that he already knows that P. And if he is not assuming that he himself knows that P, he can’t be assuming that P. Hence P cannot be presupposed. (Stalnaker, 1974: 208)

To account for this informativity, it has to be the case that R(w) contains worlds in which P is false. But this means that any selected possibility /Ru, vS must be such that Ru(v) 6¼ R(w). Because any 14 The contrast between the sentences is, of course, closely related with the contrast between the following sentences, of which only the latter gives rise to Moore’s paradox:

(i) The cat is on the mat, but Harry doesn’t believe it. (ii) ?The cat is on the mat, but I don’t believe it. (said by Harry)

16

ROBERT

VAN

ROOIJ

selected possibility /Ru, vS verifies the ‘fact’ that P is presupposed, the presupposition is satisfied in its local context. Still, it is not a presupposition of the whole conditional because /R, wS will not verify the ‘fact’ that P is presupposed.15 Notice that according to this analysis we don’t need to assume anything like local accommodation. Still, the effect of our analysis will be very similar to the standard local accommodation approach. The reason is that the possibilities /Ru, vS with respect to which the antecedent will be evaluated will be closely related with the possibility resulting after accommodation of ‘original’ possibility /R, wS with the presupposition P of the antecedent: R(w) \ P ¼ Ru(v). Now consider the following type of example discussed by Landman (1986b): (5) If John regrets that it snows heavily, then at least it snows. The standard satisfaction account predicts wrongly, again, that this kind of conditional inherits the presupposition of its antecedent. Landman (1986b) proposes to avoid this false prediction by adopting the following rule: if f presupposes fu then the assertion f-c presupposes fu unless fu already provides an answer to the question whether the consequent c is true. This rule has much in common with Gazdar’s (1979) cancellation approach and that’s exactly what we wanted to avoid. On our analysis of conditionals, however, we don’t need the rule, but just an additional constraint on the selection function. Notice first that our analysis would not predict that out of context (5) is inappropriate in case it is not yet presupposed that it snows heavily: in special circumstances, what is presupposed in the selected possibilities need not be the same as what is presupposed in the actual possibility. As we saw above, these special circumstances can have something to do with the informativity of the antecedent of the conditional. Landman’s examples show, however, that they can involve the informativity of the consequent as well. It is natural to demand (cf. Gazdar’s clausal implicatures) that the truth-value of the consequent of an indicative conditional is not yet settled by what is actually presupposed. Thus, if /R, wS is the actual possibility, both the consequent and its negation should be consistent with R(w). But this means that for the interpretation of (5), R(w) cannot yet contain the information that it snows. Because the antecedent presupposes it, we 15 Given our analysis of modal subordination below, it seems natural to assume that for AWB to have value /1, þS in /R, wS, it has to be the case that [[BA]]R,w ¼ /1, þS.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

17

predict that for each selected possibility /Ru, vS with respect to which the antecedent has to be interpreted, it holds that R(w) 6¼ Ru(v). Again, the analysis has the same result as what is supposed to happen if we allow for cancellation or local accommodation, but the idea behind my proposal is very different. Denials Following Stalnaker (2002), I defined what is presupposed by a in w in terms of what a believes is common ground. Agents a and b might believe different things, in particular about what is common ground. Indeed, there is no reason to assume why Ra(w) ¼ Rb(w). Thus, in a conversation between a and b there might be several presupposition states around. Because it is the speaker who is responsible for what she says, the presuppositions of what she says should in the first place be satisfied with respect to her own presupposition state. This, at least, is normally the case. However, so I want to argue, this is not so for sentences like: (2) Frank doesn’t know that the earth is flat, because the earth isn’t flat. This example is problematic for the standard satisfaction approach, because it both presupposes that the earth is flat, and asserts that the earth is not flat. If truth and presupposition satisfaction should be analyzed with respect to the same context, how should we account for such examples? The standard answers, as we have seen, are cancellation and local accommodation. In this section I want to suggest that (2) does indeed give rise to the presupposition that the earth is flat, as the standard satisfaction account predicts, but that it is not the presupposition of the speaker, but rather that of the addressee. The idea is that a sentence like (2) is typically uttered after the other participant in the conversation has made clear (perhaps, but not necessary, by an explicit claim) that he presupposes that the earth is flat. In terms of van der Sandt (1991), this means that (2) is typically used as a denial. To account for the intuition that sometimes a speaker indicates that a presupposition is made not by himself, but rather by the other participant in a conversation, we will index the presupposition operator by the relevant agent. Thus, qjP will mean that agent j presupposes that P is the case. As might be expected, we will represent (2) by the following formula: (6) :(qjPXknow(f, P))X:P

18

ROBERT

VAN

ROOIJ

Assuming that Kf is the reflexive accessibility relation that models what Frank knows, the first conjunct is analyzed as follows:  ½½:ð@j P^knowðf ; PÞÞM ;w ¼ h1; þi iff ½½@j P ^ knowðf ; PÞM ;w ¼ h0; þi iff 8v 2 Rj ðwÞ : ½½PM ;v ¼ 1 and ½½knowðf ; PÞM ;w ¼ 0 iff 8v 2 Rj ðwÞ : ½½PM ;v ¼ 1 and 9u 2 K f ðwÞ : ½½PM ;u ¼ 0 Notice that this first conjunct is obviously compatible with the second one: if, and only if, w doesn’t make P true and weRj(w) both conjuncts can be true. The point of what the speaker says by (2) is in fact that the other participant presupposes something that is false: the sentence as a whole can be true only in case Rj is non-reflexive. Thus, we might say, the sentence is used as a presuppositional denial. We can conclude that to account for denials we don’t have to assume that presuppositions are cancelled or locally accommodated: they have to be satisfied, but not necessarily by the information state that represents what the speaker presupposes. Disjunction The standard satisfaction analysis of presuppositions predicts that the following sentences presuppose that John used to smoke.16 (3) a. b.

Either John never used to smoke, or he stopped smoking. Either John stopped smoking, or he never used to smoke.

Karttunen (1974) suggested that this problem can be overcome when we assume that the context with respect to which we evaluate the appropriateness of one disjunct is the initial context updated with the negation of the other disjunct. However, as already noted by Karttunen himself, this won’t work for a sentence like (7) Either John stopped smoking, or he just started doing so. Gazdar (1979), Soames (1979, 1982) and van der Sandt (1982, 1988), among others, propose to account for all of (3a), (3b) and (7) by assuming that presuppositions of parts of complex sentences can sometimes be cancelled,17 for reasons of inconsistency. In this case, because the presupposition of the one disjunct would be in contradiction with what is asserted (or presupposed) by the other. 16

Although if we treat AWB as an abbreviation for :(:AX:B) this isn’t really true for (3a). 17 Although certainly Gazdar would not describe matters this way.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

19

To account for cancellation, Gazdar, Soames and van der Sandt must assume that what is presupposed by the embedded clause is also entailed by it. This avoids the implausible claim that presupposition triggers sometimes have no effect at all on the semantic/pragmatic meaning of the sentence. Not happy with the prediction that all triggers should have not only presuppositional but also assertive content, however, Heim (1983) and van der Sandt (1992) argue that in cases like (3a), (3b) and (7) the triggered presuppositions should be locally accommodated. In the previous section we saw that we can account for cases where others have made use of local accommodation, and now we will propose a similar approach by following a suggestion made by Landman. Landman (1986b) proposes a solution in the spirit of the satisfaction approach by making use of Robert’s (1989) analysis of modal subordination. Landman argues that to account for (7), we should make the assumption that the embedded clauses should not be interpreted with respect to the main context, but with respect to two mutually exclusive subordinated contexts, possibly created by an earlier use of a disjunctive sentence that has split the context. But how can this be done without giving up our assumption that what is presupposed can be represented by a single accessibility relation? Before we answer this question, it is useful to first discuss a simpler case.

Modal Subordination Possibility In the Section on Standard Implementation we followed Veltman’s (1996) analysis of possibly: the embedded sentence of ‘possibly A’ should be interpreted with respect to the same context as the whole sentence. It is easily seen, however, that sometimes this predicts too strong presuppositions. It is predicted that (8a) presupposes (8b). (8) a. b.

It is possible that John stopped smoking. John used to smoke.

However, if it has already been established that it is possible that John used to smoke, (8a) need not presuppose (8b). Consider (9a) and (9b). (9) a. b.

It is possible that John used to smoke, and it is possible that he just stopped doing so.

The most natural explanation for the fact that (9b) does not give rise to the presupposition that John used to smoke seems to me that the embedded sentence in (9a) introduces a subordinated context into the

20

ROBERT

VAN

ROOIJ

discourse that contains this information. How can we account for the introduction of these subordinated contexts without giving up the strategy of representing what is presupposed in terms of a single accessibility relation? Until now we have assumed with Stalnaker that what is presupposed can be determined in terms of what is believed by the participants of the conversation. Discourse has an effect on what is believed, and thus indirectly on what is presupposed. According to Kamp’s Discourse Representation Theory, discourse has a more direct effect on what is presupposed. Here, a context, or DRS, represents what is explicitly agreed upon by the participants of a conversation. This suggests that what is presupposed should be represented by a separate and primitive accessibility relation. As before, I will denote this accessibility relation by R, but now I assume that it represents what is established to be presupposed in a discourse. Moreover, the assumption that R represents only what conversational partners explicitly agree upon suggests that it – in contrast to what we had earlier – is by default fully introspective:18 R is not only serial and transitive, but also Euclidean. This means that for all worlds u and v accessible from w, it holds that /u, vSAR. I will show now that in terms of such a primitive accessibility relation we can account for sequences like (9a)–(9b). The basic idea is very simple: possibility statements introduce questions into the discourse. However, because we assume that what is presupposed is a presuppositional attitude and should be represented by an accessibility relation, we can implement this idea in an appealing way. Following Groenendijk and Stokhof (1984), I will represent a question by a partition or equivalence relation, and with Groenendijk (1999) I will assume that the dynamic effect of a question is to add a partition to the context by eliminating arrows.19 

Upd(BA, M) ¼ {/u, vSAR| [[A]]M,u ¼ /1, þS iff [[A]]M,v ¼ /1, þS} 18

See also Fernando (2004) for an analysis of context where full introspection is assumed. He does not account for this, however, in terms of an accessibility relation. 19 This update rule is defined on the assumption that weR(w), because otherwise we would falsely predict that if w is an A-world (or :A-world), the use of the possibility statement would predict that only other A-worlds (or only other :A-worlds) would be accessible from w. In general we cannot make this assumption, of course. Fortunately, we can use a technical trick to solve this problem. Assume that if wAR(w), we don’t go to new pointed model /Ru, wS, but rather to the pointed model /Ru, w*S with a new world w*. This new world is exactly like w of the original pointed model, except that w*eR(w*). Because our technical problem has a simple solution, I will ignore this complication in the main text. (Thanks to Frank Veltman and Henk Zeevat for discussion on this point).

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

21

According to the update function, possibility statements split the context by disconnecting A-worlds from :A-worlds. They do so by eliminating arrows between A-worlds and :A-worlds consistent with what is presupposed (although A-worlds and :A-worlds are still accessible from the actual world w). Thus, if Ri was Euclidean before the update with BA, it won’t be Euclidean anymore afterwards. Possibility statements will be interpreted from now on as follows: 

[[BA]]M,w ¼ /1, þS iff (vAR(w) : [[A]]M,v ¼ /1, þS

According to the appropriateness condition it holds that if A presupposes P, BA can be used appropriately only if it is assumed to be possible that P is presupposed. Because out of context (or so we assumed) it holds that ’vAR(w) : R(v) ¼ R(w), under normal circumstances the truth condition just states that BA presupposes the same as A itself. However, it can also account for the sequence (9a)–(9b). After the interpretation/update of (9a) there is a world u consistent with what is presupposed in the actual world w in which John used to smoke and in which it is presupposed that John used to smoke. It is important to see that from this world only worlds are accessible in which John used to smoke. But this means that the embedded sentence of (9b) can be interpreted appropriately as well. Under which circumstances does it make sense to assume that it is possible that P is presupposed? On our assumption that by default, or out of context, the accessibility relation Ri is fully introspective, this is the case only if P itself is presupposed. Thus, if (9b) was used out of context, it is correctly predicted that John used to smoke should be presupposed already in the actual world. As a result, we predict that possibility statements normally presuppose the presuppositions of its embedded sentences, but not in case the possibility that this presupposition is true has been explicitly mentioned earlier in the discourse. Of course, our analysis of modal subordination has some similarity with the anaphoric ones proposed by Kibble (1994), Geurts (1995, 1998) and Frank (1997). They too are able to account for the intuition that the sequence (9a)–(9b) does not give rise to the presupposition that John used to smoke. However, I believe that my analysis is much more appealing: not only do they need an extra representation which mine does not, their analyses are also much weaker: they just introduce subordinated contexts to the discourse that can be anaphorically picked up freely by any other sentence. Only good

22

ROBERT

VAN

ROOIJ

luck can save them from making false predictions. Moreover, their analyses do not account for the following classical sequence, while mine seems to do so: (10) a. b.

It is possible that John doesn’t have children, but it is also possible that his children are not home.

However, although the update with BA partitions the context, there is a contrast between accessible A-worlds and accessible :A-words: the former are guaranteed to exist, while the latter are not. Some Other Cases Consider (7) again. (7) Either John stopped smoking, or he just started doing so. We saw above that Landman proposed to account for such examples by assuming that the disjuncts should be interpreted with respect to two mutually exclusive subordinated contexts, possibly created by an earlier use of a disjunctive sentence that has split the context. We didn’t know how to account for this before. Our analysis of possibility statements, however, suggests a straightforward solution: 

½½A _ BM ;w ¼ h1; þi

iff

½½AM ;w ¼ h1; i or ½½BM ;w ¼ h1; i and

½½BAM ;w ¼ h1; þi and ½½BBM ;w ¼ h1; þi 

UpdðA _ B; M Þ ¼ UpdðA; M Þ [ UpdðB; M Þ

Thus, I propose that AWB is appropriate iff both BA and BB are appropriate. As for the case of possibility statements, this means that normally all the presuppositions of the disjuncts are also presuppositions of the whole disjunction. However, when the context is split, this doesn’t have to be the case. This accounts for the problematic (7), if we assume that the context was split (perhaps after accommodation of the disjunctive presupposition) between, on the one hand, worlds where John smoked before, and, on the other, worlds where he did not. Notice that a context can be split in this way both by possibility operators and by disjunctions (in case the disjunctions are mutually contradictory). Of course, it can be split by questions as well. Indeed, we can simply assume that the update of R with the yes/no question A?

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

23

is the same as the update of R with BA. And this gives rise to the following correct predictions: after question (11a), both (11b) and (11c) are appropriate and do not give rise to presuppositional readings: (11) a. Did anyone solve the problem? b. It is possible that it was John who solved the problem. c. Either it was John who solved the problem, or the problem was too difficult. What is presupposed by A is normally also presupposed by :A. In the Section on No Cancellation or Local Accommodation, however, we discussed an example for which this doesn’t seem to be true, but suggested that in this case it is just presupposed by another participant of the conversation. Unfortunately, not all counterexamples to the standard implementation of the satisfaction theory can be handled this way. Consider the following example: (12) A:. B:.

I don’t have a dog. So at least you don’t have to walk your dog.

In this example, speaker B does not presuppose that speaker A has a dog, although that is what we would predict until now. It is clear that also the other participant of the conversation does not presuppose that he has a dog: what B says should not be interpreted as a denial. Still, her assertion should be interpreted in a special way, because it cannot be a standard informative assertion where some possibilities consistent with what is presupposed are eliminated. I believe this is the crucial aspect of this example. To make sense of it, I propose to slightly change the presuppositional value of negative sentences as follows (where M[R/Ru] is just like M except that R is replaced by Ru):20  ½½:AM ;w ¼ h1; þi iff ½½AM ;w ¼ h0; þi; if 9v 2 RðwÞ : ½½AM ;v ¼ h1; þi iff ½½AM ½R=ðW

2

RÞ;w

¼ h0; þi otherwise

Thus, :A should be interpreted with respect to the same presupposition set as A – and has thus the same presupposition as A has –, if it is presupposed to be possible that A is true (remember that most simple sentences entail their presuppositions). Otherwise it should be interpreted with respect to the complement of what is presupposed

20

The update function is still defined in the old way.

24

ROBERT

VAN

ROOIJ

(note that RDW2). This means that if negative sentences are used in the normal, informative, way, they presuppose the same as their positive counterparts. In special cases as in (12), however, when their positive counterpart is already ruled out, the negated clause is interpreted with respect to the complement of what is presupposed.

Conclusion In this paper I have argued that presupposition should be thought of as a propositional attitude. This allows us to account for linguistic presuppositions independently of truth conditions, which enables us to separate the question of entailment relations from the question of presupposition. Our analysis differs from other two-dimensional analyses of presupposition proposed in the seventies because it doesn’t have to assume that the assertive and presuppositional part of an utterance should be represented by separate representations. Furthermore, I have suggested how this analysis might account for some (apparent) counterexamples of the satisfaction approach by assuming that there might be more than one information state around that could satisfy the triggered presupposition. Acknowledgment The research work for this paper was supported by a fellowship from the Royal Netherlands Academy of Arts and Sciences (KNAW), which is gratefully acknowledged. I would like to thank Frank Veltman for discussion.

References Baltag, A. 1999. A logic of epistemic actions. Proceedings of the Workshop on Foundations and Applications of Collective Agent Based Systems, 11th European Summer School on Logic, Language and Information. Utrecht: Utrecht University. Beaver, D. 1995. Presupposition and Assertion in Dynamic Semantics. Ph.D. dissertation, CCS, Edinburgh. Donnellan, K. 1966. Reference and definite descriptions. Philosophical Review 82:3–32.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

25

Fernando, T. 2004. Are context change potentials functions? In H. Kamp and B. Partee, eds., Contexts in the Analysis of Linguistic Meaning, pp. 117–136. Stuttgart/Prague. Frank, A. 1997. Context Dependence in Modal Constructions. Ph.D. dissertation, University of Stuttgart, Stuttgart. Gazdar, G. 1979. Pragmatics: Implicature, Presupposition, and Logical Form. New York: Academic Press. Gerbrandy, J. 1999. Bisimulations on Planet Kripke. Ph.D. dissertation, University of Amsterdam, Amsterdam. Geurts, B. 1995. Presupposing. Ph.D. dissertation, University of Stuttgart, Stuttgart. Geurts, B. 1998. Presuppositions and anaphors in attitude contexts. Linguistics and Philosophy 21:545–601. Groenendijk, J. 1999. The logic of interrogation. In T. Matthews and D. L. Strolowitch, eds., Proceedings of the Ninth Conference on Semantics and Linguistic Theory, pp. 109–126. Stanford: CLC Publications. Groenendijk, J. and M. Stokhof. 1984. Studies in the Semantics of Questions and the Pragmatics of Answers. Ph.D. thesis, University of Amsterdam, Amsterdam. Groenendijk, J. and M. Stokhof. 1991. Dynamic predicate logic. Linguistics and Philosophy 14:39–100. Harel, D. 1984. Dynamic logic. In D. M. Gabbay and F. Guenthner, eds., Handbook of Philosophical Logic. vol. 2, pp. 497–604. Dordrecht: Reidel. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Heim, I. 1983. On the projection problem for presuppositions. In M. Barlow, D. P. Flickinger, and M. T. Wescoat, eds., Proceedings of the Second West Coast Conference on Formal Linguistics, pp. 114–125. Stanford (CA): Stanford Linguistics Association. Heim, I. 1992. Presupposition projection and the semantics of attitude verbs. Journal of Semantics 9:183–221. Herzberger, H. 1973. Dimensions of truth. Journal of Philosophical Logic 2:535–556. Kamp, H. 1981. A theory of truth and semantic representation. In J. Groenendijk, Th. Janssen, and M. Stokhof, eds., Formal Methods in the Study of Language, pp. 277–322. Amsterdam: Mathematical Center Tracts 136. Karttunen, L. 1974. Presuppositions and linguistic context. Theoretical Linguistics 1:181–194. Karttunen, L. and S. Peters. 1979. Conventional implicature. In C.-K. Oh and D. Dinneen, eds., Syntax and Semantics, Presupposition. vol. 11, pp. 1–56. New York: Academic Press.

26

ROBERT

VAN

ROOIJ

Kibble, R. 1994. Dynamics of epistemic modality and anaphora. In H. Bunt, R. Muskens, and G. Rentier, eds., Proceedings of the First International Workshop on Computational Semantics, pp. 121–130. Tilburg: Institute for Language Technology and AI, Tilburg University. Landman, F. 1986a. Towards a Theory of Information. Ph.D. dissertation, University of Amsterdam, Amsterdam. Landman, F. 1986b. Conflicting presuppositions and modal subordination. Papers from the 22nd Regional Meeting, pp. 195–207. Chicago Linguistic Society. Lewis, D. 1979. Attitudes de dicto and de se. Philosophical Review 88: 513–543. Peters, S. 1975. A truth-conditional formulation of Karttunen’s account of presupposition. Texas Linguistic Forum, pp. 137–149. Austin: University of Texas. Roberts, C. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy 12:683–721. Soames, S. 1979. A projection problem for speaker presuppositions. Linguistic Inquiry 10:623–666. Soames, S. 1982. How presuppositions are inherited: A solution to the projection problem. Linguistic Inquiry 13:483–545. Stalnaker, R. 1968. A theory of conditionals. In N. Rescher, ed., Studies in Logical Theory, pp. 98–112. Oxford. Stalnaker, R. 1970. Pragmatics. Synthese 22:272–289. Stalnaker, R. 1973. Presupposition. Journal of Philosophical Logic 2:447–457. Stalnaker, R. 1974. Pragmatic presupposition. In M. K. Munitz and P. K. Unger, eds., Semantics and Philosophy, pp. 197–213. New York: New York University Press. Stalnaker, R. 1975. Indicative conditionals. Philosophia 5:269–286. Stalnaker, R. 1978. Assertion. In P. Cole, ed., Syntax and Semantics, vol. 9: Pragmatics, pp. 315–332. New York: Academic Press. Stalnaker, R. 1998. On the representation of context. Journal of Logic, Language and Information 7:3–19. Stalnaker, R. 1999. Introduction. In R. Stalnaker, ed., Context and Content. Oxford: Oxford University Press. Stalnaker, R. 2002. Common ground. Linguistics and Philosophy 25:701–721. van der Sandt, R. 1982. Kontekst en Presuppositie. Ph.D. dissertation, University of Nijmegen, The Netherlands. van der Sandt, R. 1988. Context and Presupposition. London: Croom Helm. van der Sandt, R. 1991. Denial. In Chicago Linguistic Society 27(2). Papers from the Parasession on Negation, pp. 331–344.

PRESUPPOSITION: AN (UN)COMMON ATTITUDE?

27

van der Sandt, R. 1992. Presupposition projection as anaphora resolution. Journal of Semantics 9:333–377. van Rooij, R. 2001. Exhaustivity in dynamic semantics; referential and descriptive pronouns. Linguistics and Philosophy 24:621–657. Veltman, F. 1996. Defaults in update semantics. Journal of Philosophical Logic 25:221–261.

2

Pragmatic Strategies ROB

VAN DER

SANDT

Introduction In the seventies Gerald Gazdar developed a theory of formal pragmatics the core of which was his theory of presupposition projection.1 Though Gazdar’s theory covered a wide range of data and though, at its time, it was vastly superior over competing theories, its further development was hampered by some serious problems. One problem was the counterintuitive notion of presupposition which did not capture the intuitive view of presupposition as being information which is in some sense given or taken for granted in a discourse. A concomitant problem was his account of cancellation which was widely felt to be conceptually incoherent. In Gazdar’s view conversational implicatures constitute a stronger kind of information than presupposition: presuppositional information may be defeated by implicatures. This is at odds with our pretheoretical notion of presupposition and conflicts with the standard way this notion is characterized. Intuitively, if a sentence is felt to be presupposing, the presuppositional information is (in some sense to be made explicit) required for its utterance to be interpretable. The standard characterization of conversational implicatures is a very different one. On the

1 Gazdar (1979), the published version, is a revision of his 1976 thesis which was completed under the supervision of Hans Kamp.

Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

29

30

ROB

VAN DER

SANDT

Gricean view a conversational implicature is not a semantic requirement for interpretation, but instead an inference that is computed on the basis of the interpretation of a sentence, contextual information and conversational principles. The question then immediately arises how such inferences, which are for their existence dependent on a previous interpretation of the utterance that invokes them, can possibly defeat information which should already be part of the linguistic or nonlinguistic context for the inducing sentence to have an interpretation. This problem is the more pressing since the received view tells us that conversational implicatures are the paradigm cases of defeasible inferences. Thus, if there is any direct interaction between presupposition and conversational implicature, one would expect the former to override the latter and not the other way around. Gazdar’s theory has a dynamic flavour. However, his dynamics is found only at the level of sentential discourse processing and does not extend to the interpretation of subsentential constituents. Moreover, presuppositional information is only computed and incremented in the subsequent context after the propositional content has been computed and incremented. Current theories that are based on dynamic frameworks deviate in exactly this respect and give a very different picture of how presuppositions are evoked and resolved in a context. In contrast to the Gazdarian picture they agree with Frege (1892) and Strawson (1950) that presuppositions are requirements for the interpretation of an utterance, they agree with Stalnaker (1974) and Karttunen (1974) that presuppositional information is information that is in some sense contextually given or taken for granted in a conversation. They agree with Lewis (1979) that this information may not always be explicitly given in the context of utterance, but may be reconstructed, and they thus hold that the felicity of a presupposing sentence may be established by a process of accommodation. On current dynamic assumptions this means that accommodation is not simply a side product of contextual update, which only affects the output context, but should instead be reconstructed as an operation on input contexts so as to enable an utterance to be processed even if the original context didn’t admit it. And, finally, there is by now a consensus that an adequate account of presupposition and presupposition projection is to be construed as a hybrid theory in which semantic and pragmatic principles interact. Existing views mainly differ with respect to the question what exactly counts as contextual requirements for the interpretation of presupposing sentences. They differ moreover in their views on the type of mechanism and the status of the principles that govern accommodation. Heim (1983), Beaver (1995) and other satisfaction theorists basically require that the context

PRAGMATIC STRATEGIES

31

of utterance entails the presuppositional information.2 Current anaphoric accounts view presuppositions as identifiable objects that should find some other identifiable object to serve as an antecedent for the presuppositional expression.3 And the latter are closer to the theory of Gazdar in accommodating this object in case it is not already there.4 In retrospect we may point to two major characteristics of Gazdar’s system that divorce his methodology from current theories, and that prevented analysts to develop his account into a more adequate and conceptually appealing theory of presupposition. The first characteristic to blame is what I will call his preKampian dynamism, the second his very unKampian view on the division of labour between semantics and pragmatics.5 These two problems are related. Gazdar’s system is confined to presupposition computation on a propositional level. In his simple model of discourse incrementation the dynamics does not extend below the level of full sentences, which means that it cannot handle quantificational and anaphoric links beyond the sentence boundaries. As a consequence, presupposition-, content- and implicature-expressions are incremented separately as entities in their own right. Presuppositional or other pragmatic information will thus never enter into binding relations with the semantic content of a sentence. Nor is presupposition treated in the Fregean way as a definedness condition of the presupposing sentence. Instead presupposition is treated as a purely pragmatic phenomenon which does not contribute to the semantic content of an utterance and thus cannot serve as a requirement for the interpretation of the inducing sentence. The first problem, which has become known as the binding problem, can only be solved when we adopt Discourse Representation Theory (DRT) or some other theory of dynamic interpretation. I will briefly comment on this issue in the next section and refer the reader to the extensive literature on dynamic treatments of presupposition that turned up in the nineties. The second problem, the interaction between pragmatic inferencing and semantic interpretation, will be discussed 2 The satisfaction account originates in the work of Karttunen (1974) and Stalnaker (1974). See for developments in a dynamic framework in particular Heim (1983) and Beaver (1997, 2001). 3 See, for example Van der Sandt (1992), Geurts (1995, 1999), Sæbø (1996), Krahmer (1998), Asher and Lascarides (1998), Bos (1999), Kamp (2001). See Zeevat (1992) for a hybrid account which incorporates ideas from both worlds. 4 See Geurts (1996) and Beaver (1997) for two different assessments of the state of the art. 5 See Kamp (1979) for a criticism of the classic division of labour between semantics and pragmatics which antedates DRT by two years.

32

ROB

VAN DER

SANDT

when I turn to a discussion and re-evaluation of Gazdar’s system (in particular the Section on The Status of Gazdarian Clausals). I will then put forward some tentative remarks on the relation between presupposition and questions in discourse. The general setup of this paper is as follows. I will first show that even when we restrict ourselves to Gazdar’s conceptual frame, some oddities of his system can easily be remedied. This will set the stage for a reinterpretation of the major feature which governs the behaviour of presuppositions in his system: the interaction between presuppositions and his clausal implicatures. I will claim that these ‘implicatures’ should not be seen as ordinary Gricean implicatures, but should instead be seen as constraints on the incoming context. In fact they can be reinterpreted as ‘issues’ or ‘questions’ in discourse. And, more importantly, I will point out that it is exactly the possibility of such a reinterpretation which explains why his original clausals did the work they were devised for. I will finally show how such a reinterpretation, when implemented in current dynamic treatments, may give a new view on certain accommodation effects as giving rise to a special kind of bridging inference. I will illustrate this by means of a simple solution to a well-known problem involved in the analysis of disjunctive sentences.

Gazdar’s Dynamics Gazdar’s account is formulated in an incremental model of discourse processing in which each utterance changes the incoming context by adding the information conveyed to the next context.6 As I pointed out in the previous section, Gazdar’s dynamics does not extend below the sentence level and is not applicable when presuppositions and implicatures enter into binding relations with quantified expressions. Gazdar’s presuppositions and implicatures act as entities in their own right, but they may, in view of their content, constrain the incrementation process of subsequent utterances. Utterances are construed as context-sentence pairs.7 A discourse is an ordered sequence of utterances. The contextual information 6 See Gazdar (1979) for the full exposition of his theory. Many detailed expositions are found in the literature. See, for example Soames (1982), Levinson (1983), Van der Sandt (1988), Beaver (1997, 2001). 7 Actually Gazdar construes them as indexed pairs of contexts and propositions, where the context is taken to be a set of propositions. Since one would like to conceive of ‘I am hungry’ said by me to you and ‘You are hungry’ said by you to me as different utterances expressing the same proposition, I will take utterances as pairs of sentences and contexts where the context contains the parameters needed to determine the proposition

PRAGMATIC STRATEGIES

33

accumulated at a certain stage of a discourse is a set of propositions, which comprises all the information conveyed up to that point and thus includes the contribution of the implicatures and presuppositions. Such an information set is interpreted as a commitment slate in the sense of Hamblin.8 After the utterance of a sentence the speaker is thus not just committed to the truth-conditional content of his utterance, but also to all further information that has been conveyed by non-truthconditional means. The latter comprises in Gazdar’s theory not only implicatural information but also the presuppositional contribution. In order to allow for the contextual defeasibility of presuppositions and implicatures Gazdar distinguishes between potential presuppositions (pre-suppositions in Gazdar’s terminology or protopresuppositions as I will call them) and the actual presuppositions of a sentence in a discourse. Similarly there is a distinction between potential implicatures (im-plicatures or proto-implicatures) and the actual implicatures conveyed by the utterance of a sentence. Both proto-implicatures and proto-presuppositions are generated in a purely mechanical way. Proto-presuppositions are read off from presuppositional constructions such as definite NPs, factive verbs, verbs of change and other items that are standardly taken to be presupposition inducers. They are given by the lexicon and the syntax of the language under consideration. These are the entities that may in the course of the incrementation process emerge as the actual presuppositions of an utterance. Whether they do so will depend on the information contained in the context at the moment of incrementation. The actual presuppositions of an utterance will thus always be a subset of the set of proto-presuppositions. Contextual incrementation is based on a hierarchical ordering of semantic content, proto-implicatures and proto-presuppositions, and is constrained by the requirement of consistency. Given a context, the proposition expressed by a sentence in this context, and given a previous determination of the proto-implicatures and proto-presuppositions of this sentence, contextual incrementation proceeds in three stages. We first create an auxiliary context by adding the proposition expressed to the already established contextual information. This yields a context which consists of the union of the old context with the truth conditional

expressed, enriched with an information set which plays the role of Stalnaker’s context set. This has no consequences for the discussion but simplifies the formulation of Gazdar’s system. Since Gazdar’s system is limited to modal propositional logic I will often use ‘context’ to refer to this information set. 8 See Hamblin (1971).

34

ROB

VAN DER

SANDT

content of the sentence. In the next stage we add those consistent sets of proto-implicatures which retain consistency with the already established set of propositions, thus again establishing a new context which comprises both the content and the implicatures associated with this sentence. Finally we add those consistent sets of proto-presuppositions which are consistent with the set established by the incorporation of the implicatures. Only those proto-implicatures and proto-presuppositions which don’t give rise to contradictions in this staged process are thus established as the actual implicatures and presuppositions of an utterance. This ordered process eliminates inter alia all protoimplicatures which are inconsistent with the propositional content, contextual information or each other. It also eliminates all protopresuppositions which are incompatible with the set resulting from the incorporation of the implicatures or which are inconsistent with one another. Only those proto-implicatures and proto-presuppositions which are added to the next context are taken to be the actual implicatures and presuppositions of the sentence processed. Figure 1 gives a pictorial representation of the incrementation procedure as Gazdar conceives it. The following examples may illustrate the idea behind this general setup: (1) a. b.

Either John does not own a donkey or he keeps his donkey very quiet. Either John has an insulated stable, or he keeps his donkey very quiet.

The first disjunct of (1a) invokes inter alia the proto-implicature that for all the speaker knows John may not have a donkey. This protoimplicature is consistent with the content of (1a). It will thus be incremented in the next context and be established as an actual implicature of an utterance of (1a). We then consider the contribution

c0 c0

, c0

χ, c1

, c2

sem. impl. pres.

sem. impl. pres.

sem impl. pres.

c0

c0 = c1

c1

c1

c1 = c 2

ordered incrementation FIGURE 1 Discourse.

c2

c2

c2

PRAGMATIC STRATEGIES

35

of the presuppositions in the context that has thus been established by the incorporation of the implicature. The second disjunct induces the proto-presupposition that John has a donkey. The assumption that John has a donkey conflicts with the information that has been established by adding the implicature that he may not have a donkey and is thus defeated.9 The proto-presupposition will thus not be inherited which correctly predicts that an utterance (1a) does not presuppose that John owns a donkey. In (1b) we observe a different situation. Due to the fact that the first disjunct of (1b) is logically independent of the proto-presupposition no conflicting implicature will be invoked in this case (the proto-implicatures that for all the speaker knows John’s stable may or may not be insulated, are compatible with the assumption that he has a donkey). Hence the proto-presupposition will be inherited by the next context and thus be established as an actual presupposition of the utterance. Gazdar’s system captures the fact that survival of pragmatic information depends on the context of utterance. It, moreover, correctly construes presupposition not as a property of the inducing sentence, but of its utterance in a context. However, in line with nearly all other theorists at his time he assumes semantic and pragmatic information are two different types of content that are computed and represented separately and will only be merged afterwards in a more substantial proposition.10 On this view it is this more informative propositional object which encodes the full message conveyed. Given a pragmatic view of presupposition this implies that we should be able to determine the proposition expressed by a sentence independently of the contribution of the presuppositional information. This has some unfortunate consequences. It yields a counterintuitive notion of propositional content and, even worse, gives rise to binding problems when presuppositions and implicatures entertain anaphoric relations with external antecedents. To illustrate the point let us have a look at (2): (2) The mayor opened the exhibition. 9 ‘Conflicts’ in an intuitive sense. Logically there is no contradiction. See the Section on Epistemic Modification. 10 Karttunen and Peters (1979) (though not Karttunen, 1974) is a particularly striking example. It should be mentioned that Stalnaker’s work (1974, 1978), is an exception, though the Stalnaker inspired theory of Van der Sandt (1988) and the general criticism of the standard view in Van der Sandt (1992) may wrongly suggest the opposite.

36

ROB

VAN DER

SANDT

Splitting up the presuppositional and content-contribution we obtain (3a) and (3b) for the presuppositional information and the content respectively: (3) a. b.

(x’y[mayor(x)2x ¼ y] (x’y[[mayor(x)2x ¼ y]Xopen_exhibition(x)]

that is, we are forced to encode the information that there is a mayor twice, once as part of the content expression and once as part of the presuppositional contribution. This captures the observation that presuppositional inferences of non-embedded sentences are preserved as an entailment of their inducing sentence. It also captures the fact that this information contributes to the proposition expressed when it is part of a complex structure. But note that the reason is not that ‘the presupposition’ is entailed by the subsentence that invokes it, as some pragmatic analysts have claimed. It only happens because the presuppositional information is encoded twice, once as part of the content expression and once as a defeasible proto-presupposition. It is the non-presuppositional content part that is preserved as a non-defeasible entailment. It is thus misleading to say that presuppositions cannot be defeated in entailing environments. For even if the proto-presupposition is defeated or neutralized, its non-presuppositional counterpart in the content expression will still do its semantic duty. There is thus no principled reason why Gazdar’s system cannot be redefined in such a way that the presupposition is defeated in such environments though the inference to the purportedly presuppositional information goes through. The redundant encoding is nevertheless wrong for both technical and empirical reasons. Firstly, it does not capture the fact that the nonpresuppositional information conveyed by (2) is not that there is a mayor who opened the exhibition, but that its utterance contributes to a context which should already provide a mayor and conveys there that he opened the exhibition. This means that the non-presuppositional contribution should be given by an open sentence that has to link up to the established presuppositional information. The procedure thus assumes a counterintuitive notion of propositional content. It moreover yields the wrong truth-conditions when we consider non-extensional embeddings. Consider (4a) and note that its utterance would normally presuppose that there is a mayor, i.e. (4b) and convey (4c), that is the scalar implicature that it is not necessary that he opened the exhibition: (4) a. It is possible that the mayor opened the exhibition. b. There is a mayor. c. It is not necessary that the mayor opened the exhibition.

PRAGMATIC STRATEGIES

37

SEMANTIC CONTENT ◊ open_exhibition (x) ¬ open_exhibition (x) ∃x mayor (x) (assertoric contribution (implicatural info (presupposed info dependent on the computed on the basis contextually required) presuppositional of contextual info requirements) + content) ACCOMMODATION INCREMENTATION i.e. adjustment of the i.e. mapping of the adjusted context of utterance context in to the next one FIGURE 2 Types of information.

When we encode the contribution of the presuppositional, content and implicature expressions separately, sentence (4a) is predicted to convey that there is a (unique) mayor in this world, that there is a (possibly) different one in the world where the exhibition was opened, and furthermore, by way of scalar implicature, that it does not hold for every world that there is a mayor who opened the exhibition. This is wrong and typical for all other theories that adhere to a strict division of labour between the semantic and pragmatic component.11 A proper analysis should rather run along the lines of (5) and be analysed as sketched in Figure 2: (5) a. There is a mayor . . . it is possible . . . but not necessary that he opened the exhibition b. (x mayor(x) . . . B open_exhibition(x) . . . :& open_exhibition(x) . . . We thus perceive an anaphoric process where the presuppositional expression should bind a variable in the (open) content expression and where the implicature is in turn anaphorically dependent on both the presuppositional and the content expression. The content thus depends on the presuppositional expression and takes priority over the implicature. This picture effectively reverses the order postulated by Gazdar.12 Gazdar’s theory is based on a standard static semantics. A solution of the binding problem had to wait until Kamp (1981) and 11

This also includes Van der Sandt’s (1988), though not his (1992) account. See Geurts and Maier (2003: ms.) for an extension of DRT which allows us to represent and interpret various kinds of semantic and pragmatic information in a unified framework. 12

38

ROB

VAN DER

SANDT

Heim (1982) presented their versions of dynamic semantics, recaptured meanings as relations between input and output contexts and showed how to treat anaphoricity inter- and intrasententially in a uniform way. Various proposals are found in the presupposition literature since then. I contend that the same development allows us to remove a second obstacle from Gazdar’s theory, the counterintuitive and conceptually incoherent thesis that the contribution of implicatures takes priority over the computation of presuppositions. But in order to do so I first have to discuss another feature of his theory that has been objected to by a variety of authors, the intuitively unreasonably strong epistemic modification of presuppositional, content and implicature expressions.

Epistemic Modification In the previous section I gave an informal sketch of Gazdar’s method of contextual incrementation as a kind of default inheritance mechanism. The basic constraint on update is consistency. Only consistent subsets of information are inherited by the next context. However, in many cases intuitively conflicting information does not formally give rise to inconsistency between different types of pragmatic and semantic information. In order to obtain inconsistency Gazdar takes recourse to a strong modal modification of the relevant expressions. The two modal operators figuring in Gazdar’s system are the K- and the P-operator of Hintikka’s (1962) epistemic logic. Ka j is to be read as ‘a knows that j’ and Pj as ‘it is possible, for all that a knows that j’. These are the epistemic variants of the well-known necessity and possibility operators of standard modal logic and are thus interdefinable in one another: Pa j :¼ :Ka :j Gazdar limits himself to monologue. Hence he only needs indices for one speaker which may thus safely be skipped. And though his proofs are cast in terms of Hintikka’s epistemic logic, the modalities may be interpreted by normal Kripke semantics. The K- and P-operators may be interpreted as the box and the diamond respectively. If we assume reflexivity and transitivity for the accessibility relation the system semantically corresponds to KT4. A central property of knowledge is that anything that is known by someone is true. Reflexivity guarantees the validity of the scheme Kj-j and thus that knowledge entails truth. A second property which

PRAGMATIC STRATEGIES

39

will turn out to be crucial is positive introspection: if one knows something one knows that one knows that. Transitivity guarantees just that by validating the scheme Kj-KKj. We may add two further operators, B and C for belief. These ‘doxastic’ operators are interdefinable just as their epistemic counterparts. Bj is to be read as the speaker believes that j. And Cj means that it is compatible with the speaker’s beliefs that j. This weaker notion should not obey Bj-j as there are false beliefs. It should however respect Bj-Cj (beliefs are consistent). We obtain this by skipping reflexivity and requiring just transitivity and seriality for belief. According to Gazdar the assertoric utterance of a sentence j commits the speaker to knowing that j under Grice’s maxim of Quality. It is this commitment which gives rise to a so-called quality implicature of the form Kj. Thus by the utterance of (6) The cat is on the mat. the speaker does not just commit himself to the cat being on the mat, but to knowing that this is the case. This is too strong. Surely by an assertoric utterance of j, a speaker commits himself to its truth, cannot admit disbelief in his statement and, when asked for further justification, he may back up his claim by saying that he knows (or believes) it is true. That is, by the assertoric use of j a speaker is committed to act as if j is true, but by his utterance he is neither saying nor implicating that he knows (or believes) that j. His commitment is a consequence of uttering this sentence with assertoric force, but it is not part of the presuppositional, propositional or implicatural information and should be handled at a meta level. Speaker commitment is already captured by Gazdar by interpreting a context as a commitment slate and thus is already taken care of and handled as a fact about the context. Hence the epistemic commitment of the speaker to the truth of what he says need not be encoded by adding knowledge claims to the information set that is taken to be his set of commitments. Note that Grice already objected to construing epistemic commitment as some kind of quality implicature. On my account, it will not be true that when I say that p, I conversationally implicate that I believe that p; for to suppose that I believe that p [ . . . ] is just to suppose that I am observing the first maxim of Quality [ . . . ] it is not a natural use of language to describe one who has said that p as having, for example, ‘implied,’ ‘indicated,’ or ‘suggested’ that he believes that p. Grice (1989: 42)

40

ROB

VAN DER

SANDT

Proto-presuppositions are likewise modalized with the K-operator. The proto-presupposition Gazdar takes to be induced by (2) is not that there is a mayor but the stronger expression that the speaker knows that there is a mayor. A typical utterance of (2) will thus presuppose: (7) K[there is a mayor] The remarks just made with respect to the epistemic modification of assertoric utterances apply here as well. When a speaker indicates that he presupposes that there is a mayor, he commits himself to the truth of what he presupposes, but just as in the case of plain assertions this is accounted for by interpreting a context as a commitment slate. There is thus no need to enter the description of a consequence of this commitment to the context. Gazdarian implicatures derive from the maxim of Quantity and come in two varieties: clausals and scalars. Clausal implicatures are read off from the clauses of compound sentences. Given a compound sentence j the mechanism yields for each non-entailed and nonpresuppositionally induced subsentence w a set of proto-implicatures Pw and P:w. For a disjunction of the form w3c this yields {Pw, P:w, Pc, P:c}. The proto-implicatures of (1a) thus comprise P[John has a donkey] and P:[John has a donkey]. Hence, by the utterance of this sentence the speaker will typically implicate that for all he knows it is possible that John has a donkey and that for all he knows it is possible that John does not have a donkey, which might also be paraphrased more concisely as the speaker does not know whether John has a donkey. The informal explanation is that by the utterance of a disjunction the speaker implicates that he does not take the truth of either disjunct for granted. For if he had known that either disjunct were true, he would have been in the position to make a much stronger statement. If he had known that John didn’t have a donkey, he should have said so directly, thereby producing a more informative statement. If he had known that John had one, the plain assertion of the second disjunct would have been shorter and more informative than the full disjunction. In both cases he would have been required to choose the alternative expression by the maxim of Quantity. The implicature for the second disjunct is derived analogously. The hearer will thus conclude that since the speaker asserted only the weaker statement, he was not taking the truth of either disjunct for granted and thus conversationally implicated that he didn’t know whether John had a donkey or whether he kept his animal quiet. Clausals thus capture the

PRAGMATIC STRATEGIES

41

suggestions of epistemic uncertainty that typically go with the utterance of disjunctions, conditionals and epistemic possibility. I will argue in the next section that this inference should not be seen as a standard Gricean implicature and, moreover, that if it is properly construed, we don’t need epistemic qualification in the Gazdarian way.13 The second variety of implicatures found in Gazdar are scalars. These are derived from quantitative scales in the sense of Horn (1972). Such scales are ordered n-tuples of expression alternatives the form ha1 ; a2 ; . . . ; an i, where each ai is a linguistic expression of a similar type. Given such a scale a sentence j containing a weaker expression will invoke the implicature that the corresponding inference containing a stronger expression from the same scale does not hold. So given the scale /all, most, many, some, . . . S the utterance of (8a) will implicate that not all of the children were in the garden. Such scalars normally come unmodalized. In Gazdar’s system they come modalized in a very strong way. The proto-implicature assigned to (8a) is (8c). By a typical utterance of (8a) the speaker is thus not said to implicate that not all of the children were in the garden, but that he knows that not all of the children were in the garden. (8) a. Some of the children were in the garden. b. All of the children were in the garden. c. K:[all of the children were in the garden] d. :K[all of the children were in the garden] e. :[all of the children were in the garden] This modification has generally be felt to be unreasonably strong and has been objected to by a number of authors. Soames (1982), Van der Sandt (1988) and Horn (1989) all suggest that, if epistemic force is 13

It should be remarked though that Gazdar’s definition overgenerates. This holds in particular for those implications that are read off from the complements of verbs of propositional attitude. For example, Gazdar’s definition associates with (i) the set of proto-clausals (ii): (i) John hopes that he has won the prize. (ii) {P[John has won the prize], {P:[John has won the prize]} These proto-clausals will survive a typical utterance of (i) unless the context provides information to the contrary. However, intuitively these environments do not give rise to the inference that the speaker does not know whether he won the prize. This is easily remedied by excluding verbs of propositional attitude from the contexts that generate such clausals.

42

ROB

VAN DER

SANDT

assigned, it should rather look like :Kj instead of K:j. For even if we grant that a co-operative speaker will not utter (8a) when he knows that (8b) is true, it does not follow that (8a) can only be uttered cooperatively when the speaker knows that (8b) is false. It is perfectly acceptable for the speaker to utter (8a), when he does not have information about the truth-value of (8b). Under normal circumstances a hearer will conclude from an utterance of (8a) that the speaker does not know whether all children are in the garden – not that the speaker knows that not all children were in the garden. The hearer will draw the latter conclusion only if he assumes that the speaker has complete information about the situation at hand. If this does not obtain, (8c) is too strong. The inference from (8d) to (8c) thus comes about as a result of the extra assumption that the speaker is well informed about the truth-value of the corresponding statements which contain a higher ranking expression from the relevant scale. Hirschberg (1991: 75–79) and Levinson (2000: 77–79) suggest a weaker epistemic modification in terms of the belief thus distinguishing between the epistemic commitments that come about as a result from the speaker’s commitments in view of what is asserted versus the commitments that arise from what he is merely implicating. In my view this strategy is misguided and based on a confusion between different levels of interpretation. Here I concur with Atlas as reported in Levinson (2000: 78). Just as in the case of presuppositional and assertoric information, the epistemic state of the agent should not be encoded as part of the content of his utterance or of whatever is conveyed by his utterance. Standard Gricean reasoning supports this view. When a speaker utters (8a) although he is in a position to utter the logically stronger (8b), he violates the maxim of Quantity. The fact that a speaker utters (8a) thus suggests – in light of the maxim of Quality ‘Say nothing for which you don’t have adequate evidence’ – that the speaker does not have complete information about the number of children being in the garden, and therefore avoids saying more than he knows. Only if the hearer knows that the speaker has full information in this respect, can he conclude that the speaker has chosen a logically weaker expression in order to avoid a conflict with the submaxim ‘Don’t say what you believe to be false’. Thus it is only under those circumstances that he can infer (8c). The inference on the part of the hearer that the speaker does not know, believe or whatever, comes about as a result of a reasoning process which involves inferences about the knowledge, beliefs and intentions of the speaker. These inferences as to the epistemic status of the speaker should however not be made part of the content of what he conveys.

PRAGMATIC STRATEGIES

43

Eliminating Epistemic Modifiers In the previous section I argued that the encoding of various degrees of epistemic commitment on the part of the speaker in the presuppositional, content and implicature expressions is based on a confusion between what features of interpretation should be handled at the level of speech act theory, what should be handled by the dynamics of discourse interpretation and what at the level of content. And even when interpreted at a meta level the epistemic force of knowing seems too strong. Nor does such a modification capture the differences between presuppositional, assertoric and various kinds of implicatural information. When we have a closer look at the system, we see that the derivation of strong presuppositions, strongly modalized assertions and implicatures is an artefact of the system which for its empirical predictions depends crucially on two factors: (i) the hierarchical ordering of semantic content, proto-implicatures and proto-presuppositions and (ii) the requirement of consistency for the contextual update. Let us run through the example given before: (9) a.

Either John does not own a donkey or he keeps his donkey very quiet. b. K[John has a donkey] c. P[John has a donkey], P:[John has a donkey]

The occurrence of John’s donkey in the second disjunct of (9a) induces the proto-presupposition (9b). And, since the first disjunct is neither entailed nor (proto-)presupposed, this clause generates the proto-implicatures given in (9c). Contextual incrementation now proceeds as follows. Assuming that the initial context is empty, K(9a) will be entered into an auxiliary context, thus establishing the input context for the incrementation of the implicatural and presuppositional information. Since the proto-clausals are consistent with the context just created, this context will be updated by this information thus establishing the proto-clausals as actual clausals of the utterance. The resulting context is the input for the proto-presupposition. The latter won’t survive the incrementation procedure though, since the context established by now comprises P:[John has a donkey] which is inconsistent with K[John has a donkey]. The context resulting from this ordered incrementation procedure will thus comprise the (modalized) content of the sentence, the clausals, but not the protopresupposition. The presupposition is said to be cancelled. Note that this cancellation process depends both on the epistemic modification of

44

ROB

VAN DER

SANDT

the presuppositional material and on the order of incrementation. For, if the proto-presupposition were not modalized, it would wrongly survive, since the set {P:[John has a donkey], [John has a donkey]} is consistent. If we would retain the modalization of the proto-presupposition but would switch the ordering around, i.e. if we would first increment the proto-presupposition and then the implicatural information, the proto-presupposition would equally survive, but now cancel the proto-implicature since the set {P:[John has a donkey], K[John has a donkey]} is inconsistent. This would wrongly predict that (9a) presupposes (9b) and does not give rise to the inference that the speaker does not know whether John has a donkey. The general setup of the system thus requires both a modalizing of the proto-presupposition and the stipulation that implicatural information is stronger than presuppositional information. I first give a solution to the problem of strong presuppositions and assertions and discuss the ordering of presuppositional and implicatural information in the next section. A simple solution to the first problem is obtained by simultaneously strengthening the consistency requirement for contextual update to the requirement of doxastic defensibility, and eliminating the epistemic modalization of proto-presupposition and assertoric update. For a similar account I refer to Beaver (2001: 65–67).14 I noted above that the epistemic commitments need not be encoded as beliefs or knowledge of the speaker. Commitments are social consequences that derive from the utterances a speaker actually produces. They may be reinforced, withdrawn, a speaker may stick to his commitments or violate them. It is the utterances a speaker produces that give rise to his commitments and this may be accounted for by entering the content of his utterances in a set which we then interpret as his commitment slate. Such a commitment slate can best be seen as an ‘edited summary of a dialogue’s preceding history’ (Hamblin, 1987: 229) and is close to Lewis’ (1979) notion of a conversational scoreboard. The fact that the information conveyed is part of such a commitment slate will subsequently put the speaker under a variety of obligations many of which are dictated by the Gricean maxims. He should be prepared to defend his claims, not redundantly repeat information etc. The basic point is that the fact that this information is 14 Beaver (2001) proposes to change the definition of consistency to ConsðXÞ iff fKj j j 2 Xg h >. This comes down to requiring epistemic defensibility and yields the same results as presented here. It is shown below that we may dispense with the K-operator altogether since doxastic defensibility already captures Gazdar’s data.

PRAGMATIC STRATEGIES

45

part of his commitment slate will put him under certain obligations. His commitment slate thus should represent the content of what he is committed to, but not contain the description of his commitments. This is roughly the way in which Hamblin introduced the notion. It follows that the content of what a speaker conveys by making assertoric utterances will be entered as a part of the context, but that the effect of his being committed to knowing or believing what he says should be handled on a different level. Such an interpretation is in line with the framework for speech act theory Gazdar proposes elsewhere (Gazdar, 1981). Here speech acts are essentially viewed as instructions to change the context. An assertion, for example, will change a context in which the speaker is not committed to the truth of j to one in which he is so committed. This is close to Stalnaker (1978) where we also perceive a principled division between the content of what is conveyed and the various effects this gives rise to. Gazdar (1979: 47) motivates the epistemic modalization of assertoric utterances by an appeal to Hintikka’s discussion of Moore’s paradox. Moore noted the oddity of the following statement. (10) The cat is on the mat, but I don’t believe it. Clearly, what is expressed by this sentence is consistent. A speaker may say that the cat is on the mat without believing it. However, on no occasion of use will his utterance of (10) be felicitous. Hintikka explains the oddity by pointing out that a speaker is expected to believe or at least to be in a position to conceivably believe what he says. This presumption is violated in (10). In Gricean terms: by an utterance of (10) the speaker infringes the maxim of Quality. Hintikka’s explanation runs as follows. Though (11a) is consistent (11b) can be shown to be inconsistent, which means that (11a) is necessarily unbelievable. (11) a. b.

j4:Bj B[j 4:Bj]

The central notion is doxastic defensibility. Doxastic defensibility does not require consistency tout court, but it requires that it is consistently possible for a speaker to believe a set of sentences. Doxastic defensibility. A set of sentences fj1 ; j2 . . . jn g is doxastically defensible just in case B[j1 4 j2 4 . . . jn] is consistent.

46

ROB

VAN DER

SANDT

Replacing the B- by the K-operator we get the corresponding notion of epistemic defensibility. We may further define the notion of doxastic implication in terms of doxastic indefensibility. Doxastic implication. j doxastically implies c just in case the set {j, :Bc} is doxastically indefensible. Given the definition of defensibility it can easily be shown that {j, :Bj} or the corresponding {j, C:j} are doxastically indefensible. The minimum requirements for the accessibility relation are seriality and transitivity. In order to interpret the knowledge-operator we add reflexivity. Consider a serial and transitive standard model M ¼ hW ; Rd ; Re ; V i, where Rd ; Re  W  W . Rd is serial and transitive, Re reflexive, serial and transitive and Re  Rd (knowledge implies belief). Assume for some arbitrary w in W, M w B½j ^ C:j. Since the necessity operator distributes over the conjunction this means (i) M w Bj and (ii) M w BC:j. Rd is serial, so according to (i) there will be a w 0 ðwRw 0 Þ such that M w 0 j and according to (ii) M w 0 C:j. The latter means that there will be a w 00 ðw 0 Rw 00 Þ such that M w 00 :j. But R is transitive. Thus wRw 00 , and according to (i) we also have M w00 j. Thus M w00 ?. Thus for no w; M : M w B½j ^ C:j. Hence {j, C:j} is doxastically indefensible (Figure 3). Since Re  Rd we see that K½j ^ P:j is also inconsistent. We thus have the same result for the knowledge operator. The set fj; C:jg is thus doxastically and fj; P:jg epistemically indefensible. We also have that the sets fj; C:cg and fj; P:cg are respectively doxastically and epistemically indefensible whenever j c. For suppose that fj; C:cg were defensible in this situation. This would mean that for some w; M : M w B½j ^ C:c, but since j c we would also have M w B½c ^ C:c which as we just saw cannot obtain.

B, BC¬

∀

C¬ 

∃

FIGURE 3 Doxastic indefensibility.

,¬ 

PRAGMATIC STRATEGIES

47

This is all we need to capture the cancellation predictions of Gazdar when stripping the K- and P-operators from the presuppositional, content and implicature expression. For the moment I will retain the epistemic modification of Gazdar’s clausals. I will discuss their status in the next section. But I will leave out the modal operators from the presuppositional expressions and the scalars. Nor will I encode Gazdar’s Quality implicature as part of the assertoric update. By the utterance of j the context will thus not be updated with the information that the speaker knows j, but just with the content-expression. I will, however, strengthen the requirement for contextual update. Instead of requiring simply consistency I will require that contextual incrementation does not give rise to indefensibility of the resulting context. This is in line with the previous discussion. The context should not contain a description of the speaker’s commitments, but simply the content of what he is committed to, and the latter will only consist of information which he can possibly believe or know. It turns out that redefined along these lines the resulting system yields the same predictions as Gazdar’s original formulation with respect to the behaviour of presuppositions and implicatures. It will turn out that this has some additional benefits. I will first run through the main cases where the interaction between contextual, implicatural and presuppositional information yields the empirical predictions Gazdar wants to account for and conclude with a short summary of definitions. Throughout the revised version I will use doxastic operators for the encoding of protoimplicatures, proto-presuppositions and the contextual update. Whatever holds for them will apply a fortiori to the stronger epistemic variants. (i) Context and content defeat clausal and presuppositional information Consider first the effect of prefixing the K-operator to the assertoric update. A typical example of cancellation would occur in a modus ponens argument. Suppose we enter Kp to the context. A subsequent utterance of p-q would add K (p-q) which in this context entails Kq. The implication will moreover invoke the following set of proto-implicatures fPp; P:p; Pq; P:qg. Since both fKp; P:pg and fKq; P:qgare inconsistent, the proto-implicatures P:p and P:q will be cancelled. Note that we get the same result if we skip the K-operator and require defensibility for the result of the contextual update. The sets fp; C:pg and fq; C:qg are indefensible and the relevant clausals will

48

ROB

VAN DER

SANDT

equally get cancelled.15 Cancellation of presuppositional material by content typically occurs when the (unmodalized) content conflicts with the presuppositional expression. So here Gazdar’s system gives the desired results without any epistemic modification. Since inconsistent sets of formulae are a fortiori indefensible, the original predictions are preserved. (ii) The interaction between clausals and scalars Scalars may be ‘suspended’ to use Horn’s terminology.16 An example is given in (12a). Here the consequent is said to invoke the scalar that not all the boys were at the party. Since no utterance of the full sentence will invoke this scalar it is cancelled in Gazdar’s terminology.17 Gazdar achieves this effect by giving clausals priority over scalars and modalizing scalars in the strong way discussed above. In (12a) the antecedent invokes the clausal P::(12b) which yields a conflict with the proto-scalar K:(12b). Again we may leave out the modalization when requiring defensibility for the update in view of the indefensibility of {:(12b), C::(12b)} (12) a. b.

Some, if not all of the boys were at the party. All the boys were at the party.

(iii) Clausals defeat presuppositions The central predictions in Gazdar arise from the interaction between clausals and presuppositional expressions. We saw that K[John has a donkey], the proto-presupposition of (9a), was defeated by the scalar P:[John has a donkey]. It holds generally in Gazdar that a scalar P:c defeats a proto-presupposition Kj whenever j c. Eliminating the K-operator from the presupposition expression and 15 Note, by the way, that the positive clausals will pass Gazdar’s procedure unharmed. The resulting context will thus contain implicatures that are simultaneously entailed by the sentence that gave rise to them. Though this is logically not a problem, it is strange and conceptually at odds with the notion of implicature. 16 Note that the claim that subclauses give rise to the same scalars as their unembedded variants seems to give rise to a projection problem for implicatures. This is wrong. Implicatures are by definition not conventionally attached to clauses but are computed on the basis of the content of a full utterance. This simply means that for implicatures there is no projection problem in the standard sense of the word. See Geurts (1999: 19–23) for an extensive discussion. I will ignore this point here and adopt Horn’s and Gazdar’s assumptions for the moment. 17 It is assumed that the whole sentence has the form of a conditional.

PRAGMATIC STRATEGIES

49

requiring doxastic defensibility for the contextual update yields the same result. For we just saw that {j, C:c} is doxastically indefensible whenever j c. (iv) Inconsistent presuppositions defeat each other Since inconsistent presuppositional expressions that come unmodalized already conflict there is no need for further modalization. It is a simple exercise to check that the arguments can be repeated in cases where inconsistency in Gazdar is achieved by conjoining contextually established information with the content expression. It thus turns out that we may safely do without the epistemic modification in content, presuppositional and scalar expressions, provided that we simultaneously strengthen the consistency requirement to defensibility. The resulting system will have the same predictive power. This does not require epistemic but just doxastic defensibility. We thus may weaken the knowledge requirement to belief. The following definitions assume an independent definition of the set of proto-presuppositions (PP(j)) and and proto-implicatures (PI(j)). They are close to Beaver’s (1997) reconstruction.18 In line with the previous discussion I assume that proto-presuppositions and 18 See Beaver (1997: 962–963). Beaver’s definitions are closer to Gazdar in retaining the epistemic modification of the content and presuppositional expressions. This could, however, easily be remedied by adopting Beaver’s (2001) suggestion (see footnote 14) and incorporating epistemic modification in the definition of consistency. On the other hand, Beaver deviates from Gazdar and the definitions given here in defining the presuppositions of an utterance j as those proto-presuppositions that are in the final context but not entailed by j. This has the advantage that an utterance of (i) does not presuppose (iii), but has the disadvantage that the same holds for non-compound sentences like (ii), since these are supposed to entail the presuppositional information.

(i) (ii) (iii)

John owns a donkey and keeps his donkey in hiding. John keeps his donkey in hiding. John owns a donkey.

Beaver (2001: 59–61) follows Gazdar in defining the presuppositions of an utterance as those proto-presuppositions that make it to the output context, which gives exactly the opposite results. One would, however, want to retain the presupposition in (ii) but not in (i). A provisional solution is to adopt a two-dimensional approach (Karttunen and Peters, 1979; Lerner and Zimmermann, 1981) and to distinguish between (a) the presuppositional information that is invoked in view of the presuppositional expression (Gazdar’s protopresupposition), (b) its counterpart in the content expression, and then take care that the former does not survive in (i). In the next section I will suggest a solution along these lines in a revised Gazdarian model (limited to a propositional language and neglecting the binding problem).

50

ROB

VAN DER

SANDT

proto-scalars come unmodalized, and that the modality involved in the contextual update and in the encoding of the clausals is serial and transitive. I leave out the ordering between the clausals and scalars. Defensible Incrementation (i) Defensibility X is doxastically defensible (Defens(X)) iff fBj j j 2 Xg h > (ii) The defensible incrementation of X with Y (X [mmY) X [ "" Y :¼ X [ fj 2 Y j 8Z  Y ðDefensðX [ ZÞ ! DefensðX [ Z [ fjgÞÞ Contextual update of information (i) c þ j ¼ c [ fjg (ii) c þþ j ¼ ðc þ jÞ [ "" PI ðjÞ (iii) c þþþ j ¼ ðc þþ jÞ [ "" PPðjÞ

propositional content implicature presupposition

Implicature and presupposition Implicature (i) j implicates c in c iff a. cAPI(j) b. c 2 c þþþ fjg Presupposition (ii) j presupposes c in c iff a. cAPP(j) b. c 2 c þþþ fjg Before I discuss the status of Gazdarian clausals two points should be made. The first concerns the status of Gazdarian presuppositions, the second the status of implicatural information. As to the first point, the definitions given above allow us to isolate the presuppositional information that is conveyed by an utterance but that is not already part of the incoming context. That is, in a more current terminology, the material that is (to be) accommodated. Let me stress again that this does not capture the intuitive notion of presupposition as information that is required to assign an interpretation to a sentence. As to the second point, one would like to have a principled distinction between presupposed and implicated material. Though presupposed material will normally be part of the incoming context, implicated material won’t. It clearly makes no sense to convey information by way of implicature if this information is already taken for granted by the participants. However, as the system

PRAGMATIC STRATEGIES

51

stands, neither of these distinctions can be made. The new information conveyed by j in c is (c þþþ {j}) \ c. But the presuppositions of j in c are simply the proto-presuppositions of j that are part of c þþþ {j}, and the implicatures of j in c are those proto-implicatures of j that satisfy the very same condition.

The Status of Gazdarian Clausals In the previous section I noted that the strong epistemic modalization of the different pieces of informational content in Gazdar’s original system is an artefact which is dictated by the requirement of inconsistency between content, implicature and presuppositional expressions in order to account for cancellation effects. I showed that the same effect can be achieved by eliminating epistemic modification and simultaneously strengthening the requirement on defeasible incrementation from mere consistency to doxastic defensibility. The reconstruction given above showed that this allows us to dispense with all epistemic modification but for the clausals. In this section I will discuss the status of Gazdar’s clausals, the counterintuitive stipulation of priority of clausal implicatures over presuppositionally induced information, and I will suggest a reinterpretation which does not conceive of clausals as conversational implicatures. Gazdarian clausals are invoked by embedded sentences that are not entailed by their matrix. They thus do the work they are supposed to do in cases like (1a), since in here the clausal that he may not have a donkey conflicts with the presuppositional expression. However, in a paradigmatic case like (13a) no clausals are invoked and the presuppositional expression thus passes unharmed. (13a) thus comes out on a par with (13b). (13) a. b.

John owns a donkey and keeps his donkey in hiding. John keeps his donkey in hiding.

However, by an utterance of (13b) a speaker will typically presuppose that John owns a donkey. When uttering (13a) he never will. A Stalnaker type explanation is straightforward. When uttering the first conjunct of (13a) we indicate that we don’t take the information that John owns a donkey for granted. When uttering the second conjunct we do. At the moment we interpret the first conjunct, the context changes since we add asserted information to the conversational background. By its very utterance its informational status thus changes. It counts as presuppositional from now on and we

52

ROB

VAN DER

SANDT

may rely on the information just introduced in a presuppositional way. The crucial point is that given the assumption that presuppositional information should ideally be part of the incoming context, (13a) can never be interpreted as being presupposing as this would defeat the point of the assertion of the first conjunct. Thinking along Gazdarian lines we thus would like to have a means to defeat the presupposition in (13a) but not (13b). Note that clausals as conceived by Gazdar will not do the job. For firstly, the first conjunct of (13a) is entailed by the full sentence which means that no (proto-)clausal will be invoked and the incrementation procedure will thus leave presuppositional expression induced by the second conjunct of (13a) unharmed. And note, secondly, that given the interpretation of a context as a commitment slate there is no way to redefine the system so as to invoke such clausals after all. A Gazdarian clausal carries the information that both the relevant clause and its negation are compatible with the knowledge of the speaker. This would clearly be an unfortunate condition to impose on assertoric acts, and it would moreover violate Grice’s Quality maxim. Instead we would like to say that an assertion is only felicitous in a context which is compatible with the truth and the falsity of the relevant clause. But in order to say this we have to abandon the interpretation of a context as the commitment slate of a speaker. If we view a context not in the Hamblin/Gazdar way as the information the speaker is committed to, but as a common ground in the Stalnaker/ Karttunen sense, we can get the desired effect by adopting Stalnaker’s principle that an utterance must be compatible with, but not entailed by, the incoming context.19 If we furthermore associate with each embedded clause a local context in the sense of Karttunen and apply the principle with respect to the local context of each subclause of a compound sentence (thus requiring that each subclause should be compatible with but not entailed by its local context20) we simultaneously capture the purported cancellation effects Gazdar captures by invoking clausals in non-entailing environments. Note that such a reinterpretation allows us to dispense with the epistemic modification of the purported clausals.21

19

See Stalnaker (1978). Though the terminology might suggest a non-trivial similarity, this is in fact very different from Karttunen (1974) and other satisfaction theorists, who require that the presuppositional expressions of each subclause are entailed by their local context. 21 This is the strategy followed in Van der Sandt (1988) who presents a projection mechanism that is inspired by Stalnaker. The same principles figure in a modified way in his (1992) anaphoric DRT-based account. 20

PRAGMATIC STRATEGIES

53

Clausals are not implicatures To understand why the ordering as stipulated by Gazdar does the work it is supposed to do, we have to see that Gazdarian clausals should not be construed as implicatures. Clausals differ in status from standard conversational implicatures in the Gricean sense. Conversational implicatures are typically computed on the basis of the incoming context, the content of the sentence to be processed and Gricean principles. They are typically incremented in the output context. Gazdarian clausals, I claim, differ in exactly this respect. They derive from pragmatic constraints on contexts, that is, constraints on input contexts. To illustrate this point consider the difference between the implicatures invoked by sentences like (8a) and the suggestion of uncertainty that goes with disjunctions, scalars and other types of implicatures that come with an utterance. They are, to use Levinson’s terminology, ‘essentially gratuitous’.22 In this respect they differ both from presuppositional inferences and clausals. The latter come about as inferences that result from contextual requirements, that is requirements the input context has to fulfil to accept an utterance as felicitous. The difference in cancellation behaviour illustrates this. In view of their very nature scalars may be easily defeated by conflicting or unsupportive assumptions as the following example from Levinson (2000) illustrates: (14) A. B.

Is there any evidence against them? Some of their identity documents are forgeries.

Here I would rather say that the purported implicature that not all of their documents are forgeries is not even invoked. The suggestion of epistemic uncertainty that is typical for clausals is not so easily defeated, however. In fact, normal assertoric uses of sentences invoking clausals are never felicitous when the common ground conflicts with members of the clausal set. Consider the disjunction (1a) and suppose that the truth of the first disjunct were taken for granted. This would not defeat the suggestion of uncertainty but instead result in sheer uninterpretability. The underlying motivation is Gricean. For in this situation the full disjunction would convey the same information as the mere utterance of the second disjunct would have done. And the utterance of just the second disjunct is thus clearly a more efficient way to convey the desired information. The only examples of purported 22

See the discussion in Levinson (2000: 60–63).

54

ROB

VAN DER

SANDT

cancellation of clausals I am aware of are examples of non-information oriented discourse, that is types of discourse in which the assumption that the conversational maxims are in force, is lifted. This happens, for example in a modus ponens argument, or exam situations. However, such situations where conversational presumptions are lifted in advance are very different from the cases of for example scalars where an implicature is computed on the basis of the assumption that the maxims are in force. Conceiving of a context as a common ground we may interpret a Gazdarian clausal set {Pj, P:j} as signalling that both Pj and P:j are compatible with but not entailed by contextual information. Geurts (1995, 1999) pointed out that Gazdar’s clausals can be seen as unresolved issues in a discourse constraining the process of presupposition projection.23 Both in (1a) and in (15) the clausal set indicates that it is not known (believed, assumed) whether John owns a donkey. (15) It is possible that John owns a donkey. Instead of saying that disjunctions as in (1a) or epistemic modals as (15) invoke implicatures we better say that they signal an issue in the ongoing discourse. Such an issue may be addressed explicitly and be settled by the claim that John owns a donkey or by the claim that he does not. Note that if an utterance signals in whatever way that the discourse context contains some issue, it would be self-defeating if the utterance of the very same sentence would settle this issue by presuppositional means. For example, in (1a) this would imply that the speaker both indicates that he does not know whether John owns a donkey and assumes that the issue whether John owns one has already been settled. Standing issues in a discourse thus may prevent presuppositions from being projected to the main context. This gives rise to a simple constraint on presupposition projection: A presupposition may not be processed in such a way that the resulting structure would settle an issue.

This view is supported by the now well-established account of information structure that assumes a direct correlation between topics, questions and the acceptability of discourse units.24 On this view a discourse owes its coherence to the fact that utterances can be viewed 23 Just as Gazdar’s thesis Geurts’ thesis was written under the supervision of Hans Kamp. It was completed in 1995. References are to his 1999 book which is an expanded and reworked version of this work. 24 See, for example van Kuppevelt (1995) or Ginzburg (1995).

PRAGMATIC STRATEGIES

55

as answers to explicit or implicit questions. And the view is supported by the fact that there is a straightforward correlation between question/answer pairs and the information structure of a sentence. In the case of wh-questions this correlation is reflected in the focus/ background division of a sentence. By lambda-abstracting over the non-focused part we obtain the focal background and by binding the relevant variable by a wh-operator we obtain the question. This question may be contextually given in which case the utterance that answers should be congruent with the question to be felicitous, that is the focus of the answer should correspond with the wh-position in the question. If no question is found in the incoming context, the focus/ background division provides the clue for the interpreter to construct the question the utterance is supposed to answer. We will, to use the presuppositional terminology, accommodate it. It is with respect to such an accommodated question that the original utterance expresses an answer. The relevance of this fact for presupposition theory has already been observed by Strawson (1964). With respect to disjunctions Grice characterized the situation as follows: A standard employment of ‘or’ is in the specification of possibilities [ . . . ] each of which is relevant in the same way to a given topic. ‘A or B’ is characteristically employed to give a partial [ . . . ] answer to some ‘‘W’’-question to which each disjunct, if assertible, would give a fuller, more specific satisfactory answer. Grice (1989: 68)

Simons (2001) develops this idea in a Hamblin-type theory of questions. She points out that Grice’s topic can be characterized as the question the disjunction is supposed to partially answer and she shows in detail that a disjunction can only provide an (informative) answer to a question in case each disjunct provides a (partial) answer to a question. Note that (1a), the disjunction we started with, is a special case. It is typically an answer to the following yes/no-question: (16) Does John own a donkey? Note furthermore that in a partition analysis like Groenendijk and Stokhof ’s25 the cells of the partition induced by the interrogative correspond to the Gazdarian clausals if we interpret these as signalling that both the information that John owns a donkey and that he does not own a donkey are compatible with the context. And, note finally that such an interpretation of clausals as contextually given issues or questions does not construe them as conversational implicatures, but 25

See Groenendijk and Stokhof (1984).

56

ROB

VAN DER

SANDT

typically interprets them as constraints on input-contexts. We will see that a proper appreciation of this fact suggests a simple solution to the problem posed by the so-called bathroom sentences.

Bridging to Questions in Discourse Evans (1977) pointed out that sentences like (17), that is the pronominalized version of (1a), are problematic for theories that treat pronouns as bound variables.26 (17) Either John does not own a donkey or he keeps it very quiet. They are equally problematic for current theories of dynamic interpretation.27 Though the accommodation mechanism of the anaphoric account of presupposition yields the correct result for (1a) by predicting that the presuppositional anaphor his donkey will be locally accommodated in the representation of the second disjunct, it is unable to handle its pronominal counterpart. The reason is simple. On the anaphoric account presuppositional constructions allow accommodation since they normally carry enough descriptive material to establish an antecedent. In the present case the information that John owns a donkey can be retrieved from the full presuppositional construction. This material may be used to create an antecedent at some accessible level of discourse structure in case the discourse does not already provide one. The explanation for (1a) now runs as follows. The first conjunct is not accessible for anaphoric reference which precludes binding. In order to get an interpretation we thus have to accommodate. Global accommodation is not an option. Accommodation of the presuppositional information at the main level of discourse 26

The problem was first pointed out and discussed by Geach (1962: 129–130). The example that has become famous in the linguistic literature is Partee’s ‘Either there is no bathroom in this house or it is in a funny place.’ Hence the label ‘bathroom’-sentence. 27 Krahmer and Muskens (1995) give a technical solution by reducing the problem posed by ‘bathroom’-sentences to the fact that double negation creates inaccessible contexts in DRT. They note that a disjunction can be rendered as an implication which yields ::j1-j2 for :j13j2 and then present a semantics for DRT in which double negation acts as a ‘hole’ for anaphoric reference. This reduces the disjunction problem to the double negation problem. There are several problems with this proposal though. One is that accessibility in such contexts does not always require that the negative element(s) are explicitly given, as pointed out by Geurts (1999: 74 ff, 128 ff). A second problem is that it does not generalize to the epistemic lists of alternatives such as in (20a) as pointed out by Van Rooy (2001: 639).

PRAGMATIC STRATEGIES

57

would violate conditions on discourse acceptability; it would result in a structure that corresponds with the following infelicitous discourse: (18) John owns a donkey. Either he does not own a donkey or he keeps it very quiet. The plain assertion of ‘John owns a donkey and keeps it very quiet’ would have conveyed the same information as (17) but in a less redundant way. The only accessible position left is the second disjunct and the material will thus be accommodated there, yielding an interpretation which can be paraphrased as: (19) Either John does not own a donkey or he owns a donkey and keeps it very quiet. This correctly accounts for the interpretation of (1a). However, in conjunction with the claim that presuppositional expressions are essentially anaphors this solution is problematic. If presuppositional expressions are treated on a par with pronouns it is imperative to provide a unified treatment of both (1a) and (17), that is, we have to give a uniform explanation for the interpretation of the explicit presuppositional construction and its pronominalized counterpart. Since pronouns are devoid of descriptive content and thus lack the capacity to accommodate, the explanation given for (1a) does not carry over to (17). In the latter the pronominal variable will not find an accessible antecedent and thus remains free leaving this structure without an interpretation. The problem is aggravated when we see that the phenomenon is not confined to disjunctions, but also occurs in conjunctions as the following:28 (20) a. b.

It is possible that John does not own a donkey, but it is also possible that he keeps his donkey very quiet. It is possible that John does not own a donkey, but it is also possible that he keeps it very quiet.

As in the disjunctive case, an analysis by means of accommodation yields the right result in (20a) but is not applicable in the case of (20b). And again we intuitively perceive a link from the anaphoric expression in the second conjunct to a formally inaccessible position in the first. We would thus, as in the disjunctive case, like to have an 28 These examples originate in Liberman (1973), and have been extensively discussed in the earlier presupposition literature.

58

ROB

VAN DER

SANDT

explanation which holds both for the case where the anaphoric expression is given by a full description and for its pronominalized counterpart. Note, moreover, that the modal examples are similar to the disjunctive counterparts in that the conjuncts give a specification of two alternative possibilities which typically give an answer to the same yes/no-question. And note finally that in both cases the specification of the possibility that John does not own a donkey and the possibility that he owns one but keeps it in hiding, should be exhaustive with respect to the knowledge of the speaker in order for him to know that the full sentence is true. The parallelism between disjunctions on the one hand and modalized conjunctions as in (20) on the other is made explicit in Zimmermann (2000) and Geurts (2005) who argue that disjunctions can be analyzed as (exhaustive) lists of epistemic possibilities. On this account the logical form of ‘not j or w’ comes out as }:j 4 }w and (1a) may be thus paraphrased as (20b). Adopting this analysis allows us to give a uniform analysis of the disjunctive case and such a list of epistemic conjuncts. I want to suggest that the key to the solution is found in Kamp and Reyle (1993: 187–190). They observe that a disjunction of the form ‘A or B’ can be paraphrased as ‘A or otherwise B’. In such a paraphrase of (1a) the otherwise would refer to the other case, that is the case described by the first disjunct. According to Kamp and Reyle this licenses us to postulate a representation for what is referred to by otherwise and then extending that representation by incorporating it in the representation of [: he keeps it quiet]. Geurts (1999: 75) adopts this idea and proposes to insert the material into the second conjunct as an inferred antecedent. Both explanations are incomplete though. On Kamp and Reyle’s view the question arises where the material encoding ‘the other case’ comes from. It cannot simply be constructed as the complement of the first disjunct for then we would end up with a double negation and thus an inaccessible antecedent after all. We thus have to assume that ‘the other case’ refers to the corresponding unnegated sentence and that the material thus retrieved may be inserted. And on Geurts’ account we are faced with the question: inferred in what way? However, if we take the remarks by Grice (p. 51 above) seriously and once we see that the disjunction is to be interpreted as an answer to the (explicitly given or implicitly assumed) yes/no-question ‘Does John own a donkey?’, the answer is straightforward. In order to interpret the disjunction the interpreter has to accommodate the question. And adopting the standard Hamblin-type account of questions we find the alternative possibilities in the cells of the partition of the question thus accommodated.

PRAGMATIC STRATEGIES

59

To illustrate such an analysis let us first consider the nondisjunctive case: (21) a. b.

Does John own a donkey? *He keeps it very quiet. Does John own a donkey? Sure. But he keeps it very quiet.

The first discourse is out. The question does not imply that the speaker assumes that there is a donkey and the indefinite is not available for pronominal reference. The second discourse is fine. The pronoun gets its value from the answer which in turn is one of the cells of the bipartition the question induces. This is uncontroversial. The situation in the disjunctive case is basically the same. The second disjunct gives an answer to the question by being in the second cell of the partition. Together with the incoming context this cell figures as a provisional context for the interpretation of this disjunct. It follows that the use of the pronoun is licensed since John’s donkey inhabits each of the worlds of this provisional context. The second disjunct thus gets a determinate interpretation in its provisional context which entails that John owns a donkey. The proposal I want to make thus agrees both with Kamp and Reyle that otherwise refers to the other case and with Geurts that the interpretation is secured in view of a bridging inference. But unlike these authors I want to claim that ‘the other case’ is not constructed out of the first disjunct. Instead ‘the other case’ is given by one of the cells of the partition induced by the antecedently constructed question. Of course this does not imply that the indefinite in the question is accessible for anaphoric reference. In order to interpret the disjunction we have to turn to the cells of the bipartition individually. The first disjunct corresponds with the cell which presents the negative answer. We then turn to the alternative cell, thus deriving (22) If John owns a donkey, he keeps it very quiet. as the bridging inference. If this analysis is on the right track the relevant interpretation does not come about by some anaphoric link across the disjuncts, nor is it due to straightforward insertion of material from the first disjunct into the second. The relation is indirect in that the full disjunction allows us to create some question by accommodation. It is the question thus accommodated which figures as the context of interpretation of the disjunction and which allows us to interpret the pronominal element.

60

ROB

VAN DER

SANDT

Conclusion In 1979 Hans Kamp gave the following formulation of the received view on the relation between semantics and pragmatics in a theory of meaning: (i) The concept, or concepts, characterized by the recursive component of the theory belongs to semantics. (ii) Moreover no notion to which this component refers belongs to pragmatics. Kamp (1979: 265) He went on to state and argue that there is ‘no good reason why semantics and pragmatics should be separate in the sense that [the principle stated above] tries to articulate’. Kamp’s argumentation was based on the analysis of non-assertoric speech acts. The same view, but now based on the analysis of generalized conversational implicatures, was recently put forward forcefully by Steve Levinson in Levinson (2000). In this paper I discussed Gazdar’s pragmatic theory as a particularly elegant elaboration of the received view. Gazdar’s theory is based on the assumption that semantic content can be determined without calling on pragmatic principles. A sentence in a context determines a proposition in the classical sense. Pragmatic processing only enters the scene after this semantic object has been established thus allowing us to determine a more substantial informational object, the information conveyed. Subsequent work in discourse representation theory and, in particular, the developments in presupposition theory and related areas in the nineties have shown that this picture is in need of a thorough revision. The idea that the semantic and pragmatic components of a theory of meaning can be seen as separate modules in the sense that a sentence in a context yields a fully determined proposition which may then be ‘enriched’ or ‘strengthened’ by pragmatic inferencing, thereby only affecting the output context, turned out to be untenable. Instead, at the very moment an utterance is processed principles of a Gricean nature play a crucial role in determining the background with respect to which this utterance will get an interpretation. This is abundantly clear when we have a closer look at the all pervasive phenomenon of presupposition accommodation, and, in particular, when we look at the factors that constrain this process. In this paper I went back in history, took Gazdar’s modular view as a starting point and showed first that, even when we adhere to

PRAGMATIC STRATEGIES

61

Gazdar’s original architecture, some obstacles can easily be removed. I showed that there is no reason to stick to the strong epistemic modalization of the various semantic and pragmatic components and, furthermore, that there is no need to retain the counterintuitive ordering between implicatures and presuppositions. Adopting a dynamic view on meaning and reinstating presuppositional information as information that is contextually required to establish an interpretation, I showed that Gazdarian clausals should not be conceived of as standard Gricean implicatures. They should, instead, be seen as constraints on input contexts regulating the resolution process of inter alia presuppositional information. I suggested that they are better conceived of as questions or unresolved issues in discourse. And I showed that it is exactly the possibility of such a reinterpretation that explains why Gazdar’s original clausals did the work they were devised for. This illustrates why the semantic autonomy hypothesis cannot be upheld. In cases where presuppositional information is not yet there, we need to accommodate. Accommodation is an operation on input contexts and this process is constrained by factors which are uncontroversially pragmatic and Gricean in nature. Acknowledgment I would like to thank Bart Geurts, Emar Maier, Jennifer Spenader and the participants of the Kamp workshop on Presupposition for their helpful comments.

References Asher, N. and A. Lascarides. 1998. The semantics and pragmatics of presupposition. Journal of Semantics 15:239–300. Beaver, D. 1995. Presupposition and Assertion in Dynamic Semantics. Ph.D. thesis, University of Edinburgh, Scotland. Beaver, D. 1997. Presupposition. In J. van Benthem and A. ter Meulen, eds., Handbook of Logic and Language, pp. 939–1008. Amsterdam: Elsevier. Beaver, D. 2001. Presupposition and Assertion in Dynamic Semantics. Stanford, CA: CSLI Publications. Bos, J. 1999. The treatment of focus particles in underspecified discourse representations. In P. Bosch and R. van der Sandt, eds., Focus. Linguistic, Cognitive, and Computational Perspectives, pp. 121–141. Cambridge: Cambridge University Press. Evans, G. 1977. Pronouns, quantifiers and relative clauses I. Canadian Journal of Philosophy 7:467–536.

62

ROB

VAN DER

SANDT

¨ ber Sinn und Bedeutung’. Zeitschrift fu Frege, G. 1892. ‘U ¨r Philosophie und philosophische Kritik 100:25–50. Translated by Max Black in Geach and Black, 1970, pp. 56–78. Gazdar, G. 1979. Pragmatics: Implicature, Presupposition, and Logical Form. New York: Academic Press. Gazdar, G. 1981. Speech act assignment. In A. K. Joshi, B. L. Webber, and I. Sag, eds., Elements of Discourse Understanding. Cambridge: Cambridge University Press. Geach, P. T. 1962. Reference and Generality. An Examination of some Mediaeval and Modern Theories. Ithaca, NY: Cornell University Press. Geurts, B. 1995. Presupposing. Ph.D. thesis, University of Stuttgart, Stuttgart. Geurts, B. 1996. Local satisfaction guaranteed: A presupposition theory and its problems. Linguistics and Philosophy 19:259–294. Geurts, B. 1999. Presuppositions and Pronouns. Amsterdam: Elsevier. Geurts, B. 2005. Entertaining alternatives: Disjunctions as modals. Natural Language Semantics 13:383–410. Geurts, B. and E. Maier. 2003. Layered DRT. ms. Ginzburg, J. 1995. Resolving questions I and II. Linguistics and Philosophy 18:459–527. Grice, P. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press. Groenendijk, J. and M. Stokhof. 1984. On the semantics of questions and the pragmatics of answers. In F. Landman and F. Veltman, eds., Varieties of Formal Semantics, pp. 143–170. Dordrecht: Foris. Hamblin, C. 1987. Imperatives. Oxford: Blackwell. Hamblin, C. L. 1971. Mathematical models of dialogue. Theoria 37:130–155. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. thesis, University of Massachusetts, Amherst, MA. Heim, I. 1983. On the projection problem for presuppositions. In M. Barlow, D. P. Flickinger and M. T. Wescoat, eds., Proceedings, Second West Coast Conference on Formal Linguistics, pp. 114–125. Stanford, CA: Stanford Linguistics Association. Hintikka, J. 1962. Knowledge and Belief. Ithaca, NY: Cornell University Press. Hirschberg, J. 1991. A Theory of Scalar Implicature. New York: Garland. Horn, L. R. 1972. On the Semantic Properties of Logical Operators in English. Ph.D. thesis, University of California, Los Angeles, CA. Horn, L. R. 1989. A Natural History of Negation. Chicago, IL: Chicago University Press. Kamp, H. 1979. Semantics versus pragmatics. In F. Guenthner and S. Schmidt, eds., Formal Semantics and Pragmatics of Natural Language, pp. 255–287. Dordrecht: Reidel.

PRAGMATIC STRATEGIES

63

Kamp, H. 1981. A theory of truth and semantic representation. In J. A. Groenendijk, T. Janssen, and M. Stokhof, eds., Formal Methods in the Study of Language, pp. 277–322. Dordrecht: Foris. Kamp, H. 2001. Computation and justification of presuppositions. In M. Bras and L. Vieu, eds., Semantics and Pragmatics of Discourse and Dialogue. Experimenting with Current Dynamic Theories. Amsterdam: Elsevier. Kamp, H. and U. Reyle. 1993. From Discourse to Logic: Introduction to Modeltheoretic Semantics in Natural Language, Formal Logic and Discourse Representation Theory, vol. 1. Dordrecht: Kluwer Academic Publishers. Karttunen, L. 1974. Presuppositions and linguistic context. Theoretical Linguistics 1:181–194. Karttunen, L. and S. Peters. 1979. Conventional implicature. In C. Oh and D. A. Dinneen, eds., Syntax and Semantics 11: Presupposition, pp. 1–56. New York: Academic Press. Krahmer, E. 1998. Presupposition and anaphora. Stanford, CA: CSLI Publications. Krahmer, E. and R. Muskens. 1995. Negation and disjunction in discourse representation theory. Journal of Semantics 12:357–376. Lerner, J.-Y. and T. Zimmermann. 1981. Mehrdimensionale Semantik. Die Pra¨suppositionen und die Kontextabha¨ngigkeit von ‘Nur’. Technical report, Universita¨t Konstanz, Sonderforschungsbereich Linguistik, 99, Konstanz. Levinson, S. C. 1983. Pragmatics. Cambridge: Cambridge University Press. Levinson, S. C. 2000. Presumptive Meanings. Cambridge, MA: The MIT Press. Lewis, D. K. 1979. Scorekeeping in a language game. Journal of Philosophical Logic 8(3):339–359. Liberman, M. 1973. Alternatives. In Papers from the 9th Regional Meeting, pp. 356–368. Chicago, IL: Chicago Linguistic Society. Sæbø, K. J. 1996. Anaphoric presuppositions and zero anaphora. Linguistics and Philosophy 19:187–209. Simons, M. 2001. Disjunctions and alternativeness. Linguistics and Philosophy 24:597–619. Soames, S. 1982. How presuppositions are inherited: A solution to the projection problem. Linguistic Inquiry 13:483–545. Stalnaker, R. C. 1974. Pragmatic presuppositions. In M. K. Munitz and P. Unger, eds., Semantics and Philosophy. New York: Academic Press. Stalnaker, R. C. 1978. Assertion. In P. Cole, ed., Syntax and Semantics, vol. 9: Pragmatics, pp. 315–332. New York: Academic Press. Strawson, P. 1964. Identifying reference and truth-values. Theoria 30:96–118. Strawson, P. F. 1950. On referring. Mind 59:320–344. van der Sandt, R. A. 1988. Context and Presupposition. London: Croom Helm. van der Sandt, R. A. 1992. Presupposition projection as anaphora resolution. Journal of Semantics 9:333–377.

64

ROB

VAN DER

SANDT

van Kuppevelt, J. 1995. Discourse structure, topicality and questioning. Journal of Linguistics 31:109–147. van Rooy, R. 2001. Exhaustivity in dynamic semantics: Referential and descriptive pronouns. Linguistics and Philosophy 24:621–657. Zeevat, H. 1992. Presupposition and accommodation in update semantics. Journal of Semantics 9:379–412. Zimmermann, T. E. 2000. Free choice disjunction and epistemic possibility. Natural Language Semantics 8:255–290.

3

Have You Noticed that Your Belly Button Lint Color is Related to the Color of Your Clothing?$ DAVID BEAVER

Abstract Karttunen identified a class of semi-factive verbs. This was erroneous, but enlightening. Stalnaker and Gazdar explained Karttunen’s data as involving cancellation of presuppositions as a result of pragmatic reasoning, an account reformulated by van der Sandt. In this paper I present a large number of naturally occurring examples bearing on the question of how factive verbs interact with implicatures, and show that many of these examples are problematic for existing accounts. I end by presenting suggestive evidence involving the relation between presupposition and information structure.

Pre-Fight Coverage In the red corner stands a presupposition trigger. In the blue corner, a quantity implicature. Who wins? Stalnaker (1974) says the implicature $

This paper has been circulated since 2002, and appears here modulo only minor corrections and improvements. All data was collected during 2002, and links to web sources in the text appear as then current. Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

65

66

DAVID BEAVER

is stronger. Gazdar (1979) says the implicature gets the first punch. Knockout. van der Sandt (1992) says the presupposition runs away, cowering in a dingy corner of an embedded discourse representation structure (DRS). All suggest the presupposition has no chance: conversational principles will soon swagger about the ring while the champion implicature gives interviews. The pundits have spoken, but what really happens? This paper is about the interaction between presuppositions and implicatures triggered by factive verbs (Kiparsky and Kiparsky, 1970). I will be concerned with cognitive factives, which is the class of factive verbs used primarily to indicate what information the subject has or how the information is acquired or lost. Members of this class in English include ‘‘know,’’ ‘‘realize,’’ ‘‘discover,’’ ‘‘notice,’’ ‘‘recognize,’’ ‘‘find out,’’ ‘‘remember,’’ ‘‘forget,’’ ‘‘be aware,’’ ‘‘be unaware,’’ ‘‘admit,’’ ‘‘intuit,’’ and a subclass of sensory factives ‘‘sense,’’ ‘‘see,’’ ‘‘smell,’’ ‘‘hear,’’ ‘‘detect,’’ ‘‘observe.’’ The other major class of factive verbs, the class I will largely ignore, is the emotives, factive verbs used primarily to convey the subject’s emotional attitude toward information. This class includes ‘‘regret,’’ ‘‘be annoyed,’’ ‘‘be upset,’’ ‘‘be glad,’’ ‘‘be happy,’’ ‘‘be ecstatic.’’ After reviewing what has been said on the interaction between presuppositions and implicatures of factives, I will go on to consider naturally occurring data. The bulk of this paper consists of examples occurring on the world wide web located using the Google search engine. The logic behind using the internet rather than a more structured linguistic corpus is simple: first, my goal in this paper is not to establish quantitative results but to show existence proofs, i.e., that certain types of example do occur naturally; second, there is a huge difference of scale. In many cases the patterns I have searched for are quite rare, and existing linguistic corpora are not large enough to be useful. To give an example, a search for the pattern ‘‘I am not aware that’’ produced six examples in a corpus of ten years of the New York Times, which took several minutes using the Unix command fgrep. The same request produced four examples using the online British National Corpus, and took about 20 seconds. But Google was able to locate 15,000 examples of this string in 0.14 seconds. Some of these examples are below. While some of the data I will present is in agreement with the theoretical predictions outlined above, much is problematic. In fact, I will eventually suggest that in concentrating on the interaction between presupposition and implicature, we might just be watching the wrong fight.

HAVE YOU NOTICED . . .

67

On the Record Stalnaker was considering data due to Karttunen (1971), data involving factive verbs embedded in hypothetical and question contexts. General conversational principles tell us (i) that when a hypothetical is uttered there is normally doubt as to whether the hypothesis is true, and (ii) when a question is uttered, there is normally uncertainty as to the answer. Stalnaker argues that these principles, combined with the semantic content of the utterance, help determine what a speaker might be presupposing. An occurrence of a factive verb, a presupposition trigger, provides us with evidence that the propositional complement of the verb is being presupposed. But, according to Stalnaker, we should ignore that evidence if it conflicts with the above two conversational principles. Here are some examples, those in (1) and (3) being from Karttunen (1971): (1) a. Did you regret that you had not told the truth? b. Did you realize that you had not told the truth? c. Did you discover that you had not told the truth? (2) a. Did she regret that you had not told the truth? b. Did she realize that you had not told the truth? c. Did she discover that you had not told the truth? (3) a.

If I regret later that I have not told the truth, I will confess it to everyone. b. If I realize later that I have not told the truth, I will confess it to everyone. c. If I discover later that I have not told the truth, I will confess it to everyone.

(4) a.

If she regrets later that I have not told the truth, I will confess it to everyone. b. If she realizes later that I have not told the truth, I will confess it to everyone. c. If she discovers later that I have not told the truth, I will confess it to everyone.

Here the factive verbs are ‘‘regret,’’ ‘‘realize,’’ and ‘‘discover,’’ and the issue is whether the complement ‘‘I have not told the truth’’ is presupposed in each example. According to both Karttunen and Stalnaker, examples (1a,b) and (3a) carry this presupposition,1 while 1

More properly, for Stalnaker it is speakers rather than sentences that presuppose.

68

DAVID BEAVER

(1c) and (3b,c) do not. According to Stalnaker, uses of the factive verbs, as in all the remaining examples (2a,b,c) and (4a,b,c), should be expected to carry the factive presupposition. Consider (3b). Abbreviate ‘‘I have not told the truth’’ as f. Stalnaker argues that since the hypothesis that the speaker will realize f is open, and the speaker has indicated this by use of the conditional, the speaker cannot already believe f. So although use of a factive verb leads to a presumption that f is being presupposed, in this case there can be no presupposition. Similarly for (3c). However, when the trigger is ‘‘regret’’ as in (3a), the argument does not go through: the fact that it is open whether the speaker will regret f does not tell us that the speaker does not already believe f, so the presupposition goes through. Likewise for all of the third person cases in (4a,b,c): it can be open what attitude ‘‘she’’ will have toward f in the future although the speaker and hearer already believe f. So the presupposition survives. The reasoning for the question cases in (1) and (2) proceeds along the same lines: the presupposition that f survives unless that would foreclose the addressee’s answer.2 Gazdar (1979) presented a formal theory in which quantity (and other) implicatures cancel presuppositions. Gazdar proposes that a hearer starts with an information state consisting of a set of propositions which have been established, then adds the ordinary semantic content of a new utterance, then adds any potential implicatures which are consistent with this set (and each other), and then adds any potential presuppositions which are consistent with that set (and each other). The presuppositions of an utterance are just those potential presuppositions which survive this vetting procedure. In (3), the potential implicature that the speaker does not know whether he or she will realize f is added to the information state before the potential presupposition triggered by ‘‘realize.’’ For Gazdar, presuppositions are epistemic, and in this case the potential presupposition is that the speaker knows f. Assuming some further axioms that enforce monotonic growth of information, the potential presupposition is not 2 The distinction between ‘‘realize’’ and ‘‘discover’’ in (1b,c) emerges, according to Stalnaker, only once we take into account the different aspectual classes of these two verbs. He claims that (1b) carries the presupposition because the question ‘‘Did you realize . . . ?’’ refers to some specific time in the past, and the question of whether the addressee realized at that time that f is not foreclosed by the addressee knowing now that f. On the other hand ‘‘Did you discover . . . ?’’ is claimed to quantify over all prior times. If the addressee knows f now, then he or she must have discovered it at some prior time, and so would be forced into a positive answer. To return the addressee the option of a negative answer, we must accept that f is not presupposed.

HAVE YOU NOTICED . . .

69

compatible with the implicature, so the potential presupposition never gets added. In Gazdar’s terms, it is cancelled. What we see in Gazdar’s treatment of this type of example is a codification of Stalnaker’s account. Stalnaker’s ‘‘presumptions’’ that the speaker is presupposing something become Gazdar’s ‘‘potential presuppositions’’; what Stalnaker called ‘‘indications’’ that the speaker was making are recognized as Gricean implicatures, and formalized via a mechanism that generates ‘‘potential implicatures’’; the fact that indications are stronger than presumptions in the examples Stalnaker discusses becomes, in Gazdar’s model, a general principle that we update with implicatures before presuppositions, and only add presuppositions if they do not conflict with anything added earlier. Note that Gazdar does not explicitly deal with the question cases above. But for the purposes of argument, I will suppose that he would treat polar questions as involving an epistemic operator ‘‘?’’, such that ‘‘f?’’ does not entail f or its negation. In this case, we expect clausal implicatures to be triggered by questions, and, modulo specifying the lexical meaning of the factive verbs themselves, we would be able to formally reproduce Stalnaker’s analysis of second person questions. The last theoretical perspective on presupposition/implicature conflict that I want to mention is that of van der Sandt (1992). In this account, there is no such thing as cancellation. Following observations of Heim (1983), van der Sandt proposes a model in which the question is not whether a presupposition is projected or cancelled, but whether it is accommodated globally or locally. There is a general preference for global accommodation, but accommodation must respect general conversational principles of informativeness and consistency. Put differently, the presupposition wants to contribute to the main context, but runs away and hides in an embedded context on pain of producing conflict with Gricean principles. The result is that although van der Sandt does not spell out implicatures in the way Gazdar does, his model predicts local accommodation whenever Gazdar’s model would predict cancellation, at least in cases of cancellation due to clausal implicature or the need to preserve consistency with the main entailment of the utterance. Gazdar also analyzes scalar implicatures, which are not formally dealt with in van der Sandt’s account, but all the cases considered above can either be seen as involving clausal implicatures (for hypotheticals and questions), or as involving potential inconsistency with the main entailment (for the negation examples). So, once we have identified cancellation with local accommodation, the predictions of van der Sandt’s model are exactly the same as Gazdar’s in the cases discussed

70

DAVID BEAVER

above. In fact, apart from a small group of cases that will be discussed below, the predictions are the same for all examples to be considered in this paper. I refrain here from a detailed formal presentation of the models of Gazdar and van der Sandt: apart from the original sources cited above, the reader is directed to Soames (1982) and Mercer (1992) for discussion of and extension to Gazdar’s model, Zeevat (1992), Kamp and Rossdeutscher (1994), Krahmer (1998), Kamp (2001a, 2001b), Beaver (2002) and, especially, Geurts (1999) for discussion and development of van der Sandt’s model, and Beaver (1997, 2001) for discussion of both models and comparison with a range of alternatives.

Implicature Wins as Predicted Is there naturally occurring data which accords with the models described? Although the literature is based entirely on artificial examples, it is easy to find naturally occurring cases. As a first illustration, consider the use of the phrase ‘‘I am not aware’’ by such luminaries as John Stuart Mill, Theodore Roosevelt, Sherlock Holmes, General Myers (Chairman of the Joint Chiefs of Staff), and Deputy Postmaster General John Nolan (relevant occurrence of a factive verb is boxed for ease of identification): (5) First person, cognitive factive under negation, cancellation: I am not aware that any community has a right to force another to be civilised. J. S. Mill, On Liberty, ed. R. B. McCallum (Oxford, 1946), p. 83. (6) First person, cognitive factive under negation, cancellation: I think Great Britain is now showing great courtesy and forbearance. I believe that she has done things to our ships that ought not to have been done, but I am not aware that she is now doing them. Theodore Roosevelt, letter to Sir Edward Grey, January 22, 1915. (7) First person, cognitive factive under negation, cancellation: ‘‘I think that I may go so far as to say, Watson, that I have not lived wholly in vain,’’ he remarked. ‘‘If my record were closed tonight I could still survey it with equanimity. The air of London is the sweeter for my presence. In over a thousand cases I am not aware that I have ever used my powers upon the wrong side.’’

HAVE YOU NOTICED . . .

71

Arthur Conan Doyle, ‘‘The Final Problem,’’ in Memoirs of Sherlock Holmes, Harper and Brothers, New York, 1894. (8) First person, cognitive factive under negation, cancellation: QUESTION FROM INTERFAX: Can we expect Russian military forces to be among the Coalition forces deployed at Manas airport? GENERAL MYERS: I am not aware that there will be Russian forces here. I mean I can’t rule it out, but I am personally not aware that there will be Russian forces here as part of the coalition. Press Conference, Bishkek, Kyrgyz Republic, February 18, 2002. (9) First person, cognitive factive under negation, cancellation: Responding to a reporter’s question Wednesday about the 911 call, Deputy Postmaster General John Nolan said, ‘‘I am not aware that he [Morris] saw the Daschle letter . . . aware he saw some letter.’’ CNN, November 8, 2001. What if we spread our net wider than ‘‘aware’’? There is no shortage of similar examples of cancellation of first person factives under negation. Witness the following uses of ‘‘know,’’ ‘‘trouver’’/‘‘discover,’’ and ‘‘notice’’ due to Dickens, Rousseau and Roosevelt again: (10) First person, cognitive factive under negation, cancellation: ‘This night, my husband being away, he has been with me, declaring himself my lover. This minute he expects me, for I could release myself of his presence by no other means. I do not know that I am sorry, I do not know that I am ashamed, I do not know that I am degraded in my own esteem. All that I know is, your philosophy and your teaching will not save me. Now, father, you have brought me to this. Save me by some other means!’ [Louisa Bounderby, ne´e Gradgrind] Charles Dickens, Hard Times, Bradbury & Evans, 1854. (11) First person, cognitive factive under negation, cancellation: Pour avoir commence´ tard mettre en exercice ma faculte´ judiciaire, je n’ai pas trouv´e qu’elle eu ˆt perdu sa vigueur; et quand j’ai publie´ mes propres ide´es, on ne m’a pas accuse´ d’eˆtre un disciple servile, et de jurer in verba magistri. Jean Jacques Rousseau Les confessions, H. Launette & Companie, Paris, 1889 (originally published 1782, 1789).

72

DAVID BEAVER Though it was late before I began to exercise my judicial faculties, I have not discovered that they had lost their vigor, and on publishing my own ideas, have never been accused of being a servile disciple or of swearing in verba magistri. Translation: W. Conyngham, Tudor Pub. Co., New York, 1935.

(12) First person, cognitive factive under negation, cancellation: She is so bright and amusing and now seems perfectly happy, and is not only devoted to Archie and Quentin but is very wise in the way she takes care of them. Quentin, under parental duress, rides Algonquin every day. Archie has just bought himself a football suit, but I have not noticed that he has played football as yet. Theodore Roosevelt, Letter to Kermit Roosevelt, October 15, 1904. And here is a morbid example of cancellation involving the factive ‘‘find out’’: (13) First person, cognitive factive under negation, cancellation: I haven’t had any dealings with Tri-State since 1996. Hopefully, I won’t find out that any of the bodies found there were ones I sent. I know I did send 11 bodies around 1991–1992. I find myself thinking about what a perverse mind one must have to just pile lifeless, decomposing bodies on one another so he can save money. Bob Perry, Perry Funeral Home owner, as quoted in The Post, Alabama, February 25, 2002. Cancellation of presuppositions when a cognitive factive trigger in the first person is embedded in a conditional also occur in natural contexts. For example, this seems to be common in a previously undescribed genre of academic disaster text: (14) First person, cognitive factive in conditional, cancellation: Cheating is not allowed in this class. If I notice that you appear to have trouble keeping your eyes on your own paper during a quiz or the final exam you will be asked to sit in the front row to complete the quiz or exam. If it is evident that you have copied from a neighbor you will receive a 0 on that particular quiz or exam and will be subject to disciplinary procedures within the

HAVE YOU NOTICED . . .

73

College of Business and Public Administration and Drake University. Marie Klugman/Stuart Klugman, Stat 60, Department of Actuarial Science and Statistics, Drake University.3 (15) First person, cognitive factive in conditional, cancellation: Excessive absence (over four absences) will result in a grade of F. **Unexcused absences over the two class limit will lower your grade by a half letter for each class.** If I notice that you leave class during the break without consulting me, I will mark you absent for that class. Tardiness will count as absence if you fail to tell me (at the end of class) that you are present. These policies are not negotiable. Albert Rouzie, Course Policies, English 153, Writing & RhetoricSpecial Topics, Ohio State University, 1999.4 (16) First person, cognitive factive in conditional, cancellation: You may use your laptop to take notes; however, you should not use your laptop for anything else. If this is a problem for you, I would encourage you to use a pen or pencil for your notes. You should not be connected to the network during lecture periods, and I may disconnect you from the network without warning if I notice that your network cable is in use. Christopher C. Taylor, General Course Policies, Milwaukee School of Engineering, 1998–2002.5 (17) First person, cognitive factive in conditional, cancellation: If I discover that you have turned in plagiarized work, I will report this immediately to the Dean of Studies and you will receive an automatic grade of F for that piece of work, and, in most cases, a final grade of F for the course. Nicola Denzey, Grading Policies: The Bible in contemporary American Film, Skidmore College, Summer Session, 2001.6

3

http://www.drake.edu/cbpa/acts/stat60.htm http://www-as.phy.ohiou.edu/Brouzie/fall153/policy.html 5 http://people.msoe.edu/Btaylor/resources/policy.htm 6 http://www.skidmore.edu/Bndenzey/re330_film/filmcontract.htm 4

74

DAVID BEAVER

(18) First person, cognitive factive in conditional, cancellation: If I discover that your work is plagiarized, I will be forced to notify the Dean. Beth Berkowitz, Requirements, C1001 – Literature Humanities; Section 52.7 (19) First person, cognitive factive in conditional (plus quantificational adverb), cancellation: (1) Please ask for additional clarification on any grade that you do not understand. I will never penalize you in any way for doing this, even if I discover that your actual grade should have been recorded as lower. Theodore D. George, Social and Political Philosophy, Phil 332–500, Spring 2002.8 (20) First person, cognitive factive in conditional, cancellation: What if I discover that my book has missing pages half way through the semester? Caribou Bookstore, help page.9 Second person question cases like those described by Karttunen and Stalnaker also appear naturally. People and groups offering help have a penchant for asking very suggestive questions, as shown in the following examples. Note that while in some cases it might be claimed that the presupposition is projected for the intended audience of the text, in other cases it is clear that the truth of the presupposition remains open. In (26), the potential presuppositions conflict with each other, so it is clear that neither is actually presupposed. (21) Second person, cognitive factive in question, cancellation: Have you ever found yourself ‘‘disillusioned’’ with part of your life? Was there a hope or dream that didn’t turn out as expected? Have you discovered that some things you thought were true 7

http://www.columbia.edu/itc/lithum/berkowitz/course_ requirements.html 8 http://www-phil.tamu.edu/Philosophy/Faculty/george/332SYL02A. html 9 http://www.cariboobookstore.com/outerweb/store_information.asp

HAVE YOU NOTICED . . .

75

didn’t turn out to be? Illusions are self-created and do supply some benefit, but they require substantial energy to maintain and can interfere with your relationships. Bernice L. Ross, Lessons for Living and Learning, ULiveandLearn.com, Inc., Copyright dated 2000–2002.10 (22) Second person, cognitive factive in question, cancellation: What can you do with this new found knowledge? Have you discovered that you often remind yourself of your mother when she scolded you as a child (and you swore you would never be like her!); then you can make your Action Plan – I will not say those words to my child, I will find a better way to communicate. Have you discovered that you are often shy and allow others to step all over you? Then your Action Plan might be – I will stand up for myself the next time I feel I’m being used. Bobbie Ann Pimm, Your Action Plan, Copyright 2001.11 (23) Second person, cognitive factive in question, cancellation: Answering yes to some or all of the questions in a section might mean that you’ve been taken advantage of. [ . . . ] Did an agent sell you an ‘‘investment’’ or ‘‘pension’’ plan that turned out to be insurance? Have you discovered that the ‘‘cash value’’ of your policy has decreased or disappeared? Has your agent made promises that turned out to be untrue? James Hoyer, Attorney’s at Law, Consumer Checklist.12 (24) Second person, cognitive factive in question, cancellation: Did you ever realize that you did something only because you were ‘supposed’ to, or may be because that’s what everyone else was doing. Brenden Clarke, Life re-examination leads to fundamental questions, Eastern Echo, Fall Edition, 2000.

10 http://www.uliveandlearn.com/lessons/lessonlist.cfm?kc ¼ 6&u ¼ fr 11 http://www.bobbieann.net/YourActionPlan.html 12 http://www.jameshoyer.com/problem_lifeinsurance.html

76

DAVID BEAVER

(25) Second person, cognitive factive in question, cancellation: Did you learn something about yourself? Did you discover that you were stronger than you thought? Did you discover that you were not as strong as you thought? Harley King, Grief Support Writing: How to Write and Share Your Story of Pet Loss.13 (26) Second person, cognitive factive in question, cancellation: What surprised you? Did you discover that some things you thought were true have turned out to be false? Did you discover some things you thought were false that you now know are true? Journal and Report on METRO-APEX Participation.14 Sometimes presuppositions in second person questions are cancelled quite wistfully: (27) Second person, cognitive factive in question, cancellation: When you were a kid, did you ever realize that you were going to hate a teacher on the second day of school, only to realize soon thereafter that you were stuck with her and her goddamn for-agrade book report dioramas for the foreseeable future no matter what you did? Well, that’s my daily trip past the mirror, buddy! This lumpen mass isn’t going anywhere fast, and I do not have the patience to see this health kick through. Jimski, The World is Out of Stuff, October 3, 2000.15 (28) Second person, cognitive factive in question, cancellation: Did you ever realize that you Just simply couldn’t be without someone And that you would risk your whole damn life For their tender precious love You don’t want to do anything Without their hand to hold You simply cannot move 13

www.petloss.com/writing.htm http://www.usc.edu/schools/sppd/classes/ppmt/220/journal/ summary.html 15 http://www.jimski.net/oct00.html 14

HAVE YOU NOTICED . . .

77

Without their arm around you You just can’t find your strength Without that special touch Its like your soul is entirely shut off And has just went blank Sugar_spice_15, Suicidal Love.16 As a final example in this section of examples that basically accord with the Karttunen–Stalnaker generalization, a second person modal factive (followed by a doubly embedded first person factive under a second person factive) on which to reflect: (29) Second person, cognitive factive under negation and modal, cancellation: One day Chuang Tzu and a friend were walking by a river. ‘‘Look at the fish swimming about,’’ said Chuang Tzu, ‘‘They are really enjoying themselves.’’ ‘‘You are not a fish,’’ replied the friend, ‘‘So you can’t truly know that they are enjoying themselves.’’ ‘‘You are not me,’’ said Chuang Tzu. ‘‘So how do you know that I do not know that the fish are enjoying themselves?’’ Recounted by John Suler, Knowing Fish (undated web page), Department of Psychology Rider University.17 It is clear that Stalnaker’s argument, and thus also the models of Gazdar and van der Sandt, make appropriate predictions in a wide array of naturally occurring examples. But we will return to these very same examples at the close of this paper to investigate whether those models miss an important generalization.

The Third Person The formal models (Gazdar’s and van der Sandt’s) are clear in their predictions: while appropriately tensed first and second person uses of cognitive factives should lead to cancellation, third person uses of cognitive factives should, ceteris paribus, be immune to this effect. 16

http://www.4degreez.com/poetry/10984/1018070725.html Suler notes a number of reactions to this Nagelian story, including: ‘‘To think like a fish, you have to drink like a fish.’’ 17

78

DAVID BEAVER

I will now present counterexamples, third person uses of cognitive factives in which cancellation occurs. As mentioned above, there is one type of example of relevance for which the predictions of Gazdar’s model are different from those of van der Sandt’s model. These are cases in which there is a free variable in the presupposed proposition. A constraint which van der Sandt terms trapping prevents the presupposition from being accommodated outside of the scope of the binding operator, which can produce the effect of cancellation (i.e., local accommodation) when Gazdar’s theory would predict projection. This situation is not common in the corpus of factives I have collected, but it is sometimes the case that a defender of van der Sandt’s model could argue that a free temporal variable is present in the presupposition, and that this is responsible for forcing local accommodation. Consider the following two examples: (30) Third person, cognitive factive in conditional, cancellation: What can an individual do if he becomes aware that pets kept for sale are being maltreated? Carrie Jane Canniffe B.L., David Burke B.L., Barra Faughnan B.L., ISPCA Legal Handbook, Irish Society for the Prevention of Cruelty to Animals.18 (31) Third person, cognitive factive in conditional, cancellation: We must look at the regulation of our affairs, a Cheann Comhairle, to ensure strict adherence to the long-held tradition of the veracity of any statement put on the record; and if anyone becomes aware that misleading or inaccurate information is put on the record, that the long-standing practice of correcting the record is strictly applied. Mr. Howlin, Flood Tribunal Terms of Reference: Motion (Resumed), July 1, 1998.19 The factive complement ‘‘pets kept for sale are being maltreated’’ in (30) is ambiguous: it can refer to maltreatment at the time of utterance, or maltreatment at some indefinite time at which ‘‘an individual becomes aware.’’ It is only in the latter case that the binding of the temporal variable becomes relevant, for then the sentence could be 18

http://www.ispca.ie/content/legal.html http://www.irlgov.ie/debates/1jul98/sect11.htm

19

HAVE YOU NOTICED . . .

79

paraphrased as ‘‘if for some time t an individual becomes aware at t that pets kept for sale are being maltreated at t, what can an individual do at (or after) t?’’ Global accommodation of ‘‘pets kept for sale are being maltreated at t’’ would be blocked on pain of unbinding t. A similar analysis could be given for (31). In these cases, the example remains problematic for Gazdar’s analysis, which does not include any trapping constraint, but, with the dependent temporal variable interpretation, is not a counterexample to van der Sandt’s theory. What of the other temporal interpretation? If the factive complement refers to the utterance time, van der Sandt’s account makes exactly the same prediction as Gazdar’s, namely that projection/global accommodation will occur. This is clearly an incorrect prediction. In the remainder of this section many more cases of cancellation of a third person cognitive factive complement will be presented. In a few cases it may be arguable that there is a bound variable reading as in the example above, but even in these cases a non-bound variable reading is available, and cancellation appears to be independent of which reading is selected. Thus the data in this section counter-exemplifies the Gazdar and van der Sandt models equally, and, as will be discussed, much of it is also problematic for Stalnaker’s account. The following six examples are all cases where the writer explicitly indicates that a person who comes to be in possession of information should pass it on. The implication in all cases is that the writer does not yet have the information, so it is clear informally why cancellation occurs. But the formal models we have considered, Gazdar’s and van der Sandt’s, rely on clausal implicatures to produce cancellation in the first and second person cases we considered above. For these third person cases, the clausal implicature is not enough: (32) Third person, cognitive factive in conditional, cancellation: They also work well to deter rabbits & foxes from digging into the chook-pen (Hen-run). Dig a shallow trench the width of a single mattress, then place the springs flat in the trench. Drive your fence posts in the mid-line, so half the spring is outside & half inside the pen. I haven’t tried this with wombats, though, & if anyone discovers that the method is also wombat-proof, I’d really like to know! Radical Recycling! 20

20

http://www.merlin.net.au/Barachne/mattress.html

80

DAVID BEAVER

(33) Third person, cognitive factive in conditional, cancellation: If anyone discovers that Cook-n-Stirs are available to the US market please let the list know. Marlan Green, Electric paste pot!? 21 (34) Third person, cognitive factive in conditional, cancellation: Whether it’s still available seems unlikely. This is a British book, part of ‘The New ‘‘GILCRAFT’’ series – Number Two’. The publisher is C Arthur Pearson Ltd., Tower House, Southhampton St, Strand London. If anyone discovers that this book IS still available, please contact me at the above address. Bill Nelson, Games (FAQ 11), rec.scouting.22 (35) Third person, cognitive factive in conditional, cancellation: If anyone discovers that one of our volunteers is charging money for being a volunteer, please notify me ASAP. Tom Elliott, GenWeb, Waldo County, Maine, November 30, 2000.23 (36) Third person, cognitive factive in conditional, cancellation: LICENSEE shall notify UNIVERSITY if LICENSEE becomes aware that this Agreement is subject to any US or foreign government reporting or approval requirement. Contractual document, University of Arizona.24 (37) Third person, cognitive factive in conditional, cancellation: If Member becomes aware that a Project Deliverable may require the use of a patented invention then Member will provide notice to this effect to NACHA within thirty (30) days. Patent policy excerpted from and based upon American national standards institute (ansi) procedures for the development and coordination of American National Standard.25 21

http://palimpsest.stanford.edu/byform/mailing-lists/ bookarts/2000/01/msg00217.html 22 http://www.faqs.org/faqs/scouting/games/part1/ 23 http://www.rootsweb.com/Bmewaldo/newstuff.htm 24 http://www.ott.arizona.edu/acrobatpdfs/17-Miscellaneous.pdf 25 http://www.project-action.org/EXHIBIT_AB.doc

HAVE YOU NOTICED . . .

81

The next three examples involve information that might be obtained by scientists or detectives: (38) Third person, cognitive factive in conditional, cancellation: Example: If you believe, as I do, that ‘racial’ discrimination is wrong, you might be tempted (as I have) to claim that discrimination is morally wrong because it is scientifically wrong. That is, one mustn’t discriminate on the basis of race because there are, in fact, no real differences between people of different races (and besides, ‘race’ isn’t even a valid scientific category). But, what if scientists discover that there are in fact differences between the ‘races’. Would that mean that discrimination is now OK? Edward H. Hagen, ‘‘More thoughts on Evolutionary Psychology and political (in)correctness.’’ The Evolutionary Psychology FAQ, Institute for Theoretical Biology, Berlin.26 (39) Third person, cognitive factive in conditional, cancellation: If scientists discover that worms with ultra-long life spans are metabolically dynamic and not just hibernating in supersuspended animation, they could then attempt to induce similarly efficient metabolic activity, or a dauer stage, in humans. Kyle Roderick, Superworm May Hold Key to Human Longevity.27 (40) Third person, cognitive factive in conditional, cancellation: If detectives discover that those who investigated the shooting ignored evidence or otherwise acted improperly in an effort to support the officers’ version of events, it would move the scandal beyond the Rampart Division and squarely into Parker Center. Matt Lait and Scott Glover, LAPD Task Force Probes ‘95 New Year’s Eve Shooting, Los Angeles Times, February 3, 2000. As regards (38)–(40), it is obvious where the theories of Gazdar and van der Sandt go wrong. While it does not follow as a matter of logic that because scientists do not know something, I also do not know it, in many areas of knowledge this is commonly regarded as a strong default. Similarly, my ignorance of a crime does not follow logically from that 26

http://www.anth.ucsb.edu/projects/human/epfaq/galileo.html www.thirdage.com/news/archive/980808-02.html

27

82

DAVID BEAVER

of detectives. But if we add the additional conversational assumption that the authors of (40) are not publicly implicating that they have information which they have not yet provided to the police, then we can deduce that the factive presupposition must be cancelled. Although Stalnaker does not present any third person examples involving cancellation, it might be argued that all the examples presented so far in this section support his general strategy, for they show that complex conversational reasoning is relevant to presupposition cancellation. The crucial difference between the formal models of Gazdar/van der Sandt and Stalnaker’s model is that the formal models rely on generalized conversational implicatures, where Stalnaker’s model does not. If we take Stalnaker’s account to be essentially Gricean, then what this type of example shows is that quite often the right predictions must take particularized conversational implicatures into account (i.e., implicatures particular to a specific occasion of use, rather than depending on a general property like the presence of a lexical scale). Once we move from a theory in which generalized implicatures determine presupposition cancellation to an account in which particularized implicatures are needed, we are on a slippery slope, for we lack any predictive theory of particularized implicatures. Worse still, it seems unlikely that particularized implicature, as understood by Grice (1989), is even up to the job. Particularized implicature rests on the speaker having signaled in some way that a special interpretation is needed, for example, by flouting a conversational maxim. But in many of the following 17 additional cases of cancellation involving embedded third person cognitive factives, there is no clear signal (trigger) of this sort. In many cases, one is faced as a reader with a choice of whether to cancel the presupposition, and one’s decision appears to be based on no more than the question: is it more plausible that the speaker is presupposing the complement, or that the speaker is not? This, at least, is the case for the written form: at the end of the paper I shall argue that a hearer, as opposed to a reader, is faced with an easier decision. (41) Third person, cognitive factive under negation, cancellation: 160.3 In response to a question by Judge Hunton whether the Pulaski County lawsuit had settled; Dale Evans stated that they did a claim, but he is not aware that it ever went into a lawsuit. Karen M. Beeks Court Secretary, Minutes of the special meeting of the Washington County Quorum Court, May 6, 1999.28 28

http://nw-ar.com/minutes/1999/wcqc050699.html

HAVE YOU NOTICED . . .

83

(42) Third person, cognitive factive under negation, cancellation: Mr. Wynn asked if PennDot’s review letters were a reaction to a submission Home Depot made with these changes on it, which generated these review comments. Mr. Furnacola believes that there is written verification from PennDot requesting the change. Mr. Wynn is not aware that PennDot has ever dictated access on a Township roadway, since it is not in their jurisdiction. Lynda Seimes, Township Secretary, Hilltown Township Planning Commission, February 18, 2002.29 (43) Third person, cognitive factive under negation, cancellation: Mrs London is not aware that there have ever been signs erected to stop use of the route, nor that there has ever been any obstruction to stop use of the route. County Environment Director, Definitive Map Review 1996/ 2000, Public Rights of Way Committee, Parish of Aveton Gifford, 2000.30 (44) Third person, cognitive factive in conditional, cancellation: A subsidiary advantage is that I can’t make any exceptions for anyone without getting into trouble with everyone else. I really can’t, because if anyone discovers that an exception had been made my reputation for fair dealing would be blown away. Richard Kostelanetz, Anthologies on anthologies, 1999.31 (45) Third person, cognitive factive in conditional, cancellation: If anyone becomes aware that a chemical is being used in a new way or there is new information about its potential hazards, NICNAS needs to be advised and a secondary notification may have to be made. Australian government document.32

29

http://www.hilltown.org/Minutes/PCfeb2002.pdf http://atschool.eduweb.co.uk/bheber1/gnvq-ict/External/DCC/ reports/ed0094hq.html 31 http://www.richardkostelanetz.com/retro6.html 32 http://www.nicnas.gov.au/obligations/compliance/pdf/ compliance.pdf 30

84

DAVID BEAVER

(46) Third person, cognitive factive in conditional, cancellation: 3. If Susan knows that her eyes are dark brown, then A) She believes her eyes are dark brown B) Her belief that her eyes are dark brown is justified C) This belief is true D) All of the above E) None of the above. Stefan Sencerz, Introduction to Philosophy (Phil 1301), TEXAS A&M, 2001.33 (47) Third person, cognitive factive in conditional consequent, cancellation: Investigation method (1/2) 1. When observer gets RESET_ NOTIFICATION, it should check if the reset occurred at owner’s local bus. 2. If it occurred, the observer should wait for notification from the owner. 3. Then, if the observer can get notification from the owner, the observer knows that the owner has still stayed in the net. 4. If not, the observer knows that the owner has already disappeared since the reset occurred. P1394.1 working group, Isochronous owner observation, Proposal for March 19 and 20, Sony corporation.34 (48) Third person, cognitive factive in conditional, cancellation: In the 1980s jazz composers were seen as the new innovators, relegating improvisers to secondary status. If one accepts the fact that composition was the key element in jazz in the 1980s, however, the central figure of the decade was clearly Duke Ellington, although he died in 1974. Francis Davis, Online summary of ‘‘Large-Scale Jazz,’’ Atlantic Monthly, August 1987. (49) Third person, cognitive factive in conditional, cancellation: If one accepts the fact that the Wayfarer Songs existed around 1884 only in a piano version, the conclusion would present itself that the instrumental song inclusion in the Symphony No. 1 represents the first orchestra version of the song sections.

33

http://www.tamucc.edu/Bsencerz/exercise4.htm http://grouper.ieee.org/groups/1394/1/Documents/br016r00.pdf

34

HAVE YOU NOTICED . . .

85

Mathias Hansen, Concluding Remarks, Urtext edition of Gustav Mahler: Lieder eines fahrenden Gesellen, Peters Music Publishers, Berlin, 1980.35 (50) Third person, cognitive factive under modal, cancellation: Whatever the reason, she knows she can’t just bounce back, and she decides to leave. For others, the decision is based purely on fear. Perhaps she knows that if she were to stay another day, her life or the lives of her children would be gone. Dorthy Stucky Halley, When violence hits home, Information Guide for Abused Women in Kansas, Hubris Communications, June 1, 1998. (51) Third person, cognitive factive under modal, cancellation: Perhaps God knows that we will never reach the stars and so they are, for now at least, just expressions of His creativeness; just as paintings are but expressions of the artist’s creativeness. The Skeptic.36 (52) Third person, cognitive factive under modal, cancellation: What I mean is, suppose the query is ?[x, y] ( . . . ), and suppose that the querying engine has found, say, [[x/a y/a] [x/b y/c]]. Has it finished looking? Well, there are several senses. Maybe it knows that there are no other bindings of y that would be paired with the x/a binding, for example, but it hasn’t yet finished checking all the x bindings. Or perhaps it knows that there are no other x bindings that would pair with the y/c binding, but it hasnt finished checking other y bindings, and its not yet sure about all the x bindings for the case y/a. Pat Hayes, e-mail to Richard Fikes, [email protected] Mail Archives, October 25, 2001.

35

http://www.edition-peters.com/urtext/mahler/gesellen_lieder/ nachwort_engl.html 36 http://www.angelfire.com/va/theSkeptic/feedback.html

86

DAVID BEAVER

(53) Third person, cognitive factive under modal, cancellation: What’s this! It seemed obvious for him to continue trumps, but perhaps he knows that diamonds are running and is trying to set up a trump spot for partner. I don’t believe this, but you never know. Jeff Goldsmith, Partnercide, August 4, 1997.37 (54) Third person, cognitive factive under modal, cancellation: Perhaps she knows that she really is to be married, and she, too, is now sad at the end of childhood. Andrew Moore, Studying Doris Lessing’s Flight, 2001.38 (55) Third person, cognitive factive under modal, cancellation: What the heck was X talking about in the parking garage? Was he referring to the Colonists, Rebels or the Evil! Consortium members when he says that Mulder was going to lead them there? Is this a throwaway line used to set up the much more important idea of Mulder not having what it takes? Or perhaps X is aware that the Incorporeals had done something to Scully and he didn’t want that knowledge to fall into the Colonist’s (or more likely, Rebel) hands? Michelle Bush, One Breath, 2002.39 (56) Ironic third person, cognitive factive under modal, cancellation: The first line of this tablet speaks of the gathering of Mycenologists which is to take place in Texas. It reads, ‘‘mu-ke-no-ro-ko | te-ka-sa-de | i-jo-te.’’ This can only be translated as ‘‘The Mycenologoi going to Texas.’’ Interestingly enough, the remainder of the text is identical to that of PY An 1. It is possible that Hand 1 at Pylos was using the text of An 1, a list of rowers going to Pleuron, as a template for this tablet. We cannot be certain why he stopped this list after only one line. Perhaps he realized that the number of attendees was too great. 37

http://www.gg.caltech.edu/Bjeff/html/partner.html http://www.shunsley.eril.net/armoore/gcse/flight.htm 39 http://www.geocities.com/mbush1us/TheEpisodes_SeasonTwo_ OneBreath.htm 38

HAVE YOU NOTICED . . .

87

Or, better still, perhaps he realized that none of the attendees were coming from the region of Pylos. Kevin Pluta, NEWS FLASH!!, The 11th Mycenological Colloquium, as foretold by Hand 1, November 2000.40 (57) Third person, cognitive factive under modal, cancellation: Perhaps she realized that she was unable to write anything better than ‘‘Goblin Market,’’ or perhaps her ‘‘failure’’ to surpass herself is explained by her turn away from poetry to children’s stories and religious materials. David Cody, Christina Rossetti’s Literary Career.41

When Presuppositions Win So far all the naturally occurring examples we have looked at were cases where a factive presupposition is cancelled, and we considered both cases where standard accounts predict the cancellation, and cases where they do not. Now we will consider cases where standard accounts predict cancellation, but in actuality the factive presupposition is projected. In many examples, projection of the presupposition produces a marked, one is tempted to say ‘‘weird’’, effect. This is strongest in cases involving embedding under negation. If a sentence implies a speaker’s knowledge of f, and the negation of that sentence is uttered, we expect to be able to conclude that the speaker does not take f to hold. However, when a first person factive presupposition projects out of negation, this is not the case: (58) First person cognitive factive under negation, projection: I do not realize that I am in Wisconsin, as this is a strange place to me. Attributed to M. Enos, Historical society of the Upper Mojave Desert, Vol. 14 No. 2, February 16, 1999.42

40

http://www.utexas.edu/research/pasp/cipem/tablet.html http://65.107.211.206/crossetti/rossetti5.html 42 www.ridgenet.net/Bbrucew/Feb99.html 41

88

DAVID BEAVER

(59) First person cognitive factive under negation, projection: Q: How did you find out you had lost some sight? You would think I’d know, given I’m an optometrist. But because of the stroke, I am not aware that I don’t see everything. My wife Robyn noticed that I left food on the left side of the dinner plate. I was still hungry but I thought I had finished the meal. ‘‘John,’’ Recovery after a stroke, Guide Dog Association of New South Wales and A.C.T., 1997–1999.43 (60) First person cognitive factive under negation, projection: The sun says, ‘‘Light is my nature. What else but light could there be in me? I am not aware that I am shedding light. For me, to be is to shine. I am not aware of the strain of giving light. I do not feel that I am doing anything.’’ Acharya Vinoba Bhave, ‘‘The two aspects of Akarma – Yoga and Sannyasa,’’ Discourses On Gita.44 The above three cases seem to involve multiple identities of the speaker. ‘‘I’’ the speaker has a role like an omniscient narrator, and can have a different set of beliefs from ‘‘I’’ the person (or object) experiencing (or seeming to experience) the events described as they occur. While it is certainly true that the presuppositional theories we have looked at provide no way of correctly predicting what should happen to presuppositions in these cases, that is hardly surprising. For, to my knowledge, there is no current predictive theory of how what we might term split ‘‘I’’ sentences like those above are interpreted. We can find the split ‘‘I’’ effect in sentences involving other embeddings than negation. In the following two examples, with cognitive factives in hypothetical contexts, the split ‘‘I’’ is explicit. In (61) (for which, as with some other examples in this section, both the cancellation and projection readings are possible), His Lordship is considering what he would think if he were a pharmacist. In (62), the writer is considering what it is like to watch The Godfather critically, as if ignoring impressions and observations from previous screenings.

43

http://www.guidedogs.com.au/vision/interview_john.htm http://www.hindubooks.org/vinoba/gita/the_two_aspects_of_ akarma/page8.htm 44

HAVE YOU NOTICED . . .

89

(61) First person cognitive factive in conditional, projection: In the Explanatory Notes to Clause 41 there is the interesting statement in paragraph 201 that ‘‘NHS contracts are not normally enforceable in the Courts.’’ [ . . . ] If I were standing in the shoes of a pharmacist contemplating making such an arrangement, I would stop to reflect. If I am aware that the contract on offer cannot be legally enforced, and if, furthermore, as we see from Clause 40, it is a contract that can be brought to an end at the behest of the Secretary of State acting alone, I should seriously question the wisdom of going any further. Lord Hunt of Kings Heath, Lord’s Hansard, Column 164 April 24, 2001. [cancellation reading also available.] (62) First person cognitive factive under modal, projection: If I simply watch, what structural features of The Godfather catch my eye? The first time through, I might notice that the extended, halfhour wedding sequence functions as a prologue, but has little plot connection with the rest of the movie. I might notice that, in terms of plot, the rest of the movie traces the repercussions of a single event that’s announced at the end of the wedding sequence and occurs just after Tom returns from Las Vegas – the Don’s meeting with Sollozzo. William P. Coleman, The Godfather, The Screenplay Review 1:2, October 2000.45 In some other cases of first person cognitive factives in hypothetical contexts, it is harder to justify the availability of projection readings on the basis of split ‘‘I’’: (63) First person cognitive factive under modal, projection: Maybe I know that he’s been a-cheatin’ Maybe I know that he’s been untrue But what can I do Jeff Barry and Ellie Greenwich ‘‘Maybe I Know,’’ Long Tall weekend, They Might Be Giants, Emusic, 1999. 45

http://screenplayreview.home.mindspring.com/index.html

90

DAVID BEAVER

(64) First person cognitive factive under conditionalized modal, projection: SBut the assertions are merely assertions about syntactic structures. Ah, no; wait. You know that, and maybe I know that, if we were involved in making those assertions. But nothing in the RDF says this. Pat Hayes, e-mail to Sandro Hawke, [email protected] e-mail archive, January 15, 2002.46 (65) First person cognitive factive under modal, projection: Jonah was the kind of guy that wanted to be in charge of his own destiny – he hadn’t learned to submit. He actually thought he could change God’s plans! He thought that if he ran to Tarshish, he could stop the flow of God’s grace. Maybe I’m smarter than that – maybe I know that I can’t outrun God or stop Him from doing His will, but sometimes, I’ll have to admit, I’m tempted to take credit for what God does when I do submit. Really, what’s the difference? That’s just watching him pitch, but taking credit for the throw. Pastor Jim Wilson, Games People Play, Lighthouse Baptist Church, Seaside, CA 3-5-2000.47 (66) First person cognitive factive under modal, projection: I might know that the feeling of Love is the result of endorphins in my brain. This scientific fact came as the result of empirical study, an outgrowth of the dialectic. However . . . I experience it as LOVE . . . boundless, loss of the self into another, dripping with poetry and rose oil. Deean Kett, Jedi Knight, Faculty – The Jedi Academy, e-mail to Master Kan Kage, 6-28-2001.48

46

http://lists.w3.org/Archives/Public/www-archive/2002Jan/ 0069.html 47 http://www.freshministry.org/030500.html 48 http://pub88.ezboard.com/fjediacademyfrm24.showMessage? topicID ¼ 18.topic

HAVE YOU NOTICED . . .

91

(67) First person cognitive factive under modal, projection: Dear Femstat, I was watching TV, I believe the show was Seinfeld, I can’t remember. But I do remember that your ad came on. It said that I ‘‘might notice that other yeast infection medicines take 7 days . . . ’’ but you will cure mine in only 3. I MIGHT notice that, but I probably WON’T since I am a GUY. Now don’t get me wrong, I am not saying that you should limit your ads to ‘‘Mad About You’’ and other female shows. But if you are going to advertise on coed programs, acknowledge that out of millions of viewers, many of them are men, and do not need to be addressed. You can mention all of your selling points without talking as if all of America currently has a yeast infection. Rich Mackin, ‘‘Letter to Femstat,’’ in Dear Mr. Mackin . . . , Gorsky Press, Florida, 2001. As regards second person cognitive factives in questions, it is easy to find cases where the factive presupposition projects, and with absolutely no feeling of oddity. This is seen in all of examples (68)–(72): (68) Second person cognitive factive in question, projection: Have you discovered that Pro-Tools can play and record at half speed? This feature makes it easy to do a quick full octave pitch up or down with the aid of an outboard DAT (or other) recorder. How to Do Some Quick and Dirty Sound Design Without Getting Any Dirt Under Your Fingernails, The Motion Picture Editors Guild Newsletter Vol. 16, No. 6, November/December 1995. (69) Second person cognitive factive in question, projection: Have you discovered that a web site can enhance your business through increased sales and exposure? Let Lone Elk Projects assist you in the design and layout of your new web site. Lone Elk projects.49

49

http://www.angelfire.com/falcon/loneelk/

92

DAVID BEAVER

(70) Second person cognitive factive in question, projection: Have you discovered that DEL. is the same as DEL *.* or that DEL DOS will delete all the files in your DOS directory? Go on, try it if you are feeling suicidal. If you are not now feeling selfdestructive you will, if you do try it. Tom Coleman, Self-Assessment, PC Update: the magazine of Melbourne PC User Group, Melbourne, April 1994. (71) Second person cognitive factive in question, projection: Did you discover that this site is filled with ‘‘hidden’’ pages? Loro’s Guestbook.50 (72) Second person cognitive factive in question, projection: As the great-grandson of a Lithuanian tavern owner I offer one final question for you to ponder: ‘‘Did you ever realize that your ‘‘mouse pad’’ makes a great beer coaster?’’ J.R. Zane, Roll out the barrel, July 20, 1998.51 If in some cases of second person factives in questions the presupposition is cancelled, and in other cases the presupposition projects, then what is the status of Stalnaker’s ‘explanation’ of the cancellation cases? Yes, sometimes the questions are rhetorical, so that Stalnaker’s argument does not apply. But in other cases, I can see no feature of the example that would clearly exempt it from Stalnaker’s argumentation.

Written vs. Spoken Form Here is an intuition, based on looking at hundreds of naturally occurring examples: I doubt that there is any general principle that would enable one to predict from the written form of an arbitrary sentence involving a cognitive factive whether the factive complement is presupposed by the author. Certainly, there is a tendency for the complement to be presupposed. And certainly there are types

50

http://members.aol.com/loro/guestbook.html http://www.lithuaniangenealogy.org/pa/tales/brew.html

51

HAVE YOU NOTICED . . .

93

of sentence involving cognitive factives, notably in the first and second person, for which the complement is rarely if ever presupposed. But the gray area, the range of cases for which no small set of formal features of the text would tell you whether the complement is presupposed or not, is just too big. Most often, it is reasoning about the author’s knowledge and intentions, combined with an estimate of the inherent plausibility of the factive complement being true and known to be true, that tilts the reader’s opinion about what is presupposed one way or the other. However, I suggest that the situation is subtly different for the spoken form. Choose one of the above 76 examples involving a cognitive factive. Make it an example in which the factive complement is reasonably short, and where it is not too obvious from the content of the complement alone (i.e., without the rest of the utterance) whether the speaker would believe it or whether it is true. Do I hear (18)? Professor Berkovitz’s fearful pronouncement is repeated in (73a) along with some suggestive focus marking, and variants in (73c-d) with alternative placement of focus and alternative choice of factive subject: If I discover that your work is [plagiarized]F, I will be [forced to notify the Dean]F. b. If I [discover]F that your work is plagiarized, I will be [forced to notify the Dean]F. c. If the T.A. discovers that your work is [plagiarized]F, I will be [forced to notify the Dean]F. d. If the T.A. [discovers]F that your work is plagiarized, I will be [forced to notify the Dean]F.

(73) a.

Example (73a) suggests the professor has a suspicious mind, but that the student may be innocent. In contrast, (73b) suggests that the student is innocent, but that the professor is prepared to act as if she had not made the discovery, a split ‘‘I’’ case, we might say, or else the professor is mad. (73c) does not imply that the student is guilty. And finally, (73d) conjures up an image of complicity between the all-knowing professor and the guilty student. You may find this disturbing. While looking at naturally occurring examples, I came across a number of cases for which I was not sure whether the factive complement was being presupposed, or only came to the conclusion given above after scrutinizing a considerable portion of the text either side of the example. Here is a case where the intention of the original

94

DAVID BEAVER

writer remained unclear to me for some time, although I now favor a cancellation reading: (74) Second person cognitive factive in question: Incidentally, Jeff, how did you program the synth? Did you discover that the user interface was the hardest part? Rev. Bob ‘‘Bob’’ Crispen, e-mail to Jeff Harrington EMUSIC-L Digest, August 12, 1993 08:51:09.52 If (74) is read with stress on ‘‘discover,’’ one gets the impression that Bob is taking it for granted that the user interface was the hardest part, presumably because this is a common experience. But if stress is on ‘‘user interface’’ and/or ‘‘hardest part’’, this is not taken for granted. Certainly, Bob must think it plausible that the user interface was the hardest part, or else it would be odd to even raise the issue, but he need not be presupposing that it was the hardest. Jeff’s actual reply was ‘‘You talkin’ about that thing I made in high school? It had 16 pots for each harmonic and an electric organ keyboard for its controller. Volume and overall tuning was about it . . . .’’ He goes on to explain that he saved up for it working at Baskin and Robbins. Judging by the off-hand, ‘‘You talkin’ about’’, the simple unadorned statement of the facts about how the user interface was constructed, the minimizing ‘‘was about it . . . ’’, and the unpretentious approach to fund-raising, it seems that Jeff does not necessarily agree that the user interface was the hardest part, and presumably he did not take Bob to be presupposing that it was. But whether Bob had actually been taking it for granted, we will never know. Looking back at Karttunen’s examples, we can see that intonation also affects interpretation. If ‘‘realize’’ is stressed in ‘‘Did you realize that you had not told the truth?’’ (1b), the presupposition projects. But if ‘‘truth’’ is stressed, both presuppositional and non-presuppositional readings are available. In some cases certain choices of intonation produce infelicity rather than alternative readings. If ‘‘realize’’ is stressed in (3) ‘‘If I realize later that I have not told the truth, I will confess it to everyone,’’ the effect is quite bizarre: the speaker knows that he or she hasn’t told the truth, but has not realized this. Similarly, if General Myers had stressed ‘‘aware’’ in (8) ‘‘I am not aware that there will be 52 http://www.ibiblio.org/emusic-l/back-issues/vol056/ issue17.txt

HAVE YOU NOTICED . . .

95

Russian forces here.’’ the audience would have concluded that he was losing his grip. For he would have presupposed something of which he was denying knowledge. What conclusion can we draw from the plagiarism paradigm in (73) and these additional examples? The implication is clear: the crucial factor determining projection behavior is not the choice between a first or third person subject, but the choice between an accented or deaccented propositional complement. When I commented at the start of the paper that we might just be watching the ‘‘wrong fight,’’ what I meant is that perhaps it is not the interaction between presupposition and implicature which we should be looking at, but the interaction between presupposition and information structure. Some observations. First, when the factive complement is long, it is likely to contain some pitch accents. For the moment I make no claims about whether presupposition has any intonational correlates in the case of long factive complements. Second, some complements cannot easily be understood presuppositionally. For example, if a factive is embedded under negation and the complement contains a negative polarity item, as in for example Mill’s (5), then that will provide an independent reason (reminiscent of van der Sandt’s trapping constraint) for why local accommodation/cancellation might be preferred over global accommodation/projection. Third, the cases where projection is mandatory involve accent on the factive verb itself. But if the factive verb is being contrasted with a non-presuppositional expression, there is no presupposition, as in ‘‘She doesn’t [know]F it, she merely [believes]F it.’’ Whenever a philosopher stresses a factive verb, be it Mill, Rousseau, or Chuang Tzu, one has a tendency to wonder whether the alternatives under consideration include non-factive verbs. The question of whether such contrastive accents are distinct from foci marking new information remains controversial (Krahmer and Swerts, 2001). For limitations of time and space, I will desist from a detailed consideration of the interpretation of intonational and textual variants of each of the examples I have considered in this paper. I will also desist from a detailed theoretical analysis and comparison with prior work.

Further Work I should like to mention some lines of research which I hope to take up in connection with the data I have presented.

96

DAVID BEAVER

There have been a number of papers recently on the pragmatics of presupposition, developments of Stalnaker’s approach which are far from those of Gazdar and van der Sandt. In particular I am thinking of work of Abusch (2002) and Simons (2001, 2003, 2004). The types of explanation offered by Abusch and Simons are highly suggestive, although neither author uses naturally occurring data and neither author considers the effects of intonation. Abusch’s very original use of alternatives seems to me to be particularly ripe for adaption to the effects considered in this last section. Where Abusch uses constraints on the alternative set provided by the lexicon, I would want to also consider constraints on the alternative set provided by focus. This last move would be uncontroversial, given that focus is the best-known source of constraints on the alternative set (Rooth, 1992). Zeevat (2002), which considers the significance of oldness of information to choice of presuppositional expression, and speculates about the relevance of expression alternatives, is also relevant. Another line of inquiry centers on the interaction between topic/focus and presupposition. Strawson (1964) started the ball rolling, suggesting that definites only carried a presupposition when they were in topic (subject) position. Reinhart (1982) and Lappin and Reinhart (1988), among others, have carried this idea further, applying a variant of it to presuppositions of quantifiers. Meanwhile ´ (1984) provided several interesting examples of cancellation Hajicˇova (or, in her terminology, ‘‘allegation’’ rather than ‘‘presupposition’’) arising when presuppositional constructions are focussed. These are further discussed by Hajicˇova´ et al. (1998) and Partee (1996), and the latter provides a way of thinking of these cases in terms of accommodation. The other important strand of recent work I should like to mention is the recent empirical research of Spenader. Spenader (2001) studies naturally occurring factives in the London-Lund corpus, and shows that factives are commonly (more than half the time) used when the factive complement is not previously established to be true. Most of the time factive presuppositions have to be accommodated, whether globally or locally. This result seems very much in accord with two observations I have made in this paper, both of which suggest a weakening of the generalization that factive complements are normally presupposed. First, even third person uses of factives do not imply that the speaker is taking the factive complement for granted. Second, for spoken utterances the presupposition only seems to fully kick in when the factive complement is deaccented. I leave discussion of the more detailed results of Spenader (2002) for another occasion.

HAVE YOU NOTICED . . .

97

The title of this paper, as you may have guessed, is taken from naturally occurring text. You may have wondered whether the factive complement ‘‘your Belly Button Lint color is related to the color of your clothing’’ was presupposed by the author. Or, by now, you may have realized (!) that there are two ways of understanding the example. If you read the title aloud, and stress ‘‘noticed,’’ then it seems presupposed that belly button lint color comes from clothes. But if you do not accent ‘‘notice,’’ and place the main accent within the complement, then this is not necessarily presupposed. In fact, the question was taken from an online survey,53 and this survey was intended in an objective scientific spirit. The author definitely did not mean to take for granted anything about the life of belly button lint. Indeed, the results of the survey speak for themselves: although 37% of people have lint color related to their clothing color, ‘‘some people consistently have BBL in a color that is not present in their clothing.’’ Whether looking at the source of belly button lint or the source of presupposition, contemplating your own navel is not enough. Yet the majority of work on presupposition has strayed little further than a small circle of scholastic navels. It would be foolish to deny that my own navel has occasionally been an object of study. The only major line of work of any relevance that has been based on naturally occurring text is that of Prince (1981) and many followers. Yet while such work is relevant to the study of presupposition, it has generally been focussed on discourse structure and given-new, and has not had so much impact on study of the presupposition projection problem. The goal of this paper has not only been to study how presupposition projection and implicature interact, but also to suggest changes to the standard methodology. Thanks to modern technology, these changes can be made without us leaving the comfort of our armchairs. I do not want to suggest that there is no longer a place for introspective judgments of artificial examples. We are fortunate that as linguists we come equipped with our own languages, and rapid progress in the field is dependent on our taking advantage of this endowment to rapidly prototype theories and suggest new directions for empirical research. But if we stick entirely to artificial examples there is a danger that we will end up with artificial theories. I hope I have convinced at least some readers that the time is ripe for those interested in the study of presupposition to add additional tools to their empirical arsenal.

53 Dr. Karl, Survey Questions, ABC Science Online, Australia, http:// www.abc.net.au/science/k2/lint/why.htm

98

DAVID BEAVER Acknowledgments

I thank first Hans Kamp for inspiration, support and excellent company over the years. I am also grateful to participants at the workshop, and to the editors of the volume and publisher for their work on the manuscript.

References Abusch, D. 2002. Lexical alternatives as a source of pragmatic presuppositions. In B. Jackson, ed., Proceedings of SALT XII. Ithaca, NY: CLC Publications. Beaver, D. 1997. Presupposition. In J. van Benthem and A. ter Meulen, eds., The Handbook of Logic and Language, pp. 939–1008. Amsterdam: Elsevier. Beaver, D. 2001. Presupposition and Assertion in Dynamic Semantics. Stanford, CA: CSLI Publications. Beaver, D. 2002. Presupposition in DRT. In D. Beaver, L. Casillas, B. Clark, and S. Kaufmann, eds., The Construction of Meaning, pp. 23–43. Stanford, CA: CSLI Publications. Gazdar, G. 1979. Pragmatics: Implicature, Presupposition and Logical Form. New York: Academic Press. Geurts, B. 1999. Presuppositions and Pronouns. Vol. 3 of CRiSPI. Amsterdam: Elsevier. Grice, H. P. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press. ´, E. 1984. Presupposition and allegation revisited. Journal of Hajicˇova Pragmatics 8:155–167. Hajicˇova´, E., B. Partee, and P. Sgall. 1998. Topic-Focus Articulation, Tripartite Structures, and Semantic Content. Dordrecht: Kluwer. Heim, I. 1983. On the projection problem for presuppositions. In M. Barlow, D. Flickinger, and M. Westcoat, eds., Second Annual West Coast Conference on Formal Linguistics, pp. 114–126. Stanford, CA: Stanford University. Kamp, H. 2001a. Computation and justification of presuppositions. In M. Bras and L. Vieu, eds., Semantics and Pragmatics of Dialogue: Experimenting with Current Theories. Amsterdam: Elsevier. Kamp, H. 2001b. The importance of presupposition. In C. Rohrer and A. Rossdeutscher, eds., Linguistic Form and its Justification. Selected Papers from the SFB 340. Stanford, CA: CSLI Publications. Kamp, H. and A. Rossdeutscher. 1994. DRS-construction and lexically driven inference. Theoretical Linguistics 20:165–235. Karttunen, L. 1971. Some observations on factivity. Papers in Linguistics 5:55–69.

HAVE YOU NOTICED . . .

99

Kiparsky, P. and C. Kiparsky. 1970. ‘Fact’. In M. Bierwisch and K. Heidolph, eds., Progress in Linguistics, pp. 143–173. The Hague: Mouton. Reprinted in J. Peto¨fi and D. Franck, eds. 1973. Pra ¨suppositionen in Philosophie und Linguistik – Presuppositions in Philosophy and Linguistics, pp. 315–354. Frankfurt: Athana ¨um Verlag Krahmer, E. 1998. Presupposition and Anaphora. Stanford: CSLI Publications. Krahmer, E. and M. Swerts. 2001. On the alleged existence of contrastive accents. Speech Communication 34(4):391–405. Lappin, S. and T. Reinhart. 1988. Presuppositions of quantifiers: A processing account. Linguistics 26:1021–1037. Mercer, R. 1992. Default logic: Towards a common logical semantics for presupposition and entailment. Journal of Semantics 9:223–250. Partee, B. H. 1996. Allegation and local accommodation. In B. H. Partee and ´, P. Sgall, eds., Discourse and Meaning. Papers in Honor of Eva Hajicˇova pp. 65–86. Amsterdam: John Benjamins. Prince, E. 1981. Toward a taxonomy of given-new information. In P. Cole, ed., Radical Pragmatics, pp. 223–256. New York: Academic Press. Reinhart, T. 1982. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27:53–94. Rooth, M. 1992. A theory of focus interpretation. Natural Language Semantics 1:75–116. Simons, M. 2001. On the conversational basis of some presuppositions. In R. Hasting, B. Jackson, and Z. Zvolensky, eds., Proceedings of Semantics and Linguistic Theory, vol. 11, pp. 431–448. Ithaca, NY: CLC Publications, Cornell University. Simons, M. 2003. Presupposition and accommodation: Understanding the Stalnakerian picture. Philosophical Studies 112:251–278. Simons, M. 2004. Presupposition and relevance. In Z. G. Szabo´, ed., Semantics vs. Pragmatics, pp. 329–355. New York: Oxford University Press. Soames, S. 1982. How presuppositions are inherited: A solution to the projection problem. Linguistic Inquiry 13:483–545. Spenader, J. 2001. Presupposition or anaphora: Constraints on choice of factive complements in spoken discourse. In I. Kruijff-Korbayova´ and M. Steedman, eds., Information Structure, Discourse Structure and Discourse Semantics. Workshop Proceedings, pp. 161–174. Helsinki: 13th European Summer School in Logic, Language and Information. Spenader, J. 2002. Presuppositions in Spoken Discourse. Ph.D. thesis, Stockholm University, Sweden. Stalnaker, R. 1974. Pragmatic presuppositions. In M. Munitz and P. Unger, eds., Semantics and Philosophy, pp. 197–214. New York: New York University Press.

100

DAVID BEAVER

Strawson, P. 1964. Identifying reference and truth values. Theoria 30. van der Sandt, R. 1992. Presupposition projection as Anaphora Resolution. Journal of Semantics 9:333–377. Zeevat, H. 1992. Presupposition and accommodation in update semantics. Journal of Semantics 9:379–412. Zeevat, H. 2002. Old material. Unpublished ms, University of Amsterdam, Amsterdam.

4

Accommodation and Reaccommodation in Dialogue ROBIN COOPER

AND

STAFFAN LARSSON

Abstract The paper examines how accommodation of questions has been exploited in a dialogue system. Based on real dialogue examples we argue that there is a need for a local QUD (representing questions that are currently being addressed in the dialogue) and a global QUD representing (possibly among other things) questions that have been raised previously and that are available for reraising. In addition, the twin notions of reaccommodation and reraising are important for dialogue analysis. We present an implementation of a simplified version of reaccommodation and reraising. Consideration of how these notions should be implemented in its turn raises new theoretical questions to be explored.

Introduction In this paper1 we will first examine how the notion of accommodation has been exploited in some simple-minded dialogue systems that we have constructed using the TRINDIKIT dialogue system toolkit

1 The paper contains revised material from Cooper et al. (2000). More recent developments of this work are to be found in Larsson (2002).

Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

101

102

ROBIN COOPER

AND

STAFFAN LARSSON

(Larsson et al., 1999).2 The notion of accommodation used here goes a little beyond the normal notion of accommodation of presuppositions since it has largely to do with assumptions that dialogue participants have about what questions are under discussion. We will then look at some real dialogue examples which can be analysed as question accommodation, i.e. the dialogue participant adds a question to QUD (‘‘Questions Under Discussion’’) which has not been explicitly raised. We shall argue that there is a need for a local QUD (representing questions that are currently being addressed in the dialogue) and a global QUD representing (possibly among other things) questions that have been raised previously and that are available for reraising. We will suggest that the twin notions of reaccommodation and reraising are important for dialogue analysis. When a question is taken up a second time in a dialogue it can be introduced by a shorter or less explicit utterance than the first time it is introduced. Dialogue contributions of this type are similar to definite descriptions which have to provide just enough information to identify a unique referent from a context set. But the utterances we are considering are not by any means definite noun-phrases and what they are referring to are questions that have been under discussion previously during the dialogue. Finally, we will show how these considerations of real dialogue can feed back into the implementation of dialogue systems and present an implementation of a simplified version of reaccommodation and reraising. Consideration of how these notions should be implemented in its turn raises new theoretical questions to be explored.

Accommodation in the GoDiS System The GoDiS system (Traum et al., 1999; Engdahl et al., 2000) is a simple toy dialogue prototype for various domains; the examples in this paper are taken from the travel agent domain. It has a simple behaviour which of itself is not remarkable as dialogue systems go. The point is to reconstruct the behaviour of simple dialogue systems from a theoretically interesting perspective which allows an extremely modular approach to implementation. From a computational point of view this is interesting because the modular approach allows rapid porting of a dialogue system from one domain to another or from one language to another. In particular the rules for dialogue strategies are 2 See the TRINDIKIT homepage: www.ling.gu.se/research/projects/ trindi/trindikit.html

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

103

expressed in a separate module which can be maintained across such portings. GoDiS handles basic information seeking dialogues, i.e. dialogues where the user announces a task (such as booking a flight) and the system then asks a series of questions to set parameters (such as ‘‘Where do you want to fly from?’’, ‘‘Where do you want to fly to?’’, ‘‘When do you want to fly?’’ and so on). Once the system has gathered all the information, it can report information on a suitable flight. Systems that do this are often implemented using some kind of form-filling technique. The system has a number of slots which must be filled in. In the simplest systems the dialogue is driven by a finite state machine which takes the questions in order one at a time and ignores anything the user says that does not appear relevant to the question it has just asked. GoDiS pursues a dialogue strategy where it will try to interpret anything that the user says in terms of the questions that it has on its dialogue plan, even if it has not yet explicitly asked the question. It uses the notion of QUD taken from Ginzburg (1996a, b, 1998). If it notices that the user provides an answer to a question that the system has not asked it accommodates the question to the QUD and then proceeds to interpret the answer in the context of the question in the normal way. It is, of course, possible to get such behaviour in an ad hoc way even in a finite state-based system, by adding additional loops for taking questions in different orders or processing information not relevant to the question asked. But such a strategy would not be general – you might forget to program in some particular ordering – and it does not allow you to separate the dialogue strategy out from the questions associated with the particular domain. The GoDiS strategy allows you to specify the list of questions and expected answers and then the dialogue strategy (including question accommodation) will automatically allow the more complex behaviour. It is possible (although we have not done this) that the resulting system could be compiled to a dialogue system driven by a finite state automaton although it is an open question whether there would be any efficiency gains in doing this. What we have developed is a very simple theory which facilitates the construction of simple dialogue systems. The question we want to address in this paper is whether such a theory has relevance for the analysis of real language as one finds in actual recorded dialogues. Information States We present the basic ideas of question accommodation and some examples that we have presented in previous papers. Our approach is based on a view of dialogue analysis in terms of information states

104

ROBIN COOPER

AND

STAFFAN LARSSON

representing information about the dialogue that a dialogue participant has. Accommodation has to do with the nature of information state updates. The notion of information state we have proposed is basically a version of the dialogue game board which has been proposed by Ginzburg (1996a, b, 1998). In TRINDI deliverables such as Traum et al. (1999), Engdahl et al. (2000) we represent information states of a dialogue participant as a record of the type shown in Figure 1. The main division in the information state is between information which is private to the agent and that which is shared between the dialogue participants. The private part of the information state contains a PLAN field holding a dialogue plan, i.e. a list of dialogue actions that the agent wishes to carry out. The plan can be changed during the course of the conversation. The AGENDA field, on the other hand, contains the short-term goals or obligations that the agent has, i.e. what the agent is going to do next. We have included a field TMP that mirrors the shared fields. This field keeps track of shared information that has not yet been grounded, i.e. confirmed as having been understood by the other dialogue participant. The SHARED field is divided into three. One subfield is a set of propositions which the agent assumes for the sake of the conversation. The other subfield is for a stack of QUD. These are questions that have been raised and are currently under discussion in the dialogue. The LU field contains information about the latest utterance. This is a simple notion of information state which makes no pretensions to completeness. We are using it as a baseline for further extensions in implementations we are developing using the TRINDIKIT (Larsson et al., 1999), a toolbox which allows experimentation with different definitions of information state and update rules.

plan agenda bel private

: tmp

shared

:

com qud lu

: : :

OpenStack(Action) Stack(Action) Set(Prop) com : Set(Prop) qud : OpenStack(Question) : speaker : Participant lu : moves : AssocSet(Move,Bool) : Set(Prop) : OpenStack(Question) speaker : Participant : moves : AssocSet(Move,Bool) FIGURE 1 GoDiS information state.

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

105

The QUD here is a local QUD. That is, it represents questions that are under discussion in the current turn, treated as a stack. In our current analyses we rarely, if ever, have more than two questions on this stack and the cases where there are two questions are normally such that an answer to the topmost question on the stack will also be an answer to the question beneath it, e.g. oWhere does A wish to fly to, Does A wish to fly?W. In such a case a single answer will pop both questions off the QUD. In this paper we will argue that there should be at least one global QUD in addition to this and that local and global QUDs give rise to different dialogue behaviour. Our argument will be based on the nature of accommodation of questions to QUD.

Looking at Real Dialogues Question Accommodation Dialogue participants can address questions that have not been explicitly raised in the dialogue. In such cases, coherence is preserved if the agent is able to find a question which is relevant at that point in the dialogue which can then be accommodated onto the QUD. Here is an example3 ˚nad ska du ˚ (1) $J (travel agent): // vilken ma aka & what month do you want to travel? ¨:h tredje $P (customer): / ja: typ den: a ˚n ga ˚ng da ˚ billigt som ¨rde april / na ¨r / sa fja ¨jligt mo & yeah, about the, eh, third or fourth of April, around there. As cheap as possible. The point here is that the customer does more than answer the question that the travel agent has introduced onto QUD, namely what month does P (the customer) want to travel?. The additional information as cheap as possible is an ellipsis and addresses a different question, something like How much does P want to pay?. Note that this question, or something like it, has to be adduced in order to interpret the ellipsis. There is no mention of price in the previous part of the dialogue so this is not a matter of anaphoric interpretation. Rather the communication here is depending on the fact that both the 3 From a dialogue collected by the Lund group in the SDS project. We are quoting the transcription made in Go¨teborg by Jens Allwood’s group where it is identified as dialogue A820101. We have added some annotations of our own, such as the English translation, and removed some annotations that are not relevant to the point at hand.

106

ROBIN COOPER

AND

STAFFAN LARSSON

customer and the agent know that price is a relevant question when booking a flight. The strategy we adopt for interpreting elliptical utterances is to think of them as short answers (in the sense of Ginzburg) to questions on QUD. The use of question accommodation enables us to generalize the treatment to cases where the elliptical utterance is not an answer to an overt question. We have sometimes received the comment that it is unintuitive to consider the offering of additional information as answers to as yet unraised questions. Perhaps it is misleading to use the word ‘‘question’’ rather than ‘‘issue’’ here. The substitution of ‘‘issue’’ for ‘‘question’’ would be quite consistent with the way we are viewing QUD. Perhaps it sounds more intuitive to say that successful dialogues are those in which the dialogue participants can figure out what issues are being addressed rather than what questions are being answered and that a knowledge of what issues are being addressed can be essential for the interpretation of ellipses. Equally it seems more intuitive to say that the customer is raising the issue of price with her statement as cheap as possible rather than raising the question. We will nevertheless keep to the original terminology with QUD rather than changing it to the perhaps less happy abbreviation IUD (for ‘‘issues under discussion’’). QUD is after all something you can chew on . . . The substantive point here is that we do not at the moment see any reason to distinguish between issues and questions. However, we do see reasons to keep track of questions that have been raised and that may be raised again later in the dialogue. Let us consider another similar dialogue.4 The dialogue begins as in (2). (2) $B (travel agent): japp /// & yes? $A (customer): eh // jag undrar om eh // / / en resa till phuket // den tolfte trettonde december // & er, I wonder if . . . , er, a trip to Pukhet, 12th or 13th December. At the start of the conversation the travel agent does not have a plan for the dialogue. She may have several plans in the background for the kinds of tasks that she is capable of such as booking a flight, changing 4 From the same collection as the previous one, no. A821702 in the Go ¨teborg transcription.

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

107

the billing address on a booking, booking hotels, rental cars etc. It is not until the customer makes an initial contribution that the agent can form a plan. The initial contribution by the customer here seems to be fairly typical. It does not say exactly what the customer wants, although we as competent dialogue agents are so good at interpreting dialogues that it is not always obvious that this is the case. But we can test this by embedding the customer’s contribution in another constructed dialogue where it gets a different dialogue interpretation and would suggest rather different dialogue plans. Consider (3). (3) Husband: What would you like for your birthday, darling? Wife: er, I wonder if . . . , er, a trip to Pukhet, 12th or 13th December. Or (4) (4) Secretary: There’s a line on the travel agent invoice that I can’t match to any of your business trips Executive: er, I wonder if . . . , er, a trip to Pukhet, 12th or 13th December. Clearly, the activity (in something like Allwood’s sense (1995)) can make a big difference to the way in which we interpret utterances. This is because different issues are being addressed, i.e. different questions are under discussion. The travel agent, therefore, has to accommodate questions onto QUD in order to be able to interpret the customer’s contribution. She figures that this is an answer to questions like Where do you want to go?, When do you want to go? and that these are questions which are associated with the plan she has for booking trips and, on our analysis, she therefore loads this plan into her dialogue information state and thereby prepares to raise other relevant issues to the task. The dialogue continues5: (5) $B(travel agent): ja // & yes $A (customer): [1 fyra ]1 personer / & four people ¨r det ] 1 $B (travel agent) : [1 a & is it . . . B accommodates: How many people will travel? 5 The numbered brackets in the transcription indicate parts of the speakers’ contribution which occur at the same time.

108

ROBIN COOPER

AND

STAFFAN LARSSON

The travel agent has to accommodate the question How many people will travel? in order to be able to interpret four people. She interrupts her own speech in order to acknowledge the information as shown in (6) ¨r det charter du har (6) $B (travel agent): yes / a ¨r eller // sett da & Right . . . is it a charter you saw there or . . . $A (customer): [2 nþ ja: ]2 & well . . . (expressing doubt) $B (travel agent): [2 eller vad har ]2 ni ˚gonting / ¨nkt fo ¨r na ta & . . . or what kind of thing have you been thinking of? $A (customer): nej det det / jag tror att det ¨r dyrt med hotell utan eh / blir fo & no, it’s . . . it’s, I think it’ll be too expensive with a hotel but, er . . . B accommodates: Will it be too expensive for a hotel for this trip if it is booked together with the plane? At this point it seems that the travel agent has to accommodate the question indicated in (6) onto her QUD. (7) $B (travel agent): o / W / ja ni ska bara ha flyget / & right, you just want a flight? The question the customer raised is answered by the travel agent’s right in (7) and is thus popped off QUD. The agent continues by raising the new closely related question Does A just want a flight?. The customer acknowledges this question and answers it. (8) $A (customer): ja / & yes $B (travel agent): okej / [3 m: ]3 & okay, mm,..

Reaccommodation and Reraising At this point there is no question on QUD. But the customer is not entirely satisfied with the exchange and raises the second question again.

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

109

¨r ]3 / vi / tror inte (9) $A (customer): [3 ja fo ˚ det att det blir va ¨ldigt eh ho ¨ga du ocksa summor om man ska [4 // ]4 & yes, because . . . we . . . don’t you also think that it would be very, er, high costs if you’re . . . B has to reaccommodate the question: How expensive will it be for a hotel for this trip if it is booked together with the plane? There are two points we want to make about this utterance. Firstly, it is necessary to raise the question again. It cannot simply be addressed as a question already on QUD (i.e. local QUD). Neither the customer nor the agent could answer this question at this point in the conversation by saying Yes. But, secondly, the raising of the question has the flavour of a second mention in that it is more reduced than would have been possible with the original raising. If we move the customer’s utterance in (9) back in the conversation and substitute it for her last utterance in (6), where she first raises the question, it would not be successful. It would be incoherent, even if we remove the tell-tale also. We shall call her utterance in (9) a reraising of the question. Note that such a reraising can have limits on how reduced it can be. It would not have worked here if the customer had only said: Don’t you also think (so) . . . . There is not enough information in this utterance to identify the intended question among the other questions that have been under discussion. We therefore conclude that, in addition to local QUD, a global QUD is needed which keeps track of questions that have been under discussion. Reraisings have certain properties in common with definite descriptions in that they must provide at least enough information to uniquely identify the intended question among the restricted set of those on the global QUD.

Putting Reraising and Reaccommodation into Simple Dialogue Systems Larsson (2002) developed a more advanced version of GoDiS. This system includes simple versions of reraising and reaccommodation. When we consider how to implement reraising and reaccommodation cleanly in the information state-based approach we are using, there are different cases which will need a different implementation. We characterize these cases in (10).

110

ROBIN COOPER agenda plan bel private

: tmp

shared

:

nim com issues qud pm lu

AND

STAFFAN LARSSON

: : :

OpenQueue(Action) Stack(PlanConstruct) Set(Prop) com : set(Prop) issues : OpenStack(Question) : qud : OpenStack(Question) agenda : OpenQueue(Action) plan : Stack(PlanConstruct) : OpenQueue(Move) : Set(Prop) : OpenStack(Question) : OpenStack(Question) : OpenQueue(Move) speaker : Participant : move : Set(Move)

FIGURE 2 ISIS information state.

(10) 1. 2.

reraising open issues – reraising a question previously discussed but not yet resolved reraising closed issues – reraising a question which was previously discussed and resolved (a) revising and denying – changing the answer which resolved the question when it was last discussed. (There is also the possibility of denying the previous answer without providing a new one, thus making a closed issue back into an open one.6) (b) reraising of dependent questions – changing the answer to one question may mean that another closed issue whose resolution depends on the answer to that question must also be reraised.

In each of these cases it will also make a difference to the implementation whether it is the user or the system that is reraising the question. In order to implement this the ISIS system uses a slightly more complex information state than that in Figure 1. This is presented in Figure 2. 6 A further option, not explored in our implementation of the TA domain, is to add a new answer to a question which can be given several answers which do not contradict each other. For example, in a shopping domain the question ‘‘what item do you want to buy?’’ can be given several answers, each of which can be regarded as resolving the question by itself. For this reason, we regard reaccommodation and removal of incompatible information (retraction) as separate but related updates.

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

111

The main difference here is that we distinguish between a global and a local QUD. The local QUD is a set of questions that can be addressed using a short answer. The global QUD we use here is labelled ISSUES and contains all questions which have been raised in the dialogue (explicitly or implicitly) but have not yet been resolved. It thus contains a collection of open issues. A suitable data structure appears to be an open stack (a ‘‘stackset’’), i.e. a stack where nontopmost elements can be accessed. This allows a non-rigid modelling of current issues and task-related dialogue structure. Information states are operated on by update rules consisting of a list of conditions (‘‘preconditions’’) and a list of updates (‘‘effects’’). To this extent our update rules are similar to STRIPS operators (Fikes and Nilsson, 1971). The format we will use is given in (11). (11) rule : Rule name class : Rule name 8 Condition1 > > > < Condition2 pre : > ... > > : Conditionn 8 Update1 > > > > < Update2 eff : ... > > > > : Update

m

Rules are grouped into classes and there is a control algorithm which determines when the various classes of rules should fire. If the conditions hold when the rule is tried, the updates will be applied to the information state. The relations and operators available for formulating conditions and updates, respectively, are determined by the type of the information state. Thus if we consider the type of the ISIS information state represented in Figure 2, we find that the value of the path /SHARED/QUD will be an open stack (a stack all of whose elements are accessible). Thus the update operations associated with this path will be limited to those operations that can be performed on open stacks. When performing updates such as reraising which affect the structure of the dialogue, it is important to provide appropriate signals to the user to enable efficient coordination. Allwood (1995) uses the

112

ROBIN COOPER

AND

STAFFAN LARSSON

term Interactive Communication Management (ICM) for all communication dealing with the management of dialogue interaction. This includes feedback, sequencing and turn management. Sequencing ‘‘concerns the mechanisms, whereby a dialog is structured into sequences, subactivities, topics etc. . . . ’’. We regard ICM related to reraising and reaccommodation as sequencing ICM. We now give examples of each of the cases of reraising in (10) as implemented in the system and explain what update rules are used to achieve the appropriate reraising. We will simplify slightly the update rules which are generated from the code for reasons of exposition. Reraising Open Issues (12) is an example of a dialogue between a user ($U) and the system ($S), where the user reraises an issue which was previously raised but not resolved. Also subsequently the system reraises an issue which is relevant for addressing the issue that the user reraised. In order to address the issue of price the system needs to get information from the user concerning where she wishes to travel from. (12) $SW Welcome to the travel agency! $UW price information please $SW You asked about price. How do you want to travel? $UW a flight to paris $SW Okay. by flight. to paris. What city do you want to go from? $UW do i need a visa $SW You asked about visa. What country are you from? $UW actually, let’s talk about price again User reraise $SW You reraised the price issue. So, what city do you want to go from? System reraise The first case of reraising here, where it is the user who reraises the price issue, can be handled by a variant of the normal update rule for integrating user questions, with an additional condition that the question q asked by the user is already on ISSUES. It is useful for the system to indicate that it realizes that the issue has been discussed before. This can be done by producing reraising ICM (dialogue move icm : reraise :Q) by saying ‘‘You reraised the issue of price’’.

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

113

(13) rule : integrateUsrAsk class : integrate 8 > < fstð$/private/nim/; AÞ pre : A ¼ ousr; askðQÞ4 > : $domain :: planðQ; CÞ 8 delð/private/nim; AÞ > > > > > pushð/shared/qud; QÞ > > > > > pushð/private/agenda; icm : acc posÞ > > >  >  > > < pushð/private/agenda; icm : und pos : usr issueðQÞÞ eff : if then elseðinð$/shared/issues; QÞ; > > > ½raiseð/shared/issues; QÞ > > > > > pushð/private/agenda; icm : reraise :QÞ; > > > > > pushð/shared/issues; QÞÞ > > > : pushð/private/agenda; respondðQÞÞ This rule7 belongs to the class integrate which consists of those update rules which are used by the system to integrate the user’s moves into its information state. 7

The actual rule generated from the code is:

rule : integrateUsrAsk class : integrate 8 $/private/nim/fst ¼ A > > > > < A=fst ¼¼ usr pre : A=snd ¼ askðBÞ > > > > : $domain :: planðB; CÞ 8 delð/private/nim; AÞ > > > > > pushð/shared/qud; BÞ > > > > > > pushð/private/agenda; icm : acc posÞ > > > > > pushð/private/agenda; icm : und pos : usr issueðBÞÞ > < eff : if then elseðinð$/shared/issues; BÞ > > > ½raiseð/shared/issues; BÞ > > > > > pushð/private/agenda; icm : reraise :BÞ > > > > > pushð/shared/issues; ðBÞÞ > > > > : pushð/private/agenda; respondðBÞÞ

114

ROBIN COOPER

AND

STAFFAN LARSSON

The first condition checks the value of the path /PRIVATE/NIM/ (represented by the preceding ‘$’). NIM stands for ‘‘non-integrated moves’’ and the value of the path is an open queue, i.e. a queue all of whose members are available for access. ‘fst’ finds the first member of this queue, i.e. the earliest move that has not been integrated and binds it to the variable A. (We use upper case letters to represent variables, Prolog style.) Moves are represented as pairs the first member of which is the speaker associated with the move (either the user (‘usr’) or the system (‘sys’)) and the second member of which is the move itself, in this case the move of asking the question Q. The second condition checks that the non-integrated move A is of this form, i.e. that it is a question asked by the user. Finally, in the third condition the system checks in its DOMAIN resource that it in fact has a plan to answer this question. The plan is not actually loaded into the information state by this update rule which is only concerned with putting the question on the system’s agenda. But there is no sense putting a question on your agenda to respond to if you do not know how to answer it. Note that knowing how to answer a question is not the same as knowing what the answer is. In the example we are discussing answering a query about price information involves the system in addressing several questions to the user. It has a plan but the plan may not ultimately be successful, e.g. if the user fails to answer a crucial question. If the conditions are met the following updates are carried out: 1. ousr, ask(Q)W is deleted from the queue of non-integrated moves. 2. Q is pushed onto QUD and thus becomes the question under discussion. 3. The system puts an action to indicate positive acceptance (‘‘Okay’’) on its agenda and then places an action to indicate positive understanding (‘‘You asked about Q’’) after that. 4. If Q occurs in the value of /SHARED/ISSUES (i.e. if Q is an issue that has been raised but is still open, i.e. it has not been resolved) then Q is raised to the head of ISSUES and an action to reraise Q is placed at the end of the system’s agenda. Otherwise, if Q is not already on the list of open issues, it is put there. 5. Finally an action to respond to Q is placed at the end of the system’s agenda. The second case of reraising in (12), where the system reraises the issue of where the user wants to travel from, requires a selection rule for the system. This is a kind of update rule which adds new moves to the system’s agenda on the basis of material in the information state other than user moves that have not yet been integrated. (As we saw in (13)

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

115

integration rules may also add new moves to the agenda.) The class of this rule is thus ‘select_action’. (14) rule : reraiseIssue class : select action ( inð$/shared/issues; QÞ pre : not$domain :: planðQ; PÞ ( eff :

pushð/private/agenda; icm : reraiseÞ pushð/private/agenda; raiseðQÞÞ

The conditions of this rule check if there are any issues which have been raised previously by the system, but not yet resolved by the user. Issues are divided up between those for which the system has an associated plan in its domain resource and those for which it does not. For example, the ‘‘price issue’’ is one for which there is a plan: the system has to ask where the user wants to go, where from, when etc. However, there is no plan associated with the question of where the user wants to go from. This question is simply part of the plan for the price issue. Thus, when the system finds this question in the list of open issues (the first condition of this rule) and it finds that it does not have a plan for this issue (the second condition), it plans to reraise the question. Firstly, it plans to signal that it is reraising a question by putting icm:reraise on its agenda. In this simple implementation this corresponds to prefixing ‘‘so’’ to the next utterance, which is an ordinary raising of the question (placed on the agenda by the second action in the rule). In a more sophisticated implementation one could consider abbreviating the original raising of the question to make an appropriate reraising, e.g. ‘‘So, from what city?’’ But we do not have a general method for doing that yet, even in these simple dialogues. To do it properly would probably require more detailed syntactic and semantic analysis of contributions than we currently have in the implementation which uses key phrase spotting and template-based generation. This rule will not be applied if the system thinks that the user has rejected the question by refusing to answer it. The mechanism for this would involve not placing the question on ISSUES if the user appears to have rejected it. Thus the question would not be considered to be an open issue since it has not been successfully raised, i.e. confirmed as a question under discussion by both dialogue participants. This is one possible alternative way of interpreting the user’s visa question in (12), and if this approach were taken the system’s repeat of the question

116

ROBIN COOPER

AND

STAFFAN LARSSON

‘‘What city do you want to go from?’’ would not be regarded as a reraising and hence would be formulated as if it had never been asked before. We suspect that the choice between these two interpretations of the user’s visa question might depend on quite subtle issues concerning timing. For example, if the user’s question ‘‘Do I need a visa?’’ overlaps with the system’s question ‘‘What city do you want to go from?’’ it is possible that the user has not even heard the system’s question and that therefore a reraising would be inappropriate. Reraising Closed Issues In (15) we give an example of a dialogue with the system where the user reraises an issue which has been resolved (15) $SW Welcome to the travel agency! $UW do i need a visa for going to paris ? I’m from sweden. $SW You asked about visa. to paris. You are a citizen of sweden. Yes, you need a Visa. $UW okay, can i get some price information $SW You asked about price. $UW yeah, a flight from london on april the fifth as cheap as possible $SW Okay. by flight. from london. in april. the fifth. cheap. Okay, the price is 123 crowns. $UW Okay. Umm . . . i forgot, did i need a visa ? User reraises. $SW OK. Returning to the issue of Visa. Yes, you need a Visa. This can be handled with an integration rule similar to that in (13). (16) rule : integrateUsrReraise class : integrate 8 fstð$/private/nim/; AÞ > > > > > A ¼ ousr; askðQÞ4 > > > > < $domain :: planðQ; CÞ pre : not inð$/shared/issue; QÞ > > > > > > inð$/shared/com; DÞ > > > : $domain :: relevantðD; QÞ

117 ACCOMMODATION AND REACCOMMODATION IN DIALOGUE 8 delð/private/nim; AÞ > > > > > pushð/shared/qud; QÞ > > > > > pushð/shared/issues; QÞ > > > > > pushð/private/agenda; icm : acc posÞ > <   eff : pushð/private/agenda; icm : und pos : usr issueðQÞÞ > > > pushð/private/agenda; icm : reraise :QÞ > > > > > pushð/private/agenda; respondðQÞÞ > > > > > delð/shared/com; DÞ > > > : if doðinð$/private/bel; DÞ; delð/private/bel; DÞÞ This rule has additional conditions that are not present in (13). The first of these (the fourth condition) checks that the question the user has raised is not an open issue. The fact that it is not an open issue does not mean of itself that it is an issue that has been reraised. It could be the first raising of a new issue. It is the fifth and sixth conditions here that check that the issue has previously been resolved. These check that there is a fact D in the shared commitments (as the system sees them) which according to the system’s domain resource is relevant to the question which the user has raised. Relevance in this system has a very simple interpretation, namely being a possible answer to the question which itself is expressed very simply, e.g. city names are possible answers to questions about destination. Nevertheless, this strategy works for finding a closed issue only because of the extremely simple structure of the dialogues and the domains. It would fail as soon as a fact in the database could be the answer to more than one question in the domain. A more general treatment of reraising would need to keep track of closed issues explicitly in the same way we have kept track of open issues. Thus we envision two global QUDs, one for open issues as we have now and another for closed issues. In the actions the system’s reraising of the question mirroring the user’s is now not conditional on the question being an open issue (sixth action). The question is placed both on QUD and ISSUES (second and third action). Finally, the system has to revise its view of the shared commitments since clearly the user does not regard the fact D as shared, otherwise she would not be asking the question. The eighth action deletes D and the ninth action deletes it from the system’s private beliefs if it has been entered there. Clearly these deletions could involve a much more complex process of belief revision if we were to deal with dialogues of greater sophistication.

ROBIN COOPER

118

AND

STAFFAN LARSSON

We have not implemented a general update rule which would enable the system to initiate a reraising of a closed issue. Systems are not expected to forget, change their mind or have second thoughts. Thus they do not need to reraise closed issues, except as a response to the user revising shared information as we will see below. Revising and denying. In general, if the user provides an alternative resolution of an issue which has been previously resolved, this triggers a reraising of that issue. If the previous answer is incompatible with the new one, the old answer is removed. This allows the user to change his/her mind during the dialogue. Here is an example dialogue with the system: (17) $SW Welcome to the travel agency! $UW price information please $SW You asked about price. How do you want to travel? $UW a flight, april the fifth $SW by flight. in april. the fifth. Okay. What city do you want to go to? $UW london $SW Okay. to london. $UW actually, i want to go on the fourth $SW the fourth. What city do you want to go from? Initially, integration of the answer (the rule in (18)) will fail since there is incompatible information in SHARED/COM. The update rule for integrating answers includes the conditions in (18). (18) rule : integrateAnswer class : integrate

8 fstð$/shared/qud; QÞ > > > > fstð$/shared/issues; QÞ > > > > > > fstð$/private/nim/; AÞ > > > < A ¼ oX; answerðCÞ4 pre : > $domain :: relevantðC; QÞ > > > > > $domain :: combineðQ; C; DÞ > > > > > not inð$/shared/com; EÞ and $domain :: incompatibleðD; EÞ > > : ...

eff : f. . .

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

119

The first two conditions check that the question under discussion, Q, (i.e. first on QUD) is the most salient open issue (i.e. first on ISSUES). The third to fifth conditions find the first non-integrated move and check that it is an answer move with content C which according to the system’s domain resource is a relevant answer to Q. The domain resource will also contain information concerning how to combine the answer C with Q to form a proposition, D. (In the case where C is a short answer providing an argument to the lambda-expression Q this will be the result of function argument application. Longer answers involve a slightly more complex process.) The system will then try various accommodation strategies, including accommodation from /SHARED/COM formulated in (19). (19) rule : accommodateCom2Issues class : accommodate 8 inð$/private/nim/; oX; answerðAÞ4Þ > > > > > > < inð$/shared/com; PÞ pre : $domain :: questionðQÞ > > > $domain :: relevantðA; QÞ > > > : $domain :: relevantðP; QÞ eff : fpushð/shared/issues; QÞ This accommodation rule looks for an answer A among the moves which have not yet been integrated (first condition). It then looks for a proposition among the shared commitments established in the dialogue so far (second condition) which according to the system’s domain resource is an appropriate answer to some question for which A is also an answer (third to fifth conditions). Given that in this simple system answers can only be relevant to a single question,8 this strategy will be successful in identifying cases where we have two answers to the same question. A system that deals with more complex dialogues where this is not the case would need to keep track of closed issues in a separate list of closed issues. Thus the conditions will succeed if there is a question such that both the user answer and a stored proposition are relevant answers to it; in the example dialogue above, ‘‘departure date is the fourth’’ and 8 That is, in the full form in which they appear in $/SHARED/COM. ‘‘Chicago’’ can be an answer to ‘‘Which city do you want to go to?’’ and ‘‘Which city do you want to go from?’’ but when it has been combined with the questions the result will be ‘‘destination(Chicago)’’ and ‘‘from(Chicago)’’ respectively and it is this which is entered into the commitments.

120

ROBIN COOPER

AND

STAFFAN LARSSON

‘‘departure date is the fifth’’ are both relevant answers to the question ‘‘which day do you want to travel?’’. If such a question is found it is accommodated to ISSUES, i.e. it becomes an open issue again. When accommodateCom2Issues has been successfully applied, the retract rule in (20) will remove the incompatible information from the system’s view of shared commitments represented in /SHARED/COM. (20) rule : retract class : accommodate 8 inð$/private/nim/; oX; answerðAÞ4Þ > > > > > inð$/shared/com; P0 Þ > > > > > > < fstð$/shared/issues; QÞ pre : $domain :: relevantðP; QÞ > > > $domain :: relevantðA; QÞ > > > > > $domain :: combineðQ; A; PÞ > > > : $domain :: incompatibleðP; P0 Þ eff : fdelð/shared/com; P 0 Þ The conditions here are similar to those in (19). We look for an unintegrated answer (first condition) which is relevant to a question at the head of the list of open issues (third and fifth conditions) and for which there is already a relevant answer in the shared commitments (second and fourth conditions). Finally, we determine that the result of combining the answer with the question (sixth condition) is incompatible with the answer already found (seventh condition). If all this is true we delete the answer which is currently in the shared commitments. This will finally allow the new answer to be integrated by a rule that integrates an answer from the user, and a further rule will remove the resolved issue from QUD. It is also possible to remove the old answer by asserting its negation as in (21). (21) $SW Welcome to the travel agency! $UW price information for a flight to paris on april the fifth $SW You asked about price. by flight. to paris. in april. the fifth. What city do you want to go from? $UW actually, not the fifth $SW not the fifth. So, what day do you want to leave?

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

121

In this case, the system will explicitly reraise the issue to get a new response from the user. Again, the system will use the rule in (14) and signal reraising using ‘‘so,’’. All the rules will be applied as in the previous case, but the departure date question will not be removed since it is not resolved by the given answer. Eventually, this leads to the system reraising the question. Reraising of dependent questions. In some cases, an issue might be reraised which influences the answer to a further issue that has also been resolved. For example, the choice of price class for a flight influences the price of the flight. In this case, the influenced question also needs to be reaccommodated and answered again. (22) $SW Welcome to the travel agency! $UW what’s the price of a flight from london to paris april the fifth ? $SW You asked about price. by flight. From london. $SW to paris. in april. the fifth. $SW What class did you have in mind? $UW as cheap as possible $SW cheap. Okay. The price is 123 crowns. $UW actually, i might go for business class $SW first class. Okay. Concerning your question about price : The price is 1234 crowns. The rule that achieves the reraising of a dependent question to-ISSUES accommodation is shown in (23). (23) rule : accommodateCom2IssuesTransitive class : accommodate 8 inð$/private/nim/; oX; answerðAÞ4Þ > > > > > inð$/shared/com/; PÞ > > > > > $domain :: questionðQÞ > > > > > > < $domain :: relevantðA; QÞ pre : $domain :: relevantðP; QÞ > > > is emptyð$/shared/issuesÞ > > > > > $domain :: dependsðQ0 ; QÞ > > > > > inð$/shared/com; P0 Þ > > > : $domain :: relevantðP0 ; Q0 Þ

COM-

122

eff :

ROBIN COOPER AND STAFFAN LARSSON 8 delð/shared/com; P0 Þ > > > > < pushð/shared/issue; QÞ pushð/shared/issue; Q0 Þ > > > > : pushð/private/agenda; respondðQ0 ÞÞ

This rule is similar to (19) except that it looks for a question which depends on the question it finds corresponding to the answer provided by the user. It puts both questions onto the list of open issues and plans to respond to the dependent question. This rule, as currently implemented, is specific to the particular case treated in the system. There is, of course, a great deal more to say about what it means for one question to be dependent on another and how the system knows whether it should respond to dependent questions or raise them with the user.

Conclusion We believe that this discussion shows that the accommodation and reraising of questions plays a central role in dialogue and that these mechanisms point to some general conclusions about how to analyse the structure of dialogue. What we have here is not a simple hierarchical syntax in which the analogy of phrase structure is applied to dialogue structure. On our view dialogue structure is much more linear and involves transitions between information states representing the state of information in the dialogue (as viewed by one of the dialogue participants). However, the need for such information states also shows that a simple finite state view of dialogue (as often lies at the base of commercial implementations of dialogue systems) is not sufficient to give general explanatory accounts of dialogue. Rather we need transitions between information states which extend the ideas developed in dynamic semantics concerning information update. In particular we think that this work shows that the central idea of accommodation which is associated with presupposition in the classical semantic literature has much broader applications in formal approaches to the analysis of dialogue management. We feel that this points to a view of presupposition as related to a much more general dialogue phenomenon in which dialogue partners make assumptions about the nature of the other dialogue partners’ information state.

ACCOMMODATION

AND

REACCOMMODATION

IN

DIALOGUE

123

Acknowledgments This work was supported in part by the projects TRINDI (Task Oriented Instructional Dialogue), EC Project LE4-8314, SDS (Swedish Dialogue Systems), NUTEK/HSFR Language Technology Project F1472/1997, INDI (Information Exchange in Dialogue), Riksbankens Jubileumsfond 1997-0134, and SIRIDUS (Specification, Interaction, Reconfiguration in Dialogue Understanding Systems), EC Project IST-1999-10516 and ILT (Interactive Language Technology), Vinnova Project 2001-6340.

References Allwood, J. 1995. An Activity Based Approach to Pragmatics. Technical Report (GPTL) 75, Gothenburg Papers in Theoretical Linguistics, University of Gothenburg. Cooper, R., E. Engdahl, S. Larsson, and S. Ericsson. 2000. Accommodating questions and the nature of QUD. In M. Poesio and D. Traum, eds., Proceedings of Go ¨talog 2000, number 00-5 in Gothenburg Papers in Computational Linguistics (GPCL), Department of Linguistics, University of Gothenburg, Sweden. Engdahl, E., S. Larsson, and S. Ericsson. 2000. Focus-ground articulation and parallelism in a dynamic model of dialogue. Deliverable D4.1, TRINDI project, Department of Linguistics, University of Gothenburg, Sweden. Fikes, R. E. and N. J. Nilsson. 1971. Strips: A new approach to the application of theorem proving to problem solving. Artificial Intelligence 2:189–208. Ginzburg, J. 1996a. Dynamics and the semantics of dialogue. In J. Seligman and D. Westersta ˚hl, eds., Logic, Language and Computation. vol. 1, CSLI Publications. Ginzburg, J. 1996b. Interrogatives: Questions, facts and dialogue. In S. Lappin, ed., The Handbook of Contemporary Semantic Theory. Oxford: Blackwell. Ginzburg, J. 1998. Clarifying utterances. In J. Hulstijn and A. Niholt, eds., Proceedings of the Twente Workshop on the Formal Semantics and Pragmatics of Dialogues, pp. 11–30. Enschede: Universiteit Twente, Faculteit Informatica. Larsson, S. 2002. Issue-Based Dialogue Management. Ph.D. dissertation, University of Gothenburg, Sweden. Larsson, S., P. Bohlin, J. Bos, and D. Traum. 1999. Trindikit 1.0 manual. Deliverable D2.2, TRINDI project, Department of Linguistics, University of Gothenburg, Sweden. Traum, D., J. Bos, R. Cooper, S. Larsson, I. Lewin, C. Matheson, and M. Poesio. 1999. A model of dialogue moves and information state revision. Deliverable D2.1, TRINDI project, Department of Linguistics, University of Gothenburg, Sweden.

5

Specific Indefinites, Presupposition and Scope$ BART GEURTS

Abstract In the past few years it has been suggested by several independent sources that specificity and presupposition are kindred phenomena. My proposal is to relate specificity and presupposition to each other not by reducing the former to the latter, as in earlier accounts, but by subsuming them under a more comprehensive rubric, which I call backgrounding. I argue that this view doesn’t suffer from the problems other theories are faced with, and that it sheds new light on presupposition projection as well as on a number of phenomena that thus far lacked a systematic account.

$

Portions of this paper were presented at the 1999 ESSLLI workshop on focus and presupposition in multi-speaker discourse (Utrecht) and the 2001 workshop on specificity and sentence type (Berlin); earlier versions of the paper duly appeared in the proceedings of these workshops. The first version contained what in the meantime has appeared as a squib in Linguistic Inquiry (Geurts, 2000a), where it is argued that specific indefinites cannot in general be construed in situ, or in other words, that movement of some sort is called for; here it will be taken for granted that this is so. For comments and discussion I am indebted to Reinhard Blutner, Paul Dekker, Brenda Kennelly, Rob van der Sandt and Henk Zeevat. Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

125

126

BART GEURTS

Introduction Indefinites and Scope The principal characteristic of specific indefinites is that they have a predilection for taking wide scope:1 (1) a.

b.

After all that effort and time they now don’t know where 40 per cent of it is. (New Scientist, 24 April 1999; the neuter pronoun refers to 182 kg of plutonium dumped into the Irish Sea by the Sellafield nuclear plant.) All critics who were invited to comment on some poems written by a 2-year-old bonobo hailed them as mature masterpieces.

The indefinite NP ‘40 per cent of it’ in (1a) occurs within the syntactic scope of a negation sign and an attitude verb, but it is interpreted as if they weren’t there; for what the sentence means is something like: ‘40 per cent of the plutonium is such that they don’t know where it is.’ The same holds, mutatis mutandis, for the indefinite ‘a 2-year-old bonobo’ in (1b). Observations like these have been taken to show that specific indefinites always take widest scope, or even that they are referential expressions (e.g. Fodor and Sag, 1982), but as examples given already by Kasher and Gabbay (1976) and Farkas (1981) demonstrate, neither claim is correct: (2) a.

Now, after all that effort and time, they say they don’t know where 40 per cent of it is (New Scientist, 24 April 1999).2 b. Each student has to come up with three arguments which show that some condition proposed by Chomsky is wrong (Farkas, 1981). c. The police report might indicate that Mary wants to marry a Swede (Kasher and Gabbay, 1976).

The intended interpretation of (2a) presumably is that ‘they say that 40 per cent of the dumped plutonium is such that they don’t know where it is,’ and the same holds, mutatis mutandis, or the prepositional

1 I will argue eventually that specificity has nothing to do with scope, in the grammarians’ sense, but for the time being I will use the notion as an expository device for distinguishing between readings. 2 (1a) occurred in a caption, and (2a) in the text, of the same article.

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

127

object in (2b) and a Swede in (2c). Hence, in each case the specific indefinite is interpreted as if it occurred midway between its actual surface position and the outermost scope-bearing expression. In all these examples there appears to be a mismatch between the position at which an indefinite appears and its preferred interpretation. Following many of the more recent contributions to the literature, I will assume that this is the hallmark of specificity (e.g. Abusch, 1994; Reinhart, 1997; Winter, 1997; Van Geenhoven, 1998). Such mismatches are not the norm: indefinites are often interpreted in situ, and there is some reason for taking this to be the default option. The reason is that comparatively ‘neutral’, i.e. semantically attenuate, indefinites have a preference for in situ readings, as the following pairs illustrate: (3) a. b. (4) a. b.

Several students reported that they had been harassed by a professor. Several students reported that they had been harassed by a professor emeritus of the law faculty. Several students reported that they had been harassed by professors. Several students reported that they had been harassed by professors wearing false beards and pink gowns.

Both (3a) and (4a) are more likely to be understood with the sentencefinal indefinite interpreted in situ. It is only when these expressions become ‘heavier’ that a specific reading is enforced, as (3b) and (4b) illustrate. Note, incidentally, that (4b) belies the popular view that bare plural indefinites are always construed in situ. It may be the case that they like such readings better than most other indefinites do, but bare plurals allow for specific construals, too. On the strength of these observations it may be assumed that in situ interpretations are the rule, and specific interpretations the exception. Van Geenhoven (1998) suggests, furthermore, that wide-scope construals of specific indefinites are preferred, ceteris paribus, to intermediate-scope construals. I believe that she is right about this, though it must be conceded that intuitions are rather delicate. At any rate, the argument must be along the same lines as previously: (5) a. b.

Every city was represented by twelve athletes sponsored by a brewery. Every city was represented by twelve athletes sponsored by a local brewery.

BART GEURTS

128 (6) a. b.

Every newspaper featured multiple reviews of a gothic novel. Every newspaper featured multiple reviews of a gothic novel written by its editor-in-chief.

Setting in situ readings aside, it seems to me that in the (a) sentences there is a preference for construing the sentence-final indefinites as having wide, rather than intermediate, scope. The balance tips, however, when the indefinites are enhanced with material enabling a link with the universally quantified subject, as the (b) examples demonstrate. These observations support Van Geenhoven’s claim that, all else being equal, wide-scope readings are more easily obtainable than intermediate-scope ones. It bears emphasizing that these preferences hold ceteris paribus only, and are easily overridden by considerations of plausibility, as indeed the examples in (3) to (6) demonstrate. We thus arrive at the following preference order on the range of possible interpretations of indefinite NPs: in situ o wide scope o intermediate scope In the second half of this paper I endeavour to show how this pattern can be accounted for. No Need to Know There is a widespread opinion that in order for an indefinite NP to be used with a specific interpretation, the speaker must have a particular individual in mind (e.g. Kasher and Gabbay, 1976; Fodor and Sag, 1982; Kratzer, 1995; Manga, 1996; Yeom, 1998; van Rooy, 1999). It might be thought that this explains the unmistakable family resemblance between specific indefinite NPs, on the one hand, and definite NPs, on the other (which will be documented at some length in the next section). Just as a speaker employs the definite article to signal that an individual is given as part of the common ground between him and the hearer, he employs a specific indefinite if he wants to indicate that an individual is known to him, though not to his audience. In short, while definiteness implies givenness to speaker and hearer, specificity implies accessibility to the speaker alone. (For obvious reasons, there are no linguistic devices for signalling that an individual is accessible to the hearer alone.) This view on specificity is untenable. As Haspelmath (1997) points out, there are many languages that allow indefinite NPs to be morphologically flagged as ‘unknown to the speaker’, but the use of such flags doesn’t entail non-specificity. For example, German ‘irgendein N’ conveys that the speaker doesn’t know the N in question, but may well be used specifically:

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

129

(7) Wilma hat vor, irgendeinen Schweden zu heiraten. Wilma intends some-or-other Swede to marry. Even in the absence of explicit morphological clues, there are many cases in which it is simply false, intuitively speaking, that the witness of a specific indefinite must be known to the speaker. This is especially problematic when specific indefinites take intermediate scope, but these are not the only cases. Consider (1a), for example. It would be patently wrong to say that the author of this sentence must have had a particular portion of plutonium in mind; yet there can hardly be any doubt that the indefinite ‘40 per cent of it’ is being employed in a specific sense. Whatever it may be, having something in mind is not a prerequisite for specificity. Having arrived at this conclusion, we should ask ourselves how we can recognize specificity in the absence of telltale scope-bearing expressions. The answer to this question, I submit, is that by and large we can’t. That is to say, the chief problem for a theory of specificity is to account for the interaction between specific indefinites and further scope-bearing expressions occurring in the same sentence. Apart from that, there is little in the way of systematic evidence for a theory of specificity to build on, though the following type of data might be an exception: (8) At the party, Fred danced with an Irish woman, and so did Barney. This sentence may or may not be construed as implying that Fred and Barney danced with the same woman, and if this is to do with the fact that the indefinite ‘an Irish woman’ is either specific or non-specific, as suggested by Kasher and Gabbay (1976), then this is a case in which specificity manifests itself even in the absence of other scope-bearing expressions. Note, however, that in cases like (8), one purported reading is always entailed by the other, which makes it hard to prove that these sentences are truly ambiguous.

Similarities between Specific Indefinites and Definites It was hinted already that, in certain respects, there is a resemblance between specific indefinites and definite expressions. In fact, the similarities are quite striking, as the following observations will demonstrate, and if these facts may be taken at face value, any theory of specificity worth its salt should be able to explain why definites and specific indefinites are so much alike.

130

BART GEURTS

Scope The hallmark of specific indefinites is that they tend to take scope over anything else in the sentence, which is characteristic of definites, too. One example will suffice to illustrate this well-worn observation: (9) All critics who were invited to comment on some poems written by the Danish chancellor hailed them as mature masterpieces (cf. (1b)). This is most likely to be read as implying that Denmark has a chancellor who wrote all the poems presented to the various critics. Of course, definites can take ‘intermediate scope’, too, as (10) demonstrates: (10) All critics who were invited to comment on some poems written by their spouses hailed them as mature masterpieces. If the possessive pronoun is bound by the subject NP, it is of course impossible to obtain a wide-scope reading for the definite expression their spouses; but an intermediate reading remains feasible-indeed, it is the most natural reading in this case. One respect in which definites differ from indefinites at large is that it is quite difficult to obtain something akin to in situ readings for the former, whereas we have seen that the latter prefer such readings. Narrow-scope readings for definite NPs do occur, though: (11) That wasn’t Fred’s wife, you blockhead: Fred isn’t even married! But such examples are clearly marked (see Geurts, 1998a, for further discussion). Hence, although definites and indefinites are quite similar in the way they interact with scope-bearing expressions, their preferences in this regard are different. To summarize: definites: indefinites:

wide scope o intermediate scope o in situ in situ o wide scope o intermediate scope

Partitives As Ladusaw (1982) was the first to point out, the nominal constituent of a partitive PP must be definite or specific; non-specific indefinites and quantified NPs are not allowed in this position: (12) Fred is one of {the/several/*most/*all/*sm/*|} employees who will be fired. Here sm represents unstressed some, which has a distinct preference for a non-specific reading, like the bare plural, indicated by ‘|’.

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

131

Indefinite this Although formally this is a definite article, it sometimes appears to function as if it were indefinite (see Prince, 1981 for discussion): (13) There is this giant spider in the cupboard. When used in this manner, this-NPs function as indefinites because, intuitively, they introduce discourse entities that are new, an intuition which is corroborated by the following example: (14) ?Yesterday, our little daughter brought [a giant spider]i into the house, and now there is [this giant spider]i in the cupboard. It is not possible, apparently, for the this-NP to refer back to the spider introduced in the first conjunct, which confirms the impression that ‘this giant spider’ functions as an existential expression. Furthermore, indefinite this-NPs behave more like specific than non-specific indefinites, because they typically take wide scope: (15) a. b. c.

If this giant spider is in the cupboard, Betty will go berserk. There is a giant spider at large, and if it is still in the cupboard, Betty will go berserk. If there is a giant spider in the cupboard, Betty will go berserk.

(15a) is more or less synonymous with (15b), rather than (15c), which is precisely what one should expect if ‘this giant spider’ were specific. These observations indicate that indefinite this-NPs are expressions that are marked for definiteness but function as specific indefinites. It is hard to see how this ensemble of form and function could occur unless specificity and definiteness are kindred phenomena. Cross-Linguistic Evidence Perhaps the most telling evidence is that in language after language definiteness and specificity are lumped together into the same morphosyntactic rubric. I will give a handful of more or less arbitrarily chosen examples. 

Bemba:

In Bemba, a Bantu language, there is a class of nominal prefixes of the form consonant-vowel, and another class of the form vowel-consonantvowel. The former are used to mark non-specific indefinites, while the

BART GEURTS

132

latter alternatively convey definiteness or specificity. The following examples are from Givo´n (1978); here and in the following glosses are reproduced from the original source: (16) a.

b.

c.



Umu-ana a-a-fwaaya ci-tabo. vcv-child he-past-want cv-book ‘The child wanted a book (be it any).’ Umu-ana t-a-a`-somene ci-tabo. vcv-child neg-he-past-read cv-book ‘The child didn’t read a/any book.’ Umu-ana a-a-fwaaya ici-tabo. vcv-child he-past-want vcv-book ‘The child wanted the book’ or ‘The child wanted a specific book.’

Samoan:

Samoan is similar to Bemba in that it has two articles, one of which signals non-specific indefiniteness, while the other combines specificity and definiteness (examples from Lyons, 1999): (17) a. Sa i ai le uluga¯li‘i‘o Papa le tane a ‘o Eleele le fafine. Past exist Art couple Pres P. Art husband but Pres E. Art woman ‘There was a couple, Papa, the husband, and Eleele, the wife.’ b. ‘Au-mai se niu. take-Dir Art coconut ‘Bring me a coconut.’ 

West-Greenlandic Inuit:

In West-Greenlandic Inuit, an ergative language, transitive verbs may become intransitive by incorporating their objects. This shows itself, among other things, in the case marking on the subject, which is absolutive for intransitive, and ergative for transitive subjects. Moreover, it is only in transitive constructions that verbs bear objectagreement markers. The object of a transitive construction receives absolutive case, and may be either specific or definite, while incorporated objects are non-specific. According to Manga (1996), this is typical of ergative languages. The following sample of WestGreenlandic Inuit is from Van Geenhoven (1998): (18) a.

Angunguaq tikip-p-u-q. A.Abs arrive-Ind-Intr-3sg ‘Angunguaq arrived.’

SPECIFIC INDEFINITES, PRESUPPOSITION b.

c.

d.



AND

SCOPE

133

Angunguaq aalisakka-mik neri-v-u-q. A.Abs fish-Inst.sg eat-Ind-Intr-3sg ‘Angunguaq ate fish.’ Arnajaraq aalisaga-si.nngi-l-a-q. A.Abs fish-buy-Neg-Ind-Intr-3sg ‘It is not the case that Arnajaraq bought {a/more than one} fish.’ Angunguu aalisagaq neri-v-a-a. A.Erg fish-Abs eat-Ind-Tr-3sg.3sg ‘Angunguaq ate the/a particular fish.’

St’a´t’imcets:

St’a´t’imcets (Lillooet Salish) features an indefinite article which can only occur within the scope of a negative expression, a question, a modal and so on. In the absence of such operators another article must be used, which has a specific-definite function. The following examples are from Matthewson (1999): (19) a.

b.

(20) a.

b.

Cw7aoz kw-s a´ts’x-en-as ku sqaycw. Neg Det-Nom see-Tr-3Erg Det man ‘S/he didn’t see any men.’ ´ ts’x-en-as ku sqaycw. *A see-Tr-3Erg Det man ‘S/he saw a man.’ ´y-lhkan ptakwlh, pta´kwlh-min lts7a ti sme´m’lhats-a . . . Hu going.to-1sg.Subj tell.story tell.story-Appl here Det woman.Dimin-Det ‘I am going to tell a legend, a legend about a girl . . . ’ Wa7 ku7 ´lal ı la´ti7 ti sme´m’lhats-a Prog Quot cry Deic Det woman.Dimin-Det ‘The girl was crying there.’

This sample will suffice to show that many languages treat definiteness and specificity as related notions, which together stand in opposition to non-specific indefiniteness. In conjunction with the evidence of the preceding sections, these data raise the question what it is that definites and specific indefinites have in common. In my opinion, one of the main criteria for assessing theories of specificity should be how good their answers to this question are.

BART GEURTS

134 Specificity and Distributivity

It has been argued by Reinhart (1997) and Winter (1997) that specific indefinites which have escaped from a scope island don’t allow for a distributive interpretation. Reinhart credits Ruys with this insight; Winter attributes it (collectively) to Ruys and himself. Reinhart cites example (21a) from a manuscript by Ruys: (21) a. b. c.

If three relatives of mine die, I will inherit a house. There are three relatives of mine such that, if they all die, I will inherit a house. There are three relatives of mine such that, if any of them dies, I will inherit a house.

On the most likely reading of (21a) the indefinite ‘three relatives of mine’ is construed with narrow scope, but if it gets a specific reading and outscopes the if-clause, then according to Reinhart, Ruys and Winter, it can only be understood collectively. That is to say, if the indefinite is specific, (21a) is synonymous with (21b), not (21c). This observation is not quite correct, however; what Reinhart et al. have found is not a lawful correlation but merely a trend. Firstly, as noted by Matthewson (1999), there are native speakers of English who manage to obtain a distributive reading for (21a), and the same holds for parallel sentences of other languages. Secondly, Van Geenhoven (1998) points out that intuitions shift markedly when we vary the example. Thus it appears to be easier to get a distributive reading for the following sentence: (22) If some relatives of mine invite me for dinner, I will panic. In short, although in environments like (21a) or (22) specific indefinites seem to prefer collective construals, specificity does not entail collectivity. This is bad news for two rather different theories of specificity. On the one hand, theories that seek to deal with specificity with the help of quantifier raising will be embarrassed by the fact that specific indefinites disprefer non-distributive readings. On the other hand, theories that rely on choice-functions instead of quantifier raising will find it quite difficult to explain the distributive readings – a point which Winter (1997) emphasizes, because he is confident, apparently, that such readings don’t occur (for further discussion, see Geurts, 2000a). There is one family of theories that can account for distributive as well as non-distributive readings: these are theories which, on the one hand, resemble the quantifier-raising approach in that their account of

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

135

specificity is based on movement, while, on the other hand, they agree with the choice-function approach that indefinites aren’t quantifier expressions. Two such theories are discussed below.

The Binding Theory of Presupposition In the remainder of this paper I present a unified account of specificity and presupposition, which is based upon the binding theory of presupposition, so before we move on I want to quickly recapitulate the main tenets of that theory; for more extensive discussion, see van der Sandt (1992), Geurts (1999) and Geurts and van der Sandt (1999). The binding theory is an extension of discourse representation theory (Kamp, 1981), and consists of three principal claims. The first of these is that anaphora is a species of presupposition, and that the standard presupposition-inducing expressions (such as definite NPs, factives, transition verbs and so on) differ from pronominal anaphors mainly in that they possess a richer semantic content. This difference explains why in general presupposition inducers, unlike anaphoric pronouns, can be interpreted by way of accommodation, which is the second key notion in the theory. Finally, it is assumed that the process of presupposition projection is subject to certain constraints. It is the status of these constraints that will be especially important in the following. Formulated in procedural terms, the binding theory predicts that if an utterance contains a presupposition-inducing element, the hearer will initially attempt to bind the presupposition to a suitable antecedent, just as he would try to bind an ordinary anaphor. If the presupposition cannot be so bound, it will be accommodated, i.e. it will be inserted in some accessible discourse representation structure (DRS). In general the number of positions at which a presupposition may be accommodated is greater than one, and if it is the choice is restricted by various constraints, but before I turn to these, let me first illustrate the workings of the theory: (23) If Fred is gay, then his son is gay, too. This sentence contains (at least) two presupposition-inducing expressions: the definite NP his son, which triggers the presupposition that Fred has a son, and the focus particle too, which triggers the presupposition that someone other than Fred’s son is gay. Note that the first presupposition is ‘inherited’ by the sentence as a whole, while the second one is not: normally speaking, an utterance of (23) would

136

BART GEURTS

license the inference that (according to the speaker) Fred has a son, but not that someone else besides Fred’s son is gay. The binding theory accounts for these observations as follows. Suppose that the grammar assigns (23) the intermediate semantic representation in (24a). I assume for convenience that most interpretative problems have been cleared out of the way already, and that the only thing that remains to be done is resolve the presuppositions triggered by his son and too, which are marked out by single and double underscores, respectively.3 (24) a.

½x : FredðxÞ; ½: gayðxÞ ) ½u; v : x’s-sonðuÞ; gayðvÞ; vau; gayðuÞ

½x; u : FredðxÞ; x’s-sonðuÞ; ½: gayðxÞ ) ½v : gayðvÞ; vau; gayðuÞ c. ½x; u : FredðxÞ; x’s-sonðuÞ; ½v : v ¼ x; gayðxÞ; gayðvÞ; vau ) ½: gayðuÞ d. ½x; u : FredðxÞ; x’s-sonðuÞ; ½: gayðxÞ ) ½: gayðuÞ b.

(24a) is the initial semantic representation correlated with (23), in which two presuppositions remain to be resolved. One of these, that Fred has a son, cannot be bound, and therefore must be interpreted by way of accommodation. Now there is a general constraint on presupposition projection to the effect that any presupposition prefers to be projected to as high a position as possible, and accordingly our first presupposition is accommodated in the principal DRS, which yields (24b). The remaining presupposition, triggered by the focus particle, can be bound in the antecedent of the conditional; this results in (24c) which, assuming that Fred and his son are different persons, is equivalent to (24d). The binding theory may be summed up in the following three principles: (A) Presuppositions must be projected (i.e. bound or accommodated). (B) Binding is preferred to accommodation. (C) A presupposition must be projected to the highest possible DRS. It will be evident that none of these principles is absolute, although the first two may be more absolute than the last one. They are all subject to general constraints on interpretation, which require that an interpretation be consistent, coherent and so on. Before these principles come into 3 Officially, I maintain that the presupposition triggered by too is not quite as simple as what is assumed in the following analysis. See van der Sandt and Geurts (2001) for discussion.

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

137

play, presuppositions are merely representational structures, and are therefore completely inert. Principle A drives away the inertia by insisting that presuppositions be either bound or accommodated. Principle B captures the insight that accommodation is a repair strategy: in principle, a presupposition wants to be bound, but if it cannot be bound it will be accommodated. Principle C may be viewed as a generalization of a constraint first proposed by Heim (1983). Heim distinguishes between two types of accommodation: global and local. In terms of the present framework, a presupposition is accommodated globally if it goes to the principal DRS, and locally if it is accommodated in the DRS where it was triggered. Heim’s proposal is that, in general, global accommodation is preferred to local accommodation, and principle C generalizes this in two ways. Firstly, this principle applies not only to accommodation but also to projection in general. This makes some difference from an observational point of view (though not much), and it is surely more attractive conceptually speaking. Secondly, although it is possible to capture Heim’s distinction between global and local accommodation in our framework, the distinction as such doesn’t play a role in the theory. In general, there is a line of accessible DRSs in which a presupposition can be accommodated, the two ends of this chain being the main DRS and the DRS where the presupposition arises. Global and local accommodations are just convenient labels for referring to accommodation in these DRSs, but they do not denote special processes. I should like to stress that the fundamental insight underlying this treatment of presuppositions is not a controversial one. It is that presupposed information is information that is presented as given. Most extant theories of presupposition accept this premise, too. What distinguishes the binding theory from other accounts is just that it doesn’t draw a sharp line between presupposition and anaphora. Hence, although the choice of framework is essential in some respects, the gist of my analysis of specific indefinites could be expressed in other frameworks, too. To say that presupposed information is presented as given is not to say that it is given. Indeed, the concept of accommodation merely puts a label on the observation that speakers are wont to exploit (in Grice’s sense) presupposition-inducing expressions in order to convey information that is new. The point is a familiar one, I take it, but it deserves to be stressed, because it is sometimes thought that accommodation will be the weak spot of any theory of presupposition that adopts the notion, as most of them do (see Abbott, 2000 for a recent attack along these lines). Even if it could be demonstrated that,

138

BART GEURTS

say, definite NPs are regularly used to refer to entities that are new (and Abbott maintains that this has been demonstrated), that wouldn’t even begin to show that the standard view of presupposition is on the wrong track. It would merely corroborate what we knew already, namely that speakers are adept at exploiting (still in the Gricean sense) linguistic devices for their purposes.

Accommodating Indefinites In the past few years, it has been suggested by several independent sources that specificity should be handled in terms of, or at least in conjunction with, presupposition projection (Cresti, 1995; Van Geenhoven, 1998; Yeom, 1998). This is an attractive idea, as I will try to show, but it requires a dramatic change of perspective, too, because it implies that specificity is an essentially pragmatic phenomenon. Following these developments, I will present my own unified theory of presupposition and specificity in the next section. In many respects, my account is related as well as indebted to Van Geenhoven’s, which will therefore be discussed first. Incorporation versus Accommodation The majority position in the literature on specificity is that indefinites are ambiguous between specific and non-specific readings. Van Geenhoven (1998) doesn’t take exception to this view, but she develops it in an entirely new way. According to Van Geenhoven, non-specific indefinites are ordinary predicates, which neither possess quantificational force nor introduce reference markers or anything of the sort. If the indefinite in (25a), for example, is interpreted non-specifically, it doesn’t have narrow scope; indeed, it doesn’t have scope at all because it is semantically incorporated by the verb, as suggested by the paraphrase in (25b): (25) a. b.

Every man loves a woman. Every man is a-woman-lover.

If, on the other hand, an indefinite gets a specific reading, its semantic representation is rather different. Specific indefinites are analysed in accordance with the standard DRT doctrine on indefinites, save for the fact that it is stipulated that they must be accommodated. Or in other words, if a woman in (25a) is specific, it is treated as if it were a presupposition-inducing expression whose presupposition has the peculiarity that it doesn’t want to be bound. Hence, the indefinite is

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

139

dealt with in two steps. Firstly, the grammar produces the initial discourse representation in (26a), in which the semantic correlate of a woman is marked as specific, and then this representation is fed into the projection mechanism of the binding theory, which treats the indefinite description as it would treat any (other) presuppositional expression, except that it cannot be bound. Consequently, it must be accommodated, and since there is a general preference for accommodating things at the highest level of representation, it is predicted that the resulting interpretation will be (26b). (26) a. b.

[: [x: man(x)]/every xS[u: woman(u), x loves u]] [u: woman(u), [x: man(x)]/every xS[: x loves u]]

I find this analysis appealing for a number of reasons. To begin with, it comes essentially for free, because all the machinery it employs is already in place, as it is required anyway for dealing with presupposition projection. Secondly, Van Geenhoven’s proposal explains the parallels as well as the differences between definites and specific indefinites. The reason why definites and specific indefinites are so similar is that they are interpreted by the same projection mechanism; the main difference is that definites want, and specific indefinites don’t want, to be bound. Thirdly, the theory accounts in a principled way for the puzzling pattern of interpretations discussed in the Section on Indefinites and Scope, which I repeat here for ease of reference: in situ o wide scope o intermediate scope According to Van Geenhoven, indefinites are ambiguous between a specific and a non-specific reading, and if it may be assumed that the latter prevails by default, then an in situ construal is preferred to a reading that involves movement, and if an indefinite gets a specific reading, principle C of the binding theory entails a preference for a wide scope as opposed to an intermediate-scope reading. Although Van Geenhoven’s theory hinges on the premise that specific indefinites are construed by way of movement, it should be stressed that this account has nothing to do with quantifier raising (or, for that matter, any other of the standard techniques for dealing with quantifier scope). Presupposition projection is a pragmatic affair, and therefore Van Geenhoven’s proposal can only be seen as an attempt at dealing with specificity in pragmatic terms. Quantifier raising, in contrast, takes place at or near the syntax-semantics boundary, so a theory based on raising implies that specificity is a grammatical phenomenon.

140

BART GEURTS

Objections Although I applaud Van Geenhoven’s pragmatic turn, and agree with the fundamental intuition underlying her theory, I have two objections, one of which I consider to be particularly serious. To begin with the major problem, I maintain that Van Geenhoven’s analysis is conceptually incoherent. To my mind, the very idea of a class of expressions that insist on being interpreted by way of accommodation is a contradiction in terms. Accommodation is a repair strategy by definition. A speaker who presupposes that j presents j as given, and if it is not given it is at the hearer’s discretion whether or not he wants to play along by accommodating j. Therefore, accommodation isn’t anything like an ordinary rule of interpretation; it is a fall-back option, and if one wants to postulate a linguistic category that selects for this option, there is a fair amount of explaining to do. But can’t we simply broaden the concept of accommodation by ruling that it applies not only to presuppositions but to certain other types of information, as well? We can, of course, but there is a price to pay. A broadening of the notion of accommodation entails that we forfeit a powerful explanatory lever in our theory of presupposition projection. For we then will have to come up with new answers to such questions as: What justifies accommodation?, Why is binding preferred to accommodation?, and so on. And as long as I don’t see how these questions might be answered, I am not willing to pay this price. My slightly less urgent complaint concerns Van Geenhoven’s assumption that indefinites are systematically ambiguous between specific and non-specific readings. Notwithstanding the fact that this assumption is commonplace in the literature, I don’t believe there is much independent evidence to support it, but that is as it may be, because nobody would deny that ambiguities are ugly and should be avoided at practically any cost. And, come to think of it, one should expect that a specific/non-specific ambiguity can be avoided in a framework based on the insight that specificity is a pragmatic phenomenon.

Specificity and Backgrounding My proposal is to relate specificity and presupposition to each other, not by reducing the former to the latter, as Van Geenhoven has tried to do, but by subsuming them under a more comprehensive rubric, which I call ‘backgrounding’. I will argue that this view doesn’t suffer from the shortcomings discussed in the foregoing, and, furthermore, that it

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

141

throws a new and perhaps brighter light on presupposition as well as on a number of phenomena that thus far lacked a systematic account. Foreground and Background Following Foley and van Valin (1985) and Foley (1994), among others, I understand the opposition between foreground and background purely in terms of informational prominence, where prominence is a relational rather than an absolute notion. By uttering a sentence a speaker typically conveys a considerable amount of information, only a small portion of which is central to his concerns. The remainder is backgrounded information: ancillary matter that merely serves to anchor the foregrounded information to the context, or information which is brought in en passant. Backgrounded information is not necessarily unimportant, but it is of secondary interest in relation to foregrounded information. Thus the notion of background is primarily a negative one: backgrounded information is what remains when foregrounded information is taken away. It may well be, therefore, that it is impossible to provide a single positive description covering all sorts of background information. But no matter how many reasons for, or ways of, backgrounding there may be, I will suggest that at least some interpretative mechanisms do not discriminate between them. A further, and crucial, negative characteristic of my notion of background is that it doesn’t entail givenness; only the converse is true. Backgrounded information may be given, or presented as given, but new information is not necessarily foregrounded. For example, enclosing new information in (intonational or orthographic) parentheses often serves to indicate that it is of secondary importance, which is to say that it is backgrounded, not that it is presented as given. My notion of background is clearly related to Abbott’s (2000) ‘nonassertion’ and Horn’s (2000) ‘assertoric inertia’. The basic intuition in each case is that the main point of an utterance enjoys a special pragmatic status, while the remainder is, in some sense, downgraded. What distinguishes my concept from the other two is mainly that its interpretative effects are more explicit (see below). Apart from that I prefer to avoid the notion of assertion in this connection, because otherwise I would have to assume, contrary to what I take to be linguistic common sense, that assertions may occur in syntactically embedded positions. Although the distinction between foreground and background may be signalled by intonational means, I don’t want to make any substantial claims about the relationship between intonation and foreground/background. However, I should like to note that the

142

BART GEURTS

correlation between intonational prominence and foregrounding is imperfect, at best. This observation is not new, but I feel it bears emphasizing nonetheless. Consider the following example: (27) The course on postmodern theology will be given by [the dean]F. Suppose, for enhanced clarity, that this is an answer to the question ‘Who is teaching the course on postmodern theology this year?’, so we can be sure that the non-focused part of (27) is given, and therefore backgrounded. Now of course the focused part is (presented as) given, too, simply by virtue of the fact that it is a definite NP. But surely everything in this statement cannot be given? The solution to this puzzle is not so hard to find: the focus on the dean doesn’t highlight the dean, but rather the fact that it is he who will be teaching the course on postmodern theology. The dean is given; that he will play a certain role is foregrounded. If backgrounded information need not be given, there is no reason why it couldn’t be marked as new. I want to suggest that this is not just an abstract possibility: it does happen that backgrounded information is marked as new; for this is precisely what specificity comes down to. Intuitively, specific indefinites carry information that is used to set the stage for the utterance’s main point. Like presuppositions, specific indefinites are separate units of communication, to be integrated into the context before the remainder of the utterance is processed; this is why semantically incorporated indefinites cannot be specific.4 Accessibility and the Buoyancy Principle An utterance is always interpreted within a context, and broadly speaking utterances and contexts interact with each other in two ways: the context affects the interpretation of an utterance, which in its turn changes the context in which it occurs. In DRT the context of utterance is pictured as a line of accessible DRSs, and therefore the notion of accessibility is of central importance to DRT (as it is, mutatis mutandis, to all dynamic theories of meaning). What, exactly, is accessibility? From a technical point of view this question is not so hard to answer, but when we interpret the question as being about the theoretical status of the accessibility relation, many different answers are possible. In Kamp’s 4 I suspect, moreover, that any indefinite that is not semantically incorporated is eo ipso specific. This position entails, however, that generic indefinites are generally specific, and thus runs up against the currently prevailing view, as expressed, for example, by Krifka et al. (1995), that indefinite generics in the singular are never specific. I believe this view can be disproved, but will not try to do so here.

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

143

(1981) original version of DRT, accessibility was associated with anaphoricity in the sense that it was only used for constraining the interpretation of anaphora: an anaphoric pronoun had to find its antecedent in an accessible DRS. In later versions of the theory, the notion of accessibility gradually assumed a much broader significance. Thus, as we have seen in the Section on The Binding Theory of Presupposition, in the binding theory of presupposition accessibility demarcates what is given at the point where an expression occurs. I believe that an even broader view is called for, and that the accessible domain must be seen as the background against which an utterance is interpreted, where ‘background’ is to be understood as explained above. When we thus broaden our perspective on the significance of accessibility, it is only to be expected that some of the principles of interpretation hitherto cast in terms of accessibility will have to be generalized. This applies, in particular, to principle C of the binding theory, which I propose to supplant with the following: The Buoyancy Principle Backgrounded material tends to float up towards the main DRS. Strictly speaking, the Buoyancy Principle isn’t part of our theory of presupposition projection, because it is not specifically about presuppositions, so all that remains of the original binding theory is two ‘axioms’, one saying that presuppositions want to be bound, the other, that presuppositions that cannot be bound may be accommodated. The theory’s predictions aren’t affected by this change, although they are now seen in a somewhat different light. In particular, I am no longer committed to the claim that presuppositions tend to take ‘wide scope’ because they are presuppositions; it is rather because they are backgrounded, and therefore subject to the Buoyancy Principle, that they gravitate towards the principal DRS. But as far as the theory of presupposition is concerned, the proposed modification isn’t exactly a volte-face. Still, this relatively minor amendment may turn out to be more consequential than one should think, because it invites a rethinking of the binding theory’s treatment of at least some presupposition triggers, as I will argue in the Section on Second Thoughts about Presuppositions (and Sundry Other Matters). The concept of buoyancy itself is discussed at greater length in Geurts (2000b). Explaining Specificity In keeping with DRT orthodoxy, I regard indefinites as propertydenoting expressions that receive existential import when they occur in

144

BART GEURTS

argument positions. The main advantage of this division of labour is that it makes for a uniform analysis of indefinites occurring in argument positions and indefinite non-arguments, such as predicate nominals, for example. To illustrate, it allows us to maintain that a ventriloquist has the same meaning in both of the following sentences: (28) a. b.

Barney is a ventriloquist. Betty is married to a ventriloquist.

In (28a) as well as in (28b), a ventriloquist merely denotes a property, but only in the latter case is this property applied to a reference marker introduced by the verb. I will assume that, if this happens, the reference marker in question is labelled as new. There are various ways of accounting for this feature (if it is one), but that is a topic I don’t want to go into here. Unlike Reinhart, Van Geenhoven and many others, I deny that indefinites are ambiguous between a specific and a non-specific reading: indefinites always denote properties. If an indefinite occurs as an argument it may be construed as specific or non-specific depending on whether is backgrounded or not, which is to say that the choice is a pragmatic one. Of course, to say that a given aspect of interpretation is a pragmatic one is not to deny the possibility that it is conventionally marked in some languages. In this respect, specificity is in the same boat as definiteness, which is a pragmatic notion, too, and is conventionally marked in some, though by no means all, languages. Following the general consensus, I take it that by default indefinites are construed non-specifically, and the most natural way of accounting for this is by assuming that, all things being equal, an indefinite will tend to be construed as part of the foreground because it carries new information. I still deny, of course, that new information is always foregrounded, but it is only natural that the former status tends to be escorted by the latter. It is only under special circumstances that new information is backgrounded, and if this happens, the expression in question is specific. We are now all set to explain the main facts about specificity, beginning with the interaction between indefinites and (other) scopebearing expressions. We have just seen why indefinites prefer to be construed non-specifically; this is, I suggested, because they tend to be part of the foreground. But if they are backgrounded, the Buoyancy Principle applies, which is to say that, other things being equal, they will take wide scope, and only if all things aren’t equal will they take intermediate scope. This is precisely the order of preferences that we

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

145

wanted to account for. Secondly, the similarities between definites and specific indefinites fall into place, too, because both types of expressions convey backgrounded information. Thirdly, and by the same token, it is only to be expected that there will be languages which lump together specificity with definiteness, assigning the two functions a single article or case marker, say. On the present account, such conventional devices receive a straightforward interpretation: they signal that something is part of the background. Thus a vcv-prefix in Bemba, for example, isn’t ambiguous in any way; it just serves to indicate that the expression it attaches to is backgrounded. The partitive constraint is explained along the same lines. It is reasonable to suppose that, in an expression of the form ‘Det a of b’, the main duty of b is to help identify the intended a, and is therefore backgrounded (cf. e.g. Kuno, 1987). So, properly understood, the partitive constraint is not that b must be either definite or specific, but rather that it must be backgrounded. This explains why definites and specific indefinites can occur in partitive constructions, while quantifiers and non-specific indefinites cannot. Definites and Indefinites in Attitude Contexts Consider the following sentence: (29) Barney believes that the university hangman is after him. On the de dicto reading of the definite NP, this statement entails that Barney believes that the university employs a hangman, who is after him. Thus construed, the sentence does not imply that there is a university hangman, as it does on the de re reading, which says of the university hangman that Barney believes that he is after him. Now compare (29) with the following minimal variant: (30) Barney suspects that the university hangman is after him. Here we have the familiar de re and de dicto readings, as in (29), but now we discern a third type of reading, as well, which can be paraphrased like this: (31) Barney believes that the university employs a hangmani and he suspects that hei is after him (where a hangman is to be read de dicto). Intuitively speaking, this reading is de re and de dicto at the same time. It is de re in that, some way or other, the university hangman is not

146

BART GEURTS

interpreted as dependent on suspects, though it is confined to Barney’s beliefs – which is remarkable of course, because there is no explicit reference to Barney’s beliefs in (30). These observations about definite NPs extend not only to other presupposition-inducing expressions, but to indefinite NPs, as well: (32) Barney suspects that an evil spirit is haunting him. Here, too, we have three readings, one which is de re, another which is de dicto and a mixed reading which can be paraphrased thus: (33) Barney believes that there is an evil spiriti and he suspects that iti is after him (where an evil spirit is to be read de dicto). In Geurts (1998b, 1999) I have ventured to give an in-depth analysis of the interaction between attitude verbs and presupposition-inducing expressions, and as that account carries over to specific indefinites without further ado, I will only give a quick outline here. The main idea is that the meanings of attitude verbs such as suspect, hope, want and so on, are defined in terms of their subjects’ doxastic states (cf. Stalnaker, 1984; Heim, 1992). For example, to say that a suspects that j is to say, approximately, that among the worlds compatible with what a believes, those worlds which a considers to be more likely all make j come out true. More generally, doxastic states provide the context against which we attribute not only suspicions but also hopes, desires and so forth. In DRT terms this means, for (30) and (32), that a representation of Barney’s doxastic state acts as the intermediate context between the main context and the suspicion that is being attributed to him. Therefore, a presupposition triggered within the syntactic scope of suspect will project to the main context by default, or if that option is barred for some reason or other, it will go to the DRS representing Barney’s doxastic state. Similarly, a specific indefinite will either be construed in the main context or as part of Barney’s beliefs. Hence, the various readings available for (30) and (32) are predicted in a uniform fashion. I have shown elsewhere that attitude reports are a crucible for theories that always want to construe indefinites in situ; such theories, I have argued, even fail to separate between de re and de dicto readings (Geurts, 2000a). The type of interpretation discussed in the foregoing is problematic not only for in situ accounts, but also for theories that treat specificity as a scope phenomenon. For if specificity could be dealt with in terms of scope, we should find the classical de re and de dicto readings, but not the mixed construals exemplified by (32). Specific

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

147

indefinites cannot be construed in situ any more than presuppositional expressions can; movement of some sort is called for. Nor is specificity a matter of scope in the grammarians’ sense: it is a pragmatic phenomenon, again like presupposition.5 Summing Up It will be evident that this analysis of specificity owes a great deal to Van Geenhoven’s proposal. But my account improves upon Van Geenhoven’s by giving a coherent picture of the relation between specificity, on the one hand, and presupposition and definiteness, on the other, while foregoing the premise that indefinites are ambiguous between specific and non-specific readings. Apart from providing a principled way of dealing with specificity, the present theory offers another attraction as well, in that it may shed new light on matters not directly related to specificity. It is to these matters that we now turn.

Second Thoughts about Presuppositions (and Sundry Other Matters) Being an extension of standard DRT, the binding theory regards presuppositions as elements that would like to be bound by an antecedent. This is a view that agrees with pre-theoretical intuitions about the definite article, for example, but it doesn’t seem right for some other expressions and constructions that are standardly categorized as presupposition inducers. I want to propose that at least some of these are better viewed as instances of backgrounding. Lexical ‘Presuppositions’ Intuitively speaking, the notion that presuppositions are anaphoroid elements does not seem to be quite appropriate for dealing with lexical inferences like the following, which have often been said to be presuppositional in nature (here ‘c’ is to be read as ‘implies, intuitively speaking’): (34) a. b.

Leslie is a bachelor. cLeslie is a man.

5 Many authors would balk at this and insist that presupposition is a ‘discoursesemantic’ phenomenon. Thus far I have failed to grasp exactly what that is, but as discourse semantics seems to comprise large tracts of what I take to be pragmatics, the difference of opinion is merely verbal, I suspect.

148 (35) a. b. (36) a. b.

BART GEURTS Wilma managed to fry an egg. cIt was difficult for Wilma to fry an egg. Fred accused Barney of nepotism. cNepotism is a bad thing.

It is commonly held that (34a) presupposes (34b), and this claim seems justified by the observation that this inference tends to go through even when (34a) is embedded in non-entailing environments, such as: (37) Perhaps Leslie is a bachelor. A naive account of facts like this would be to suppose that the lexical content of bachelor falls into two parts: an assertional part which specifies that bachelor is truthfully predicated only of unmarried individuals, and a presuppositional part which says, among other things, that a bachelor is a man; of course, it is the second half of the content of bachelor that triggers the presupposition in (34a) and (37). There are several problems with this naive account. First, as it stands, this analysis implies that every occurrence of bachelor gives rise to the presupposition that the individual it is being applied to is a man, and therefore it predicts, for instance, that (38) Betty is allergic to bachelors. means something like, ‘Betty is allergic to unmarried individuals who are presupposed to be men’ – which is not what we want. The solution to this problem is fairly obvious: the word bachelor should only be allowed to trigger its presupposition when it is being used predicatively. But this seems to entail that bachelor is ambiguous between a presupposing and a non-presupposing reading, which is not exactly an appealing consequence. The second problem, which is related to the first, is the following. Suppose that it is encoded in the lexicon that predicating bachelor of some individual a carries with it the presupposition that a is man. Consider now how the words bachelor and man are related to each other: the former is a hyponym of the latter, and the only distinctive feature of the word bachelor is that it applies to unmarried individuals. But at the same time that is all we are saying, as opposed to presupposing, when we call somebody a bachelor. Could this be a coincidence? I think it is pretty clear that it is not. For one thing, other hyponyms behave alike: spinster presupposes ‘female’, woodpecker presupposes ‘bird’ and so on. For another, an intuitively plausible story about this phenomenon is readily available: if a speaker wants to announce that Leslie is unmarried and

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

149

has even the slightest doubt about Leslie’s sex he would say that Leslie is unmarried rather than risking (34a). I do not want to suggest that spelling out an explanation along these lines is going to be trivial, but it is obvious that if such an account could be made to work it would be much more attractive than the one we started out with, which says, in effect, that it is a lexical accident that (34a) presupposes (34b). There is yet another problem with the suggestion that (predicative) bachelor presupposes ‘adult male’. It is that this presupposition, if it is one, is evidently not the kind of thing that seeks to be bound in anything like the way anaphoric elements seek to be bound. This becomes quite apparent when one considers how the binding theory would deal with (37), for example: (39) a. b.

[x: Leslie(x), perhaps: [: male(x), adult(x), unmarried(x)]] [x: Leslie(x), male(x), adult(x), perhaps: [: unmarried(x)]]

Assuming that (39a) is the semantic representation associated with (37) by the grammar, the binding theory predicts that the presupposition triggered by bachelor is accommodated in the principal DRS, because it cannot be bound and there is no reason (let us suppose) why it should be accommodated locally. This yields the right interpretation (and as a matter of fact I don’t know of any counterexamples to this analysis of bachelor), but within the framework of the binding theory this analysis causes something of an embarrassment. The presupposition supposedly triggered by bachelor can never be bound, as there is nothing to bind, so this presupposition would be one that, by its very nature, must always be accommodated, and as I have argued in my discussion of Van Geenhoven’s account of specificity, that is practically a contradiction in terms. The presuppositions allegedly triggered by verbs such as manage and accuse (cf. examples (35) and (36)) are dubious, too, and partly for the same reasons. Most importantly, it just doesn’t seem to be plausible, from a pre-theoretical vantage point, that the inferences licensed by these verbs should be of an anaphoric nature, and this suspicion is strengthened by the observation that it is next to impossible to come up with examples in which these purported presuppositions must be interpreted by way of binding. My proposal is to deal with the lexical inferences in (34)–(36) in terms of backgrounding instead of presupposition. According to the theory developed in the last section, backgrounded material may be given (i.e. presupposed) but backgrounding isn’t wedded to givenness, and therefore new information may be backgrounded, too. This, it

150

BART GEURTS

seems to me, is precisely what we witness in the cases under discussion. For example, if a speaker utters (34a), it is likely that the essential bit of information he intends to convey is that Leslie is not married, not that Leslie is an adult male. Therefore, the information that Leslie is a man is backgrounded, which means, I have argued, that it will gravitate towards the principal DRS, by virtue of the Buoyancy Principle. Similarly, if someone utters (35a), he conveys (35b), but he doesn’t present this information as given (not necessarily, anyway). However, by using this particular expression, the speaker does indicate that the truth of (35b) is of less concern to him than the fact that Wilma fried an egg. Hence, even if (35b) isn’t given, we may assume that is backgrounded. The same, mutatis mutandis, for (36a). I have proposed that the lexical inferences in (34)–(36) be explained in terms of backgrounding. This is not to suggest, however, that these inferences are alike in all respects, because they aren’t. Speakers’ intuitions make a fairly clear distinction between (34), on the one hand, and (35) and (36), on the other. Most speakers would say that if Leslie is a woman, (34a) is false. Whereas, if it turns out to be easy for Wilma to fry an egg, then it is not so evident what we should say about (35a). This statement would be misleading, to be sure, but many speakers would hesitate to simply reject it as false; similarly for (36a). One might say that, in contradistinction to the lexical entailment in (34), the inferences in (35) and (36) are conventional implicatures, but in view of the notorious ill-definedness of the concept of conventional implicature, that would do little more than rephrase the problem. I don’t have particularly strong opinions on how the differences between (34) and (35)–(36) can be accounted for, nor am I convinced that this issue is extremely urgent. This, however, is as it may be, because what I proposed in the foregoing doesn’t entail that such differences couldn’t exist. But these observations reinforce the suspicion voiced in the Section on Foreground and Background, that there may be various ways of backgrounding, which may not all be equivalent. Presupposition versus Background Over the past few decades, but especially during the presupposition craze of the seventies, the label ‘presuppositional’ has been applied to such a bewildering variety of phenomena that the very notion of presupposition has become suspect, as the following passage from Neale (1990: 54) illustrates: A great range of disparate and unrelated phenomena has been dubbed ‘presuppositional’ over the years, but [ . . . ] it seems highly

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

151

implausible that any theoretically important notion will do justice to the full range of data that semanticists professing an interest in ‘presupposition’ seek to explain.

Needless to say, I am not entirely convinced that the second half of this claim is justified, but the first half certainly is. All too often, the concept of presupposition has been used, or rather abused, without even the shadow of justification. I have argued elsewhere that this abuse was caused at least in part because the diagnostic tests for presuppositionhood were (and still are) applied too carelessly, if they were applied at all (see Geurts, 1999). But in the light of the foregoing discussion I want to suggest that there may have been another factor as well, which is that the standard tests don’t allow us to make a clear distinction between presuppositional and backgrounding effects, and that at least some of the phenomena that have been categorized, to greater or lesser acclaim, as ‘presuppositional’ are better seen in terms of backgrounding. The lexical inferences discussed previously are relatively clear instances of this category, and further possible candidates for relocation will be discussed below. But first I want to raise the question how we are going to distinguish between genuine presuppositions and instances of backgrounding. This is not a trivial question because, as I hinted already, the standard litmus tests for presuppositionhood fail to distinguish between presupposition and backgrounding, as the following observations illustrate: (40) a. b. (41) a. b. (42) a. b. (43) a. b.

If Germany becomes a monarchy again, the king of France will get nervous. c There is a king of France. If there is a king of France, the king of France will get nervous. >> There is a king of France. If Leslie is rich, he is a bachelor. c Leslie is a man. If Leslie is a man, he is a bachelor. >> Leslie is a man. If the king of France gets nervous, his ministers get nervous, too. c There is a king of France. If the king of France gets nervous, then France must be a monarchy. >> There is a king of France. If Leslie is a bachelor, he is rich. c Leslie is a man. If Leslie is a bachelor, he is a man. >> Leslie is a man.

These observations suggest that there are no differences between the presupposition triggered by the definite NP ‘the king of France’, as in (40) and (42), and the lexical inference licensed by the noun bachelor, as in (41) and (43), and the parallels extend to all sorts of embedding

BART GEURTS

152

contexts. Nevertheless, I have argued, there are good reasons for believing that lexical inferences aren’t of a presuppositional nature. But none of these reasons provides us with a general criterion for discriminating between presupposition and backgrounding. According to the binding theory, presupposed information is presented as given, in the same sense that the antecedent of an anaphoric expression is given, and the theory’s treatment of presupposition is a generalization of DRT’s treatment of anaphora, which is based on the widely held view that an anaphoric expression serves to retrieve an element from the common ground. That is to say, the speaker employs an anaphor not merely to signal that a discourse entity x is given, but also as an instruction to the hearer that he should identify and recover the intended x, so that new information will have the right connections. In other words, the hearer is expected to ask himself which entity the speaker has in mind. I want to suggest that we can turn this observation into a useful test for distinguishing between real presuppositions and merely backgrounded information. The test goes as follows: If x is a genuine presupposition, then it should make sense to ask ‘Which x do you mean?’ when the speaker has just uttered a sentence implying the existence of some x. This admittedly informal criterion indicates that, for example, the following are genuine presupposition inducers: 

Pronouns:

(44) 

Definite NP’s:

(45) 

A: The banana has been stolen. B: Which banana has been stolen?

Quantifier domains:

(46)



A: He is insane. B: Who is insane?

A: Every girl has sent me a postcard. B: Which girls have sent you a postcard?

Focus particles:

(47)

A: Professor Babel has read my paper, too. B: Who else has read your paper?

On the other hand, there are various alleged presupposition inducers that fail the wh-test. The lexical inferences discussed in the

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

153

previous section are a case in point, as are factive verbs and transition verbs, for example, which are standardly listed among the presupposition-inducing expressions: 

Factives:

(48) a. Barney is proud that his daughter is an anarcho-syndicalist. b. c Barney’s daughter is an anarcho-syndicalist. 

Transition verbs:

(49) a. Betty has started taking saxophone lessons [at time t]. b. c Betty wasn’t taking saxophone lessons [before t]. The inferences in (48) and (49) originate with the factive be proud and the transition verb start, respectively, and they both exhibit the projection behaviour that is characteristic of presuppositions. But they also fail the wh-test. In the first case it would make no sense to ask which state or fact (or whatever) involving his daughter Barney is proud of, and in the second case no hearer would ever wonder which instance of Betty-not-taking-saxophone-lessons ended at time t. Hence, if the wh-test is to be trusted, the inferences exemplified by (48) and (49) aren’t genuine presuppositions, and therefore they must be explained in terms of backgrounding. Zeevat (1992) has proposed a classification of presuppositioninducing expressions which resembles my somewhat tentative distinction between genuine presupposition inducers and expressions licensing inferences that are best understood in terms of backgrounding. Zeevat’s ‘resolution triggers’ correspond to what I call ‘presuppositions’ simpliciter ; his ‘lexical triggers’, to what I prefer to treat as backgrounding expressions.6 It would take me too far afield to discuss the theory Zeevat erects on his classification, but I would like to briefly comment on one of his empirical claims, which, if correct, might be put to use for discriminating between presupposition inducers and backgrounding expressions. Zeevat views lexical triggers as ‘applicability conditions’ which must be satisfied locally, i.e. in situ; and this constraint does not hold, according to Zeevat, for resolution triggers. It follows from this that resolution triggers can, and lexical triggers

6 The correspondences are not quite perfect, though. More recently, Zeevat (2002) has presented a new and finer-grained classification of presupposition triggers, which is rather less similar to what is proposed here than in his 1992 version.

BART GEURTS

154

cannot, get de re construals. The following example illustrates both predictions: (50) Betty believes that the superintendent is a bachelor. If this statement is true, Betty can hardly fail to believe that the superintendent is a man (which is the lexical inference triggered by bachelor), but it may well be that she is not aware that the person in question is a superintendent (which is part of the presupposition triggered by the definite NP). Unfortunately however, for Zeevat as well as myself, this distinction is not as neat as it initially appears to be. Suppose that all Betty knows about the superintendent is that he or she is not married. Would (50) be true or false, under these circumstances? Speaking for myself, I believe I might accept the statement as true, but even if other speakers should disagree, they would still have to concede, I think, that the matter is not as clear-cut as it seemed to be at first. When we turn away from the standard bachelor-type cases, it becomes even clearer that Zeevat’s observation is hard to maintain. Suppose Fred tells his friend Barney: ‘Wilma fried an egg this morning.’ Whereupon Barney reports to his wife: (51) Fred believes that Wilma managed to fry an egg. Tendentious though it may be, this statement is clearly correct, and it need not imply that Fred believes that it is (or was) difficult for Wilma to fry an egg. Therefore, if Zeevat’s diagnostic applied across the board, this inference could not be a lexical presupposition (in Zeevat’s terminology) or backgrounded information (in mine). I don’t know how Zeevat would want to deal with this inference, but since I want to treat it as an instance of backgrounding, I cannot employ attitude contexts for distinguishing between presuppositions and backgrounded information. Factives Factive verbs are standardly regarded as presupposition-inducing expressions, although there is a well-known problem with this view. It is that some factive verbs, at least, do not always seem to trigger the presupposition that their complement is true: (52) a. b.

If Barney should discover that Miss Chambley is rich, he’ll propose to her. If I should discover that Miss Chambley is rich, I’ll propose to her.

Both (52a) and (52b) can be consistently uttered by a speaker who doesn’t want to commit himself as to whether Miss Chambley is rich,

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

155

but unlike (52b), (52a) appears to have a further reading, as well, implying that Miss Chambley is rich. In view of observations such as these it has been suggested that discover belongs to a special class of ‘semi-factive’ verbs, which are ambiguous between a presupposing and a non-presupposing reading. This unattractive assumption can be avoided if we approach the matter in somewhat different terms. If the complement of a factive verb can be either backgrounded or not, the Buoyancy Principle predicts that something very much like presupposition projection will occur in the former case but not in the latter. This view is an attractive one, I believe, because it seems to correlate with our intuitions about foreground versus background in factive constructions. For example, a speaker who utters (53a) may be interested primarily in the fact that Barney knew (53b) or in the fact that (53b) is true. In the former case, the information in (53b) is backgrounded; in the latter, it is foregrounded. (53) a. b.

Barney knows that his daughter is an anarcho-syndicalist. c Barney’s daughter is an anarcho-syndicalist.

Now if the same options are available in (52), we predict that backgrounding the proposition that Miss Chambley is rich will imply that Miss Chambley is rich, whereas this inference will not go through if the factive complement is foregrounded. These predictions appear to be correct. Concluding Remarks In the preceding pages I have argued that a number of expressions that are standardly categorized as presupposition inducers are better viewed as backgrounding devices. I suspect, furthermore, that this viewpoint may be of more general use, and that it may help to account for phenomena which have not as yet received a satisfactory treatment. Let me mention just two, rather disparate, examples: 

Non-restrictive relative clauses:

(54) a. b. 

Fred suspected that Betty, who had been avoiding him of late, had discovered about his collection of Neil Sedaka albums. c Betty had been avoiding Fred of late.

Felicity conditions on speech acts:

(55) a. Where is my bicycle? b. c The speaker doesn’t know where his bicycle is.

156

BART GEURTS

Although it has occasionally been suggested that these inferences are of a presuppositional nature, this position has not gained much support in the literature (exceptions are Fillmore, 1969; Keenan, 1971). Still, both types of inference seem to exhibit the ‘wide scope’ tendency that is the hallmark of presuppositions. This is harder to demonstrate for felicity conditions on speech acts than for non-restrictive relatives, because non-declaratives dislike being embedded under operators of any kind. But at least we have conditional speech acts: (56) a. If my pogo stick is in the attic, where is my bicycle? b. c The speaker doesn’t know where his bicycle is. That non-restrictive relative clauses behave similarly is easier to show, for instance, by embedding (54) under a weak modal operator, such as perhaps. Nevertheless, it seems to me that the majority view is correct, and that the inferences exemplified by (54) and (55) shouldn’t be granted the status of presuppositions. In particular, the preferred interpretation of non-restrictive relatives is plausibly explained in terms of backgrounding: non-restrictive relatives are parenthetical remarks, which are backgrounded if anything is. So the Buoyancy Principle surely applies to non-restrictive relatives, and I conjecture that it applies to felicity conditions on speech acts, too.

References Abbott, B. 2000. Presuppositions as nonassertions. Journal of Pragmatics 32:1419–1437. Abusch, D. 1994. The scope of indefinites. Natural Language Semantics 2:83–135. Cresti, D. 1995. Indefinite Topics. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA. Farkas, D.F. 1981. Quantifier scope and syntactic islands. Papers from the Seventeenth Regional Meeting of the Chicago Linguistic Society, pp. 59–66. Fillmore, C. J. 1969. Types of lexical information. In F. Kiefer, ed., Studies in Syntax and Semantics, pp. 109–137. Dordrecht: Reidel. Fodor, J. D. and I. A. Sag. 1982. Referential and quantificational indefinites. Linguistics and Philosophy 5:355–398. Foley, W. A. 1994. Information structure. In R. E. Asher, ed., The Encyclopedia of Language and Linguistics. Vol. 3, pp. 1678–1685. Oxford: Pergamon Press.

SPECIFIC INDEFINITES, PRESUPPOSITION

AND

SCOPE

157

Foley, W. A. and R. D. van Valin. 1985. Information packaging in the clause. In T. Shopen, ed., Language Typology and Syntactic Description, vol. I: Clause Structure, pp. 282–364. Cambridge: Cambridge University Press. Geurts, B. 1998a. The mechanisms of denial. Language 74:274–307. Geurts, B. 1998b. Presuppositions and anaphors in attitude contexts. Linguistics and Philosophy 21:545–601. Geurts, B. 1999. Presuppositions and Pronouns. Oxford: Elsevier. Geurts, B. 2000a. Indefinites and choice functions. Linguistic Inquiry 31:731–738. Geurts, B. 2000b. Buoyancy and strength. Journal of Semantics 17:315–333. Geurts, B. and R. van der Sandt. 1999. Domain restriction. In P. Bosch and R. van der Sandt, eds., Focus: Linguistic, Cognitive, and Computational Perspectives, pp. 268–292. Cambridge: Cambridge University Press. Givo´n, T. 1978. Definiteness and referentiality. In J. H. Greenberg, C. A. Ferguson, and E. A. Moravcsik, eds., Universals of Human Language, vol. 4: Syntax, pp. 291–330. Stanford: Stanford University Press. Haspelmath, M. 1997. Indefinite Pronouns. Oxford: Oxford University Press. Heim, I. 1983. On the projection problem for presuppositions. Proceedings of the West Coast Conference on Formal Linguistics 2:114–126. Heim, I. 1992. Presupposition projection and the semantics of attitude verbs. Journal of Semantics 9:183–221. Horn, L. R. 2000. Assertoric inertia. Paper presented at the LSA meeting, Chicago, IL. Kamp, H. 1981. A theory of truth and semantic representation. In J. A. G. Groenendijk, T. M. V. Janssen, and M. B. J. Stokhof, eds., Formal Methods in the Study of Language, pp. 277–322. Mathematical Centre Tracts 135, Amsterdam. Reprinted in J. A. G. Groenendijk, T. M. V. Janssen, and M. B. J. Stokhof, eds., 1984. Truth, Interpretation, and Information. Selected Papers from the Third Amsterdam Colloquium, pp. 1–41, Dordrecht: Foris. Kasher, A. and D. M. Gabbay. 1976. On the semantics and pragmatics of specific and non-specific indefinite expressions. Theoretical Linguistics 3:145–190. Keenan, E. L. 1971. Two kinds of presupposition in natural language. In C. J. Fillmore and D. T. Langendoen, eds., Studies in Linguistic Semantics, pp. 45–54. New York: Holt. Kratzer, A. 1995. Scope or pseudoscope? Are there wide-scope indefinites? In S. Rothstein, ed., Events and Grammar, pp. 163–196. Dordrecht: Kluwer. Krifka, M., F. J. Pelletier, G. N. Carlson, A. ter Meulen, G. Link, and G. Chierchia. 1995. Genericity: An introduction. In G. N. Carlson and F. J. Pelletier, eds., The Generic Book, pp. 1–124. Chicago, IL: University of Chicago Press. Kuno, S. 1987. Functional Syntax: Anaphora, Discourse and Empathy. Chicago, IL: University of Chicago Press.

158

BART GEURTS

Ladusaw, W. 1982. Semantic constraints on the English partitive construction. Proceedings of the 1st West Coast Conference on Formal Linguistics, pp. 231–242. Lyons, C. 1999. Definiteness. Cambridge: Cambridge University Press. Manga, L. 1996. An Explanation for Ergative versus Accusative Languages: An Examination of Inuktitut. Doctoral thesis, University of Ottawa, Ottawa. Matthewson, L. 1999. On the interpretation of wide-scope indefinites. Natural Language Semantics 7:79–134. Neale, S. 1990. Descriptions. Cambridge, MA: MIT Press. Prince, E. 1981. On the inferencing of indefinite-this NPs. In A. K. Joshi, B. L. Webber, and I. A. Sag, eds., Elements of Discourse Understanding, pp. 231–250. Cambridge: Cambridge University Press. Reinhart, T. 1997. Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20:335–397. Stalnaker, R. C. 1984. Inquiry. Cambridge, MA: MIT Press. van der Sandt, R. 1992. Presupposition projection as anaphora resolution. Journal of Semantics 9:333–377. van der Sandt, R. and B. Geurts 2001. Too. Proceedings of the 13th Amsterdam Colloquium. University of Amsterdam, Amsterdam. Van Geenhoven, V. 1998. Semantic Incorporation and Indefinite Descriptions. Stanford, CA: CSLI Publications. Van Rooy, R. 1999. The Specificity of Indefinites. MS thesis, University of Amsterdam, Amsterdam. Winter, Y. 1997. Choice functions and the scopal semantics of indefinites. Linguistics and Philosophy 20:399–467. Yeom, J.-I. 1998. A Presuppositional Analysis of Specific Indefinites: Common Grounds as Structured Information States. New York: Garland. Zeevat, H. 1992. Presupposition and accommodation in update semantics. Journal of Semantics 9:379–412. Zeevat, H. 2002. Explaining presupposition triggers. In K. van Deemter and R. Kibble, eds., Information Sharing. Stanford: CSLI.

6

Deixis, Binding and Presupposition NICHOLAS ASHER

Introduction Dynamic semantic accounts of presupposition incorporate an important and novel element that links presupposition with anaphora (van der Sandt, 1992; Geurts, 1995). This approach, which accounts for a number of important facts about presupposition, extends to integrate bridging and other discourse phenomena (Asher and Lascarides, 1998a, b). In this extended anaphoric account, presuppositions, like assertions, attach to the discourse context via certain rhetorical relations. These discourse attachments constrain accommodation and help avoid some infelicitous predictions of standard accounts of presupposition. Further, they have interesting and complex interactions with underspecified conditions that are an important feature of the contributions of most presupposition triggers. Deictic uses of definites, on the other hand, seem at first glance to fall outside the purview of an anaphoric theory of presupposition. Deictically used definites simply denote, perhaps through some additional help from the context, the individuals in the context that satisfy them. There seems little that a discourse-based theory would have to say. I will argue, however, that an extended, anaphoric account has an important role to play in analysing such uses of definites, and it can capture how these uses of definites function in conversation. In particular such accounts can clarify the interaction between the uses of such deictic definites and various conversational moves. At least Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

159

160

NICHOLAS ASHER

some deictic uses of definites generate presuppositions that are bound to the context via a rhetorical function that I’ll call anchoring, which if successful entails a type of knowing how. If this anchoring function is accepted, then the acceptors know how to locate the referent of the definite in the present context. I’ll concentrate here just on definites that refer to spatial locations, where the intuitions about anchoring are quite clear. But I think that this view extends to other deictic uses of definites and has ramifications for an analysis of de re attitudes as well.

Different Ways to Bind Presuppositions To set the stage for an analysis of anchoring uses of definites and the role that their presuppositions play there, it is useful to see how many roles an extended anaphoric theory takes the presuppositions of definites to play in standard anaphoric uses. According to ‘‘Dynamic’’ accounts like Heim’s familiarity theory (1982), definites presuppose familiar discourse referents. Such presuppositions must be satisfied in the discourse context in the Tarskian sense or must be accommodated (i.e. added) to the discourse context. Van der Sandt (1992) tells us to find these discourse referents via anaphora resolution, i.e. try to bind, and failing that, accommodate. Geurts (1995) extends this view by including propositional identity as a means of binding. Nevertheless, there is much more that can be said about binding. Consider bridging examples like the following: (1) a. b.

I met two interesting people last night at a party. The woman was a member of Clinton’s Cabinet.

(2) a. John took engine E1 from Avon to Dansville. b. Then he picked up the boxcar c. and took it to Broxburn. Let’s consider what happens to the presupposition generated by the boxcar in (2b). On standard accounts, we can’t bind the boxcar to any discourse entity in the context nor can we satisfy the presupposition that there is a boxcar in the discourse context created by the first sentence. So all the standard theories of presupposition would say that we should accommodate a boxcar in the context. This misses an important aspect of the meaning of this discourse, which we can focus on by asking ourselves, where is the boxcar? The boxcar is manifestly in Dansville. A discourse-based account like that in Asher and Lascarides (1998a) gets this interpretation of (2b) by assuming that the presuppositional material introduced by the definite contains some

DEIXIS, BINDING

AND

PRESUPPOSITION

161

underspecified elements. In particular, we suppose that definite descriptions introduce an underspecified relation, the bridging relation, between the referent of the definite and some other contextually given entity. We further stipulate that the bridging relation is set to identity if this produces a well-defined result (thus incorporating the insight by van der Sandt that binding is preferred). In (2b), however, there is no non-absurd identification of the boxcar with some other discourse entity to be had. However, in (2b), there is a discourse particle or adverbial then that determines the discourse relation between (2a) and (2b) to be one of Narration. The presence of such a discourse relation between (2a) and (2b) entails that the event described in (2b) is understood as coming after the event described in (2a) and as spatially located in the location in which the event in (2a) terminates (Dansville). The lexical semantics of pick up adds the information that in effect the boxcar is in Dansville. This suffices to determine the bridging relation in this case to be ‘in’. Thus, the boxcar is linked to Dansville and that is enough to satisfy the presupposition in this discourse context and to get the right interpretation. Details of the analysis can be found in Asher and Lascarides (1998a). With (1), we also see a need to supplement both the Heim and the van der Sandt-Geurts approach to presupposition. Here too we can’t bind the woman to any discourse entity in the context nor can we satisfy the presupposition that there is a woman in the discourse context created by the first sentence. So all the standard theories of presupposition would say that we should accommodate a woman in the context. But this misses an essential component of the interpretation of (1): the woman is one of the two people that I met last night. The discourse-based anaphoric account of presupposition gets this interpretation of (1) by a simultaneous resolution of the underspecified bridging relation and a computation of the discourse connections between the presupposition generated by the woman, the asserted component of (1b) and (1a) accounts for this anaphoric connection and the coherence of the text. More specifically, in this case again, specifying the bridging relation to identity yields an absurdity. But if we specify the bridging relation to be ‘‘an element of ’’, we get a coherent discourse and a discourse relation of elaboration between (1a) and (1b). Alternatively, specifying the discourse relation to be Elaboration will coerce the underspecified bridging relation to the appropriate value.1

1 See Asher and Lascarides (1998a) for more details, and also the Section on SDRT’s Account of Presupposition.

162

NICHOLAS ASHER

Other examples of complex presupposition binding occur when the presupposition trigger, the expression that generates the presupposition, is itself a discourse particle. That is the case with (3): (3) John lives in New York too. Kripke (2009) observes that (3) can’t be uttered in a null context, even if many people are known by the speaker and the audience to live in NY. But accounts like van der Sandt’s and Heim’s don’t predict this. On the other hand, the discourse-based account can, if we assume that the presupposition of too is that it generates a specific rhetorical function connecting the asserted content of (3) to some element of the contextually given discourse structure. In the null context there is no element of discourse structure to connect to, and so the presupposition of too can’t be fulfilled, and we are able to explain Kripke’s observation. A final example of binding with rhetorical relations reveals that not only are the Heim-van der Sandt-Geurts accounts of presupposition incomplete but they derive wrong interpretations. Consider (4). (4) a. If a farmer goes to the market, he buys a donkey. b. Yesterday, Farmer John went to the market. c. The donkey he bought was expensive. d. This time the donkey was expensive. e. This time (?)itdonkey was expensive. Van der Sandt and Geurts must accommodate the existence of a donkey in order to interpret the presuppositions of (4c,d,e). But the donkey in (4) depends on an anaphoric link between bought and went. Accommodation yields incorrect results. Heim’s (1983) theory yields only the satisfaction of an existential presupposition, not an anaphoric one. We need an appropriate instantiation of the conditional (like (4au)) (4au) If John is a farmer and went to market, John bought a donkey. which, when coupled with (4b), gives the donkey referred to in (4c). (5) If a farmer goes to Paris, he buys a donkey. Pedro went to Paris. His donkey was expensive. (6) A farmer buys a donkey whenever he visits the market. Farmer John visits the market on Wednesdays. His donkeys are merry. Similarly, for adding an argument for going-to-the-market events. We might call this inferential binding. The inferential binding in

DEIXIS, BINDING

AND

PRESUPPOSITION

163

(4a,b,c) falls squarely within the analysis given to the bridging examples and to our anaphoric theory of presuppositions. The rhetorical relation that binds the presupposition to the discourse context is the relation of Defeasible Consequence (Asher and Lascarides, 1998b), a natural generalization of Geurts’s notion of propositional binding. Defeasible Consequence holds of two propositions p, q iff q is a defeasible consequence of p.2 This relation will bind the presupposition to the conjunction of (4a) and (4b), as both are necessary to derive the presupposition that Pedro owns a donkey. In these examples, as in (2) and (1), we cannot specify the bridging relation itself to identity without absurd results. So the semantics of the rhetorical relation used to bind the presupposition must determine the bridging relation. The use of defeasible consequence to link the presupposition to (4a,b) naturally suggests that the bridging relation be set to a witnessing relation W. While one term of W should be the discourse referent introduced by the definite, it is less clear what the other term should be. If we allow reference to quantifiers themselves or their logical forms (it is after all one type of abstract object), then we can take the other term of W to be the quantifier of which the definite produces an instance. In (4), the appropriate quantifier is a donkey in (4a).

‘‘Deictic Binding’’ and Discourse Function Not all uses of definite descriptions fit comfortably into an anaphoric theory of presupposition. Definite descriptions have deictic uses within ordinary conversation. Let’s take some simple examples: (7) a.

Now pour the mixture into the pan and gently simmer for 10 mins. b. Move the window to the lower left (on a computer screen). c. Close the window in the bedroom. d. (You’ve just checked into a hotel and the clerk says:) Your room is up the stairs and right at the end of the corridor.

Many of these definites occur in contexts where, e.g., a window on the computer screen has already been introduced in a previous discourse turn. So the presupposition would be satisfied here by linking the discourse referents introduced by the two NPs. On the other hand, 2 Defeasible consequence is defined precisely via a nonmonotonic logic. See for instance Asher and Morreau (1991) or Lascarides and Asher (1993).

164

NICHOLAS ASHER

this mere anaphoric connection isn’t sufficient to carry out the conversational purpose behind these instructions. In order, e.g., to carry out the instruction in (7a), the addressee needs to be able to find the referent of the description. Similarly for (7b,c). The discourse referents introduced by the definites have to be linked or anchored to particular nonlinguistic elements in the visual nonlinguistic context. The case in (7d) is a bit different, but in a way it’s more interesting. Consider the use of the definite the stairs. On traditional accounts, we would simply accommodate that there is a corridor on an anaphoric account. But this misses the intended interpretation – viz. that the stairs be linked to some object in the environment that accomplishes the manifest goal of the speaker, which is that the addressee knows how to get to his room. As one would expect, a standard, dynamic account of presupposition, which treats the presupposition of the definite in (7d) simply by adding it to the context, misses the rhetorical function assigned to it by the speaker. This rhetorical function of the presupposition for the speaker in a given context is part of what drives the conversation. Consider what happens when this rhetorical function isn’t shared by the interpreter or addressee. If the addressee cannot locate the stairs, for instance, he can’t accomplish his goal and it’s quite appropriate for him to say: (7du) Where are the stairs? We saw earlier that the presupposition of a definite description when resolved can help determine a rhetorical function for the asserted content of a sentence (viz. (1b)), and it seems as though the presupposed material here too has an important role to play in this rhetorical function. But what exactly does it do? It’s helpful to consider the behaviour of such deictic uses of definites within longer stretches of dialogue. Consider the uses of definite descriptions referring to locations or spatially localized entities in the following two dialogues. (8) a. I: Philippe?(i) C’est Isabelle(ii). Bon, je crois que je me suis un peu perdue(iii). `? b. P: Ah bon? Et tu m’appelles d’ou c. I: Ben, j’en sais rien! Je ne vois pas le nom de la rue. d. P: Mais tu as fait comment pour y arriver? e. I: Non, en fait, je suis tout au bout de I’ avenue Jean Jaure`s, apre`s la grande place (i). C’est une grande rue et il y a un panneau qui indique ‘‘Aubervilliers’’ tout droit (ii). Je ne suis pas sur le bon chemin, non? (iii)

DEIXIS, BINDING f. g. h. i. j. k. l. m.

n.

AND

PRESUPPOSITION

165

` tu es (i). Tu as pris sous la P: Attends, je ne vois pas trop ou voie ferre´e, comme je t’avais dit? (ii) I: D’ici on voit une voie ferre´e, au dessus de la place. P: Au dessus de la place? Non, c- a peut pas ˆetre c- a! ` tu habitais, c’e´tait bien Jean Jaure`s? Et I: Ecoute, l’avenue ou bien, la grande place est juste apre`s, avec le pont suspendu. P: Non, mais c- a, c- a doit ˆetre le pe´riph et pas la voie ferre´e! I: Ben, peut-eˆtre, d’ici c- a se voit pas. P: Tu es `a coˆte´ de la cite´ de la musique? I: J’en sais rien, je connais pas la cite´ de la musique (i). Mais attends, est-ce qu’il fallait que je prenne un deuxie`me tunnel apre`s celui de chez toi ? (ii) P: Bien su ˆr, c’est la` que tu passes sous la voie ferre´e.

In this dialogue, Isabelle is trying to find her way to Phillippe’s new apartment in Paris. There are many definites referring to locations or landmarks; many cannot be bound anaphorically to previous items mentioned in the discourse. The turn (81-m) is quite interesting. la cite´ de la musique is a novel definite, with which Isabelle is obviously not familiar, since she says that she doesn’t know the cite´ de la musique. According to standard anaphoric theories of presupposition, one would at this point accommodate the presupposition of the definite, and the discourse would proceed smoothly. But that’s not what happens; the discourse doesn’t proceed smoothly. Isabelle feels as though it entails a presupposition failure, i.e. it indicates that the presupposition cannot be satisfactorily treated. With such presupposition failure, there is no possibility of carrying out commands, answering questions, or the like. It has to be the same flavour as orthodox presupposition failures in questions: (9) What does the present King of France do on Bastille Day? Just as Isabelle says exasperatedly, ‘‘I don’t know the Cite´ de la Musique’’ (8m), one could respond to this question with a similar exasperated correction: there is no present king of France! Were accommodation possible here, Isabelle would still not be able to answer the question Phillippe has asked her (Asher et al., 2001). If the accommodation of the presuppositions of novel uses of definite descriptions is not always an option, why is that so? And can we understand a successful treatment of the presuppositions of such definites as a ‘‘binding’’ of the presuppositions and bad ones where no binding is possible? This would seem to be the only option for the anaphoric theory of presupposition to handle these cases. If we go back

166

NICHOLAS ASHER

to the dialogue, there are uses of definites that could be bound via identity to a previously mentioned occurrence as in (8f.ii) P: Tu as pris sous la voie ferre´e comme je t’avais dit? But interestingly Isabelle does not use this binding alone. In order to be able to answer the question, she must be able to identify the railroad tracks in her immediate environment or as something she passed on her journey. She has to ‘‘anchor’’ the definite to some object in the (nonlinguistic) context. To that end, she offers up an object in her perceptual context that might be the appropriate anchor for la voie ferre´e. (8e) I: D’ici on voit une voie ferre´e, au dessus de la place. Interestingly again, Phillippe rejects this contextual anchoring of the railroad tracks in (8h); he identifies what she sees as the Boulevard Pe´riphe´rique. Exactly what is the nature of this anchoring? Clearly, it involves some sort of de re attitude towards the object, but this observation by itself doesn’t illuminate an important link between anchoring of a definite in an utterance and the conversational goals of the utterance or of utterances linked to it. Consider (8e.i) I: Non, en fait, je suis tout au bout de l’avenue Jean Jaure`s, apre`s la grande place. au bout de l’avenue Jean Jaure`s is a definite with a novel use. In another context accommodation might suffice, but here it does not. And the reason is that mere accommodation or even a simple de re belief of the sort one gets when one has a belief about Byron won’t lead to the satisfaction of the particular conversational goals of finding out where Isabelle is. Or perhaps, since Phillippe lived in that neighbourhood, the location denoted by the end of the avenue Jean Jaure`s could be bound to some doxastically accessible conceptual individual or discourse referent in Phillippe’s beliefs. This would be a strategy for treating the presupposition available to anaphoric theories of presupposition that integrated a dynamic semantic view of an individual’s doxastic state (e.g. in Asher, 1986 and Kamp, 1990). But Phillippe’s response to Isabelle (8f.i) makes clear that such a binding isn’t sufficient for him to attain his speech act related goal of knowing where Isabelle is. In order to satisfy his conversational goals, he needs a contextual anchoring that will support a de re knowledge claim. What would suffice is a binding of

DEIXIS, BINDING

AND

PRESUPPOSITION

167

the presupposed location to some location that he is familiar with and can locate on his ‘‘cognitive map’’. And in fact this is what Isabelle wants to do too in view of the goal they both have of getting Isabelle unlost. So anchoring does involve a de re attitude on the part of the interpreter, but the de re attitude should be understood not as a knowing that but a sort of knowing how that is needed to solve the relevant conversational goals. That anchoring must be done in order to achieve the conversational goals is also well-attested in the map corpus dialogues compiled at Edinburgh University. I give a sample here. (10) a. A: Start at the extinct volcano, and go down round the tribal settlement. And then b. B: Whereabouts is the tribal settlement? c. A: It’s at the bottom. It’s to the left of the a e extinct volcano. d. B: Right. How far? e. A: m Ehm, at the opposite side. f. B: To the opposite side. Is it underneath the rope bridge or to the a lef g. A: It’s underneath the rope bridge. And then from the tribal settlement go straight up towards the rope bridge and over the rope bridge. Then down three steps and along to above the volcano. h. B: x Eh, a d . . . Is down three steps below or above the machete? i. A: x Ah. The machete’s not on my map. j. B: x Oh. k. A: Down three lines. l. B: Right. m. A: And then along as far as the volcano but above it, and stop underneath the collapsed shelter but away from it a bit. n. B: Right. o. A: And go up to about the middle of the map. p. B: The middle of the map. q. A: And stop. r. B: Just slightly above the crevasse? s. : [s] A That’s not on my map either. m Ehm, go to your left again into about the middle. t. B: I think that would bring me over the crevasse. u. A: Well, it’s not on my map. v. B: No? x Oh.

168

NICHOLAS ASHER

In the MAP Task Corpus dialogues, agent A is trying to get B to a given goal in a game, in which both A and B have a partially accurate map of the terrain to be traversed. In this dialogue the instructions to move to a particular place can only be carried out once either that place or obstacles to be circumvented have been anchored in an appropriate way to their maps. When this anchoring doesn’t take place, then the agents can’t give or carry out directions and they have to settle on another means for conveying directions. The directions can only be carried out once the locations involved are appropriately anchored. I want to draw several conclusions from the discussion of these examples. First, like other uses of definites, deictically used definites do generate familiarity presuppositions. Second these presuppositions are not accommodated or bound in the way familiar from standard, anaphoric theories of presupposition. But that doesn’t mean that we have to throw out the machinery of the anaphoric theory. In fact, the discourse-based anaphoric theory of presuppositions is very useful: we can understand contextual anchoring as a special sort of presupposition binding; in fact contextual anchoring is a rhetorical function of the presupposed information in these deictic cases. For the discourse to be felicitous, the presupposition generated by a deictically used definite must be tied to some object in the nonlinguistic context such that the interpreter believes he knows how to identify it or make use of that object for some conversationally salient discourse purpose. The upshot of such anchoring is a mutual belief between speaker and hearer that they are referring to the same object with the description. When this doesn’t happen, we get something that bears the discourse relation of Correction or some sort of question by the other participants, as Phillippe does in (8h), and as Isabelle does in (8m). Phillippe’s goal is to get de re knowledge, to know where Isabelle is. But this goal itself is subservient to another goal namely that of getting Isabelle unlost. Let’s assume for now that Phillippe has acquired this goal. And it is this higher goal that tells us what sort of de re knowledge this really is; Phillippe needs to know where Isabelle is so that he can give her directions to get her to her destination. And to give these directions he has to construct a path from Isabelle’s present location to her destination and to do that he has to be able to fix the present location (and her destination) on some cognitive map; or perhaps more simply he has to know how himself to get from where she is to where she wants to go. So this de re claim is grounded in a plan and finally in a capacity for action. It’s not knowledge that that’s indicative or even constitutive of de re attitude claims; it’s knowing how to realize a certain goal. Boer and Lycan (1986) propose that de re knowledge be

DEIXIS, BINDING

AND

PRESUPPOSITION

169

understood relative to purposes, a proposal that I take to be basically correct. But they still analyse de re attitudes in light of knowledge that – viz. knowledge of a proposition containing an attributive description. And while this is sometimes the case, it need not be; in the map task it may be the ability to point to a location or to put an agent in a particular location that constitutes knowledge de re of that location. Boer and Lycan are interested in stopping the ‘‘regress’’ of ‘‘who is X?’’ type questions. But in so doing they conflate the issue of de re knowledge claims with their justification. De re knowledge is just a matter of having access to the object that is sufficient for accomplishing the contextually given goals at hand. The upshot of our proposal for contextual anchoring amounts to the following view of de re attitudes: there isn’t any absolute de re knowledge; there’s de re knowledge relative to various goals that one might have. Previous Approaches to Contextual Anchoring The description given of the phenomenon of contextual anchoring of the presuppositions of definites is a quite different picture of deictically used definites than that found in the philosophical and linguistic literature. Here are some approaches that might be useful to combine with the Boer and Lycan like analysis of de re attitudes I have sketched above. contextual evaluations for indexicals and demonstratives (Kaplan, 1978)  value loading (Barwise and Perry, 1983)  DRT’s external and internal anchors (e.g. Asher, 1986; Kamp, 1990) 

Each one of these theories gives us an account of the satisfaction of the definites in a nonlinguistic context. Both Kaplan and Barwise and Perry suggest that a definite may be evaluated in the present context or, for Barwise and Perry, in any ‘‘conversationally salient’’ situation. This ‘‘value loading’’, or ‘‘externalist’’ type of evaluation yields a singular proposition for the sentence containing the definite. Such an account yields a connection between deictically used definites and de re attitudes, which seems needed to account for the rhetorical function of such definites. Once an agent accepts such a singular proposition or comes to believe it, he has a de re belief. The problem is that this act of acceptance and the de re attitude as an attitude towards a singular proposition doesn’t by itself link up easily to the conversational patterns we’ve already discussed. Consider again the position of Phillippe in (8f.i). Suppose that he accepts

170

NICHOLAS ASHER

Isabelle’s assertion whose interpretation yields a singular proposition. By accepting this assertion, he comes to have a belief, in this case a de re belief. But on the other hand, we’d like to distinguish this case from the sort of attitude that Phillippe requires in order to satisfy his conversational goals. For instance, in (8i), it’s clear from his response that Phillippe doesn’t have access to that contextual evaluation, which is something that these approaches can’t explain. And this lack of access drives his response to Isabelle’s assertion, and in particular her use of the definite au bout de l’avenue Jean Jaure`s. An alternative, ‘‘internalist’’ approach to singular propositions and the attendant construal of de re attitudes is to look for some internal, cognitive aspect of these attitudes. This is also a familiar idea in philosophy, also made famous by Kaplan – though this time it’s Kaplan’s (1968) paper ‘Quantifying in’. Kaplan’s idea was that a de re attitude involves a particular sort of name, a ‘‘vivid name’’, for the object and that name is a constituent of the attitude object. Vivid names for a particular attitude holder are ones that have a lot of information associated with them, perhaps information sufficient to identify the object. But, at least on this construal, vividness isn’t necessary for some de re knowledge claims. Knowledge who, for instance, is presumably a kind of de re attitude. So now consider the de re knowledge involved in knowing who lost the battle of Hastings for the purposes of a history exam. Here the name of a long dead Anglo Saxon king will suffice; what seems important in this case is not the amount of information at the disposal of the possessor of the attitude but the way that information interacts with the conversational goals at hand. The practical activity in the example about King Harold is just being able to supply the correct answer. We could reconstrue vividness in terms of knowing how, but we would still need to supplement this with an account of how this attitude towards the referent of the definite interacts with discourse. And we lack here any connection with accounts of presupposition. A DRT approach to deictically used definites could make use both of internalist and externalist components. In the terminology of Asher (1986) or Kamp (1990), a DRT approach to deictically used definites could make use both of external anchors to simulate the truth conditional effects of singular propositions and of internal anchors that could furnish additional descriptive conditions. The definite would introduce a discourse referent x that would be linked to some object a in the context via an external anchor, which would ensure that the satisfaction of the discourse representation must proceed by assigning a to x. Additionally, the binding of the presupposition generated by the

DEIXIS, BINDING

AND

PRESUPPOSITION

171

definite could take place via an internal anchoring of x to some discourse referent in a representation that is part of the agent’s cognitive state. Such internal anchors link the interpretation of one discourse referent x in one representation R1 to the interpretation of another discourse referent y in another representation R2 in the representational theory of cognitive states that has been elaborated within the DRT framework (Asher, 1986; Kamp, 1990); more precisely, we say that a pair of assignments f, g satisfies R1 and R2 respectively given an internal anchor between x and y iff f(x) ¼ g(y). Unlike the value loading accounts, this approach focuses on the cognitive aspect of these contextual anchorings. DRT approaches also give us an anaphoric account of presuppositions. Roughly, a definite description introduces a presuppositional component into the discourse representation in which a discourse referent is introduced along with the properties that are given by the description. This discourse referent must be linked to some other discourse referent in the context, unless the presupposition is to be accommodated. We can now combine the picture of deixis in the previous paragraph with the DRT approach to presuppositions: the discourse referent introduced by the presupposition generated by a deictically used definite binds via an internal anchor to some discourse referent in the interpreter’s cognitive state. Nevertheless, a DRT approach says little about the sort of knowing how that we’ve seen is important in the examples. The uses of definites in these dialogues establish that it’s the cognitive access for certain purposes that are crucial for de re attitude claims. Like Kaplan’s own picture of de re knowledge in ‘Quantifying in’, DRT’s conception of internal anchoring lacks any tie to practical plans; in fact there aren’t any constraints on internal anchors whatsoever, which might well accord with our intuitions about beliefs of a certain kind but which doesn’t capture the particular sense of de re attitude at issue here.3 The proposal I’ve just developed as it stands is still just binding, albeit to a belief context rather than a discourse context. We need some story here of familiarity that goes beyond binding. We need an account in which, e.g., the variable associated with Isabelle’s location is linked to some cognitively accessible discourse referent in a way that allows Phillippe to accomplish his conversational goals. I turn now to a view of presuppositions of definites on which the thesis about de re knowledge claims goes hand in hand with a goal

3

For a development of such a view outside of DRT circles, see Jeshion (2001).

172

NICHOLAS ASHER

relative notion of contextual anchoring. To do this I’ll sketch in a little more detail than before the SDRT theory of presupposition and how it interacts with a theory of cognitive modelling. I’ll then return to these contextual anchorings.

SDRT’s Account of Presupposition In the SDRT account of Asher and Lascarides (1998a, b), presuppositions are, like assertions, units of information that must be integrated into the discourse context. A unit of information, however, can be integrated into the discourse context in different ways, ways which correspond to the rhetorical function or functions of that unit of information. Accordingly this leads us to a more complex notion of a discourse structure than that present, say, in DRT. A discourse structure is a pair (A, F ), where: A is a set of labels F is a set of formulas representing clauses and relations on labels (between clauses)  F : A - F  

We’ll express the effects of F on A via the notation p : K. While both assertions and presuppositions must be integrated to the discourse context, presuppositions must be linked via particular discourse relations. Asher and Lascarides (1998b) isolated two, Background and Defeasible Consequence. While there is no accommodation per se in this framework since the attachment of presuppositions is just part and parcel of building a discourse structure, the cases of accommodation in the literature correspond to linking the presupposition via the relation of Background. Defeasible Consequence generalizes the propositional binding relation in van der Sandt and Geurts, while Background imposes thematic constraints that the notion of accommodation lacks. Background (p, q) holds iff q and p entail a common topic and q specifies properties of elements in p that set the stage for or serve as an explanation for some event described in p or in some proposition linked to p (6¼q). As discussed in the introduction, presuppositions are understood as containing incomplete or underspecified elements. This is particularly true in the case of the presuppositions of definites. In discussing earlier examples, I alluded to an innovation of the SDRT view that incorporates an underspecified bridging relation B in the presupposition of a definite. The representations of presuppositions underspecify the

DEIXIS, BINDING

AND

PRESUPPOSITION

173

discourse relation by which they attach to the discourse context and the other term of that relation; as such they are explicitly anaphoric elements whose proper interpretation must resolve all of these underspecifications, i.e. find appropriate specifications for these underspecifications. Here’s an example of how a presupposition and an assertion arising from a sentence containing a definite would be analysed in SDRT. The asserted component is labelled with a, the presupposed part with p. The asserted part produces a labelled SDRS for the asserted content of a clause; snoyhrt SDRS, the ‘‘p’’ part – with speech act discourse referent pu, and condition pu : Kpu for the presupposed content of this clause – represents the presupposition. (11) The man walked (11u)

In the SDRS above, the man denoted by the definite must be (bridging) related via B, whose content is underspecified (something expressed above by the ‘ ¼ ?’ notation) to an antecedent object that is also not specified by the meaning of the definite itself. These underspecifications make it hard to find a felicitous way to utter (11) in a null context. Further, the presupposition must be bound to the context via a rhetorical relation. While both presuppositions and assertions must get integrated into the context, they do so in different ways. Assertions can sometimes simply be added to the discourse context (as when we have a discourse initial assertion). But presuppositions cannot; the presence of the underspecified discourse relation R above and the underspecified attachment point v mean that these elements must be specified in order for the presupposition to be felicitous. Presuppositions link typically with either Defeasible Consequence or Background. We’ll add here the relation of Anchoring as another relation that presuppositions can bear to other elements in

174

NICHOLAS ASHER

the discourse structure. When the components above are attached to the discourse context and the various specifications of underspecified conditions are effected (as far as possible), then we have an update of the discourse context with the information given by (11). Following Asher and Lascarides (1998b), I’ll represent update by a three place relation involving the discourse context, the new information and a ‘‘resulting’’ SDRS that integrates the new information into the discourse context. In the introduction, I also mentioned that SDRT incorporates a principle of resolving B to identity whenever feasible, thus capturing the preference for binding that is a feature of anaphoric theories of presupposition. This accounts for the simple cases of binding in (12) Whenever I see a book in a bookstore that I like, I try to buy the book. We can formalize this principle as follows: 

If Possible Use Identity:   ðK b ½B ¼ ?^ # UpdateðK t; K a; K b ½B=lxly x ¼ yÞÞ ! t; a; b½B=lxly x ¼ y

This constraint says that as long as setting the bridging relation to identity is well defined (k) then the update of the discourse context with Kb will set B to identity. As we saw in (1) or (2), sometimes we cannot resolve B to identity. In that case what happens? When we try to resolve the bridging relation to something other than identity, we do so in a way that maximizes discourse coherence. Since the update relation is nondeterministic, there are often many ways new information can be integrated into a discourse context. Sometimes the resolution of underspecified elements as in (1) will determine how the new information attaches to the discourse context. Some of these ways provide for a more coherent discourse than not. Attachment and resolution of underspecified elements always tries to maximize discourse coherence. This is the constraint known as Maximize Discourse Coherence or MDC. To give this constraint some bite, I specify some things about the preference order g on discourse structures: More specified, well-typed SDRSs are always preferred to SDRSs with less specification.  SDRSs that violate type restrictions are less preferred than those that don’t violate such restrictions. 

DEIXIS, BINDING

AND

PRESUPPOSITION

175

Defeasible consequence g background for presupposed material. Linking a presupposed SDRS with background with a more specific topic g linking it with background with a less specific topic.  Where speech act related goals or SARGs can be inferred from Cognitive Modelling, a discourse structure that is more likely to lead to SARG satisfaction is more coherent than one that is not likely to lead to SARG satisfaction.  

All of these constraints on g require probably more explanation than I can give here. But here are some short glosses (for more details see Asher and Lascarides, 2003). The first constraint just says that an SDRS with fewer underspecifications where no type constraints on predicates are violated is to be preferred to an underspecified SDRS. The second constraint says that anytime a type restriction is violated that SDRS is less preferred to other SDRSs where the type restriction is not violated. The third constraint says that some discourse relations between presupposed material and the discourse context like Defeasible Consequence are to be preferred over a relation of background between the presupposed material and the discourse context (thus encoding an anaphoric theory of presupposition’s preference for binding over accommodation). The fourth constraint tells us that the tighter the connection between the background material and the foreground material, the better the discourse coherence between those two segments, as a tighter connection between background and foreground will allow for a narrower, or more specific, topic. Thus, in an example like (1) maximizing discourse coherence or MDC will prefer those SDRSs where the bridging relation in the presupposition of the definite is set to some relation other than identity since setting the relation to identity would require the identification of a couple with a woman, which violates type restrictions. But further setting the bridging relation to be ‘‘a member of ’’ is preferred on several counts: it specifies the underspecified relation and it also gives rise to a Background relation with a more specific topic than would be otherwise possible. For the inferential binding cases like (4), MDC will specify the bridging relation to the witness relation, which holds between a discourse referent and a discourse constituent or speech act referent, as we described earlier, because that will allow us to attach the presupposition with the relation of Defeasible Consequence to the SDRS consisting of the first two sentences of (4), and that is preferred to any option on which the presupposition is not so attached. Below I give a picture of how the specifications would go. Def-cons is the relation of Defeasible Consequence, Commentary is another relation in

176

NICHOLAS ASHER

which the speaker of the second constituent expresses an attitude towards some element in the first constituent. (13)

That finishes my gloss except for the last constraint on g. That has to do with cognitive modelling, the part of our story that I turn to next. Cognitive Modelling As we’ve seen, anchoring requires linking an epistemic attitude to conversational goals. Thus, we need to be able to infer conversational goals from conversational patterns. SDRT co-opts some of the insights of Gricean pragmatics and speech act theory to link speech act related goals or SARGs to discourse structure (Asher and Lascarides, 2003). On this view, the rhetorical relations in dialogue bring considerations about why participants ask, elaborate, request, assert and respond to what is said. In turn such SARGs help elucidate and further constrain discourse structure. In order to formulate a precise notion of anchoring for deictically used definites, I will give some of the principles for discovering SARGs in that component of SDRT that supplies a rough cognitive model of discourse participants. A second feature of anchoring is that once the anchoring function of a deictically used definite is accepted by the interpreter, it appears that speaker and hearer mutually believe that the definite picks out the same object. Given that we have adopted a largely internalist view of the de re attitude involved in anchoring and that the way dynamic semantics models attitudes has nothing to say about knowing how, I’ll show how such mutual belief can be derived from axioms having to do with the beliefs of the participants. Cognitive modelling in SDRT follows the basic BDI approach in which we have modal operators for belief (with a K45 logic) (B) and intention (I ) (with a KD logic), and a mutual belief operator MBG , for any group G with the usual axiomatization. We’ll assume

DEIXIS, BINDING

AND

PRESUPPOSITION

177

distributivity of B and I over Defeasable Implication W as well. I’ll suppose that BA pp , corresponds to A believing the propositional content represented in the SDRS Kp. Goals are propositions that one intends (a simplification but good enough for our purposes here). I’ll start with the simple Grice-like axioms for belief modelling. The first axiom allows us to infer beliefs from assertions. 

Sincerity: R(a, b) W BelAgent(b)R(a, b)

A second default, competence, transfers the beliefs of one agent to another, while the constraint on acceptance gets us from acceptances to beliefs about what others have said.  

Competence: BA f W BB f Constraint on Acceptance: Accepts(a, b) W MBAgentðbÞ a

Let’s now turn to the inference of SARGs. Inferences concerning SARGs also revolve around a Gricean notion of cooperativity. One agent B is cooperative with another agent A if he adopts A’s goals. According to this, B will try to realize A’s goals and in so doing help A. This can be only a default, because there may be many times when B has conflicting goals with respect to A. So, a second level to cooperativity is to indicate if the speaker does not share the conversational goals of the other participant. These principles are expressed by the following axiom: 

Cooperativity: (a) I A ðfÞ4I B ðfÞ (b) I A ðfÞ ^ :I B ðfÞ4I B BA :I B ðfÞ

Cooperativity doesn’t tell us what an agent’s goals might be in dialogue, because it may not be possible to infer an agent’s goals from what he says. This is where particular linguistic axioms like QRG and RRG come in. 

Question Related Goals (QRG):   QAPða; bÞ4 I agentðaÞ BagentðaÞ b

This axiom states: if b is the answer to the question a, then normally the agent or speaker of a intends to be in a certain state in which b is true. This axiom applies whenever an agent asks a question. A similar axiom holds for requests (where a :! expresses a is requested).

178 

NICHOLAS ASHER

Request Related Goals (RRG): a : ! 4 I AgentðaÞ a

SARGs for assertions are more difficult to capture. We’ll assume that knowledge relevant to connecting the content of assertions, which we’ll assume here to be sincere, to their conventionally associated SARGs can be accessed by the linguistic system. Finally, we’ll assume that if we compute a SARG via Cooperativity or RRG or QRG, then if the agent’s speech act has both a presuppositional and an assertional component, the SARG computed applies to both. One final matter is that in SDRT questions can elaborate on other questions or requests. We see this in (8b) already where Phillippe’s question is intended to help elaborate a plan already implicit in Isabelle’s opening (8a). We call the rhetorical relation that (8b) stands in to (8a) Question Elaboration or Q-Elab (see Asher and Lascarides, 1998c; Asher, 1999). If we have a Q-elab, then the SARGs of the second question include the SARGs of the first. Formally, we express this as: 

SARG additivity: ðQ-elabða; bÞ ^ SARGða; dÞÞ ! SARGðb; dÞ

This ensures that Q-elab SARGs are carried along as discourse participants try to answer the original question by asking other questions. We’ll group the SARGs that are accumulated through nested Q-elabs within a cluster.

Conversational Goals and De Re Attitudes We have most of the parts in place for our presuppositional account of contextual anchoring and its attendant cognitive attitudes. We needed an account of conversational goals, and we have just seen ways of getting SARGs from various conversational moves. We determined earlier from looking at our examples that these conversational goals were essential in determining the de re attitudes that are part of contextual anchoring. Further, I argued that these de re attitudes were really grounded not in an attitude towards a proposition but in an ability. I’ll try to be more precise about what this ability consists in now using the devices available to a theory like SDRT. That means giving some analysis of this practical capacity in terms of a broadly DR-theoretic account of belief.

DEIXIS, BINDING

AND

PRESUPPOSITION

179

Let’s return first to the turn (8e-f). Isabelle first corrects her previous turn and then tries to answer Phillippe’s question in (8b). Isabelle uses a deictically used definite, ‘the end of the avenue Jean Jaure`s’ in her answer, that Isabelle further localizes with reference to ‘the big square’ (la grande place). What is the discourse relation between the presupposition of the definite and the discourse context? Presumably, the presupposition is to anchor the assertion. Interestingly, Isabelle goes on to elaborate on this location where she is, and the point of this elaboration in (8e.ii) is ostensibly to help establish the Anchoring relation between the presupposition of the definite and the discourse context. To appreciate the cognitive effects of Anchoring, let’s see what happens were Phillippe to accept the anchoring discourse move by Isabelle. Were Phillippe to accept the Anchoring relation, he would be able to determine which location Isabelle describes. We can infer that this would satisfy one of the SARGs of the utterance, as Phillippe’s knowing where Isabelle is the SARG derived from (8b) via QRG. By SARG additivity this remains a SARG through (8d).4 By Cooperativity Isabelle takes over this SARG, which she tries to satisfy with her utterance of (8e). Were (8e) to be accepted, she would have satisfied that SARG and perhaps also the associated SARG of getting her unlost. In (8f), however, Phillippe doesn’t accept the Anchoring relation, which is why he asks ‘‘Wait a minute, I don’t quite see where you are’’. So accepting an Anchoring relation between the presupposition introduced by a definite c and some element in the discourse context by an agent A requires a computable means of getting to the referent of c from the present here and now, the present nonlinguistic context of utterance, for some given purpose f. To this end, I define a path relation on discourse referents x, which is introduced by the presuppositional component of the definite’s DR-theoretic lexical entry and u1, . . . ,un relative to a SARG f and its associated cluster, Pf(A, x, u1, . . . ,un). This relation holds iff u1, . . . ,un are accessible in A’s belief state and some of the u1, . . . ,un are externally anchored to distinguishable objects in the present context (e.g. the here and now).  there is a collection of formulas T(u1, . . . ,un) characterizing correct beliefs of A concerning u1, . . . ,un such that A has a proof from T(u1, . . . ,un) that f. 

4 Actually, in SDRT-theoretic terms, (8d) attaches to (8b) via Question Elaboration or Q-elab, which automatically propagates the SARG of the first question forward, but I’ll gloss over the details of this part of the discourse structure here.

180

NICHOLAS ASHER

Thus Anchoring as a discourse relation between a presupposition introduced by a definite and some other element in the discourse context in the SDRS for an agent A entails that the Agent can satisfy a current SARG that he has. The connection to a particular de re attitude grounded in an ability comes about because in many, and perhaps in all cases, the SARG that needs to be satisfied specifies, or requires for its satisfaction, a de re attitude that is itself grounded in an ability. That is, satisfying a SARG may often involve a practical ability in addition to beliefs towards attitudes. Spelling out the entailment without specifying the SARG further seems difficult. On the other hand, in the case of knowledge which is what is at issue in the examples culled from the dialogues (8) and (10), we can be more specific. In such cases the path formula could plausibly involve a sequence of locations l1 . . . ln, such that T and l1 have the following features: T implies formulas of the form C(li, liþ1) for 1rirn, where C is the relation of Connectedness.  T determines distance and orientation information for each li and with respect to li–1 and liþ1 and  the initial location l1 is an accessible point in the present nonlinguistic context. 

The idea is that if the dialogue agent whose SDRS contains an Anchoring relation and the associated goal is knowing where someone is, then the agent should be in possession of information that will allow him to compute a path, a sequence of connected locations that will get him from his present surroundings to the location denoted by the definite. Or if the SARG is a slightly more complex type of knowing where, say, the goal is to know where something a is relative to some other location l, then the agent must be in possession of a path from the location of a to l. With this in mind, let’s once again go back to the exchange in (8e-f). Phillippe doesn’t accept the Anchoring relation. Why? Well, it’s manifestly because even though Phillippe presumably knows what the end of avenue Jean Jaure`s is, he doesn’t know where she is. In this case the demands of his SARG to know where she is, can only be satisfied if he can bind the definite to a spatially determinable object from the present context, i.e. have his belief structure satisfy a path condition between where Isabelle is and his current context, or perhaps where she needs to go. As his response makes evident, he cannot.

DEIXIS, BINDING

AND

PRESUPPOSITION

181

Other examples from our dialogues bear out the usefulness of thinking in terms of T as determining a path. In (10) speaker A uses definites that he can link to elements in his environment and intends to have B link to elements in his immediate context (the map). A has presumably already linked the discourse referents via a Path condition. B stops the flow of instructions when he cannot determine a Path relation. The path condition is really a constraint on a dialogue agent’s attitudes. Path binding is a type of internal anchor in DRT. But what is distinctive about it is its link to practical activities as defined by the discourse and by SARGs.

Contextual Anchoring as Discourse Function We have now seen how Anchoring as a discourse function has entailments concerning de re abilities. What remains to be done is to specify how we might infer Anchoring as a discourse relation and to specify formally the relationship between the discourse structure and the cognitive constraints. This will be done through a pair of axioms written in the SDRT format. Like other discourse relations, Anchoring can also specify the underspecified bridging relation in the presupposed information. Given our informal analysis, we might think that Anchoring should somehow specify the bridging relation to a path relation. Here as with inferential binding, there is a downward flow from the global discourse structure and its associated cognitive model to resolving certain underspecifications needed in the binding of presuppositions. In those examples of inferential binding, however, the bridging relation cannot be set to identity without violating type restrictions. Here the situation appears to be different. Consider the exchange in (8f-h). Isabelle tentatively accepts (8f). She uses a path condition to bind the railroad tracks to something in her immediate surroundings. The path sequence has length 1, and she has information about the direction and distance that makes it more likely that her current SARGs will be satisfied: the SARGs are that Phillippe know whether Isabelle has passed under the railroad tracks (inferred via QRG as a SARG for Phillippe and then as a SARG for Isabelle via Cooperativity), that Phillippe know where Isabelle is and that Isabelle find her way (inferred via QRG, Cooperativity, SARG additivity). But she’s not sure, so she tells Phillippe what the head of the Path sequence is in (8g).

182

NICHOLAS ASHER

Now how does the Path condition interact with the specification of the bridging relation? Given the instructions given earlier by Phillippe to Isabelle, it’s easy enough for Isabelle to set the bridging relation to identity. This would be sufficient to bind the presupposition via Background to the asserted constituent or to Background’s topic. But this won’t achieve Phillippe’s SARG, which is determined by his question – namely, this is the SARG of knowing whether Isabelle passed under the railroad tracks he told her about. Further, we can assume that Isabelle also has the SARG that Phillippe know whether she went under the railroad tracks or not. This follows from QRG and Cooperativity: QRG tells us that Phillippe has as a SARG that he know the answer to his question; Cooperativity transfers this SARG from Phillippe to Isabelle. In order to satisfy this common SARG, Isabelle has to do two things; she does indeed have to link the railroad tracks mentioned to those given in Phillippe’s instructions, and she has to bind the location of that bridge to some location in her journey or where she is now. And if this analysis is right, then we need both to have Anchoring determine a Path condition while also allowing in the relevant cases the bridging relation to be set to identity. This would result in the most coherent discourse structure according to MDC because it leads to a satisfaction of a given SARG and it is also the one mandated by If Possible Use Identity. So it appears that whenever setting the bridging relation to a Path relation would help achieve some recognizable SARG, we infer Anchoring as a discourse relation; and in turn Anchoring then determines the existence of a Path condition relation. But an inference to an Anchoring relation doesn’t clash with the principle of setting the bridging relation to identity If Possible Use Identity; rather it complements it. I have formalized axioms for anchoring using the underlying nonmonotonic logic of SDRT. We infer Anchoring by default whenever resolving B to a path relation would normally allow the agents involved to see to it that (formalized via the operator stit) their SARGs are realized. Below we use [B ¼ ?](b) to mean that Kb has the underspecified conditions B ¼ ?. 

Anchoring:   t; a; b ^ Sargðb; fÞ ^ ½B ¼ ?ðbÞ^   K b ½B! P f  ! stitðagentðbÞ; fÞ 4 Anchorða; bÞ

DEIXIS, BINDING 

AND

PRESUPPOSITION

183

Constraint on Anchoring: ðAnchor ða; bÞ ^ ½Bðx; yÞðbÞ ^ Sarg ða; fÞÞ !  

 v jPathf ða; x;~ vÞ K a ^ K b ^ BA Anchorða; bÞ ! ~

The constraint on anchoring says that an anchoring relation entails that its terms must be true propositions and further that if an agent believes Anchor(a, b), then the Path condition must be satisfied by agent A. To see how these axioms work, let’s go back to (8g) and its context once again. Isabelle first processes Phillippe’s question. She isolates out the presupposition of the definite la voie ferre´e in a constituent Ke.ii.p and the assertion Ke.ii. Via Cooperativity, we can infer that she takes on Phillippe’s SARG of knowing the answer to this question and she does her best to tell him. But in order to give him an answer, she has to be able to anchor the presupposition and thus satisfy a path condition linking the bridge to some element in her trajectory, which I assume can be reconstructed from her here and now, or is her here and now itself. So Isabelle’s SDRS looks something like this, if we ignore the processing of definites like Phillippe and you and the manner adverbial clause: (14)

In the above, u is the discourse referent for the railroad tracks introduced in Phillippe’s previous instructions and one that is presumably now cognitively accessible in Isabelle’s cognitive state. The bridging relation has been set to identity between x and v. u is some discourse referent in Isabelle’s cognitive state that is an internal anchor for the presupposed material. And it is in virtue of u that the Path condition is satisfied. But Isabelle is not sure whether the Path condition has been satisfied. So she attempts to clarify or elaborate on what the putative Path condition is that she has found for x. In SDRT we model this rhetorical function by attaching (8g) to pe.ii.p with the discourse relation

184

NICHOLAS ASHER

Elaboration (for details on this relation see Asher, 1993; Lascarides and Asher, 1993). This Elaboration also constitutes an indirect answer (Asher and Lascarides, 1998a) to the question in pe.ii, and it is precisely this Elaboration and indirect answer that Phillippe rejects in (8h). Let’s see how our approach fares with discourse initial deictic definites. Consider the initial turn (10a) where A mentions the tribal settlement. Here there is presumably no discourse referent already introduced in the discourse that could serve as a link, and so resolving the bridging relation to identity in this case is not possible. We’ll assume that B is able to anchor the presupposition generated by the the extinct volcano, but as his question demonstrates, he is not able presumably to determine a Path condition for the discourse referent x introduced by the tribal settlement. So presumably the question in (10b) is intended to help get an appropriate Path condition for x and once that question is answered B can anchor the presupposition to the asserted content of (10a). In this case since If Possible Use Identity because this axiom cannot be used, MDC resolves the Bridging relation to the Path condition, once the Anchor relation is established.

From Acknowledging Path Binding to Mutual Belief A final element in the analysis of anchoring is to account for the fact that when an Anchoring function has been acknowledged, the two participants in the conversation have the mutual belief that the Path relations link to the same location. This comes about after the interpreter accepts an Anchoring relation proffered by the speaker; this means that the Path condition is satisfied not only by the speaker but the interpreter as well. Because the Path relation must be satisfied by both the speaker’s and the hearer’s beliefs if Anchoring, we say that Anchoring is a kind of coordination. How do we acquire mutual belief in communication? Due to Fisher (1988) we know that if communication is synchronous, then mutual belief can be had. Suppose there is enough simultaneous exchange of information to have it qualify as synchronous. This is in fact encoded in our constraint on agreement: a signal of agreement to a previous contribution in which a discourse structure like Anchoring holds gets us to a mutual belief that the presupposed material is serving as an Anchor. Now consider any of the conversational turns where an Anchoring relation is proposed and then accepted (e.g. 10c-d). (10) c. A: It’s at the bottom. It’s to the left of the a e extinct volcano. d. B: Right. How far?

DEIXIS, BINDING

AND

PRESUPPOSITION

185

By sincerity we have that A attaches the presupposed material given by the extinct volcano with Anchoring to his turn. By signalling an agreement with Right B also adopts this discourse structure for A’s turn. We can now conclude given our assumptions that there is mutual belief in this discourse structure (and that if you will we have that discourse structure in the common ground). But now how do we get to that mutual belief that both path bindings link to the same object? How do we even represent this fact? We can relatively easily answer the latter question: among the beliefs of a dialogue agent A are also beliefs about other dialogue participants – let’s say for the moment just B. Given that there is a shared belief that both participants have a path binding (from Anchoring), A can internally anchor the last discourse referent u of B’s path condition as in Asher (1986) or more recent work of Kamp. Here I’m going to use the older notation and represent internal anchors as equalities within the embedded belief context. So we’ll represent this internal anchoring for A as an equality xA ¼ xB in A’s representation of B’s belief state and similarly for B. In effect this says that B’s beliefs about xB are also in effect a belief about A’s individual concept. We’ll suppose that Anchoring has been proposed and accepted as in (10c-d). So it’s mutually believed that each agent’s cognitive state satisfies the Path condition for the discourse referent introduced by the presupposition of the definite. This means:      

Assume A represents B’s cognitive state as having a formula f(xB) in it for the definite while his own has fu(xA). BA BB fðx B Þ ^ BA fðx A Þ. By competence we have BA ðBB fðx B Þ4BA fðx B ÞÞ. By distributivity of belief over W and DMP: BA BA fðx B Þ, which then in K45 yields BA fðx B Þ, and so by  K BA fðx B Þ ^ f0 ðx A Þ .

And since f entails a uniqueness clause, first order logic yields: BA x A ¼ x B Since this is derived from mutually believed information and mutual belief is closed under deduction, we get:  MBA;B x A ¼ x B  

That seems to me to suffice for internal Anchoring however it’s represented. Notice that postulating this equality in A’s belief state leads to no binding problems because A supposes through competence that there is an xB of which B has his beliefs.

186

NICHOLAS ASHER

Conclusions I’ve shown that a discourse-based, anaphoric theory of presupposition has an interesting story to tell about at least some deictic uses of definites. In many of these uses presuppositions are anaphorically bound to the discourse context via a particular discourse relation, Anchoring, whose semantics and conversational function is directly linked to the participants’ conversational goals. Anchoring entails a de re attitude, but it is one that is linked to an increased capacity for satisfying at least some conversational goals. Our investigation has confirmed the view that de re attitudes involve some sort of knowing how. We have seen how Anchoring, when accepted by all participants, leads to a mutual belief in coordinated reference – viz. that all the participants are referring to the same thing and can single it out at least insofar as that’s required for conversational purposes. SDRT gave us the framework within which to analyse the discourse function of these uses of definites, and the modest set of defaults that SDRT uses in developing a theory of conversational goals or SARGs was helpful in deducing SARGs for the Anchoring analysis. Further tasks: Presumably definites outside the context of spatial localization dialogues can also be Anchored. So one idea for further research is to see how to extend this analysis to other definites – deictically used pronouns and the like. Moreover, it seems that almost all words have presupposition-like associated information whose failure to be anchored (bound) lead to similar corrections as those we’ve studied here. Consider these metalinguistic bits of anchoring information in the examples below due to Ginzburg that are called into question by B’s responses. (15) a. b.

A: John kowtowed. B: Kowtowed?

(16) a. b.

A: Chris inebriated Pat. B: Inebriated?

References Asher, N. 1986. Belief in discourse representation theory. Journal of Philosophical Logic 15:127–189. Asher, N. 1993. Reference to Abstract Objects in Discourse. Kluwer Academic Publishers.

DEIXIS, BINDING

AND

PRESUPPOSITION

187

Asher, N. 1999. Discourse structure and the logic of conversation. In K. Turner, ed., The Semantics/Pragmatics Interface from Different Points of View, pp. 19–48. Amsterdam: Elsevier Publishers. Asher, N., J. Busquets, and A. Le Draoulec. 2001. Cooperativity in dialogue. In M. Bras and L. Vieu, eds., Semantics and Pragmatic Issues in Discourse and Dialogue, Current Research in the Semantics/Pragmatics Interface, pp. 217–246. Amsterdam: Elsevier Publishers. Asher, N. and A. Lascarides. 1998a. Bridging. Journal of Semantics 15.1:83–113. Asher, N. and A. Lascarides. 1998b. The semantics and pragmatics of presupposition. Journal of Semantics 15:239–299. Asher, N. and A. Lascarides. 1998c. Questions in dialogue. Linguistics and Philosophy 23:237–309. Asher, N. and A. Lascarides. 2003. Logics of Conversation. Cambridge University Press. Asher, N. and Morreau, M. 1991. Common sense entailment: A modal theory of nonmonotonic reasoning. Proceedings to the 12th International Joint Conference on Artificial Intelligence, Sydney, Australia. Barwise, J. and J. Perry. 1983. Situations and Attitudes. MIT Press. Boer, S. and W. Lycan. 1986. Knowing Who. MIT Press. Fisher, J. 1988. Common knowledge and synchronous communication. In Y. Moses, ed., Proceedings of the Conference on Theoretical Aspects of Reasoning about Knowledge. Geurts, B. 1995. Presupposing. Ph.D. dissertation, Universita ¨t Stuttgart, Stuttgart. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. thesis, University of Massachusetts, Amherst. Heim, I. 1983. On the projection problem of presuppositions. Proceedings of the West Coast Conference on Formal Linguistics 2:114–126. Jeshion, R. 2001. Acquaintanceless De Re belief. In J. K. Campbell, M. O’Rourke, and D. Shier, eds., Meaning and Truth: Essays in Philosophical Semantics. Seven Bridges Press. Kamp, H. 1990. Prolegomena to a structural theory of belief and other attitudes. In C. Anderson and J. Owens, eds., The Role of Content in Language, Logic and Mind, pp. 27–91. CSLI Publications. Kaplan, D. 1968. Quantifying in. Synthese 19:178–214. Kaplan, D. 1978. Dthat. In P. Cole, ed., Syntax and Semantics, vol. 9, pp. 221–253. New York: Academic Press. Kripke, S. A. 2009. Presupposition and anaphora: Remarks on the formulation of the projection problem. Linguistic Inquiry 40:367–386. Lascarides, A. and N. Asher. 1993. Temporal interpretation, discourse relations and common-sense entailment. Linguistics and Philosophy 16:437–493. van der Sandt, R. 1992. Presupposition projection as anaphora resolution. Journal of Semantics 19(4):333–377.

7

BECOME and its Presuppositions MANFRED BIERWISCH

Opening In spite of continuous debate over the appropriate characterization of various details, there is little doubt that the semantic component frequently represented as BECOME is a crucial element in the structure of natural language. Standard examples illustrating the pertinent phenomena are minimal pairs like those in (1) and (2), differing by the presence of BECOME in (b) where (a) just lacks it: (1) a. Eve is an actress. The crew slept. The shop is open. Some of the kids were sick. (2) a. Hans schlief. (Hans slept)

b. Eve becomes an actress. The crew fell asleep. The shop opens. Some of the kids got sick. b. Hans schlief ein. (Hans fell asleep) Eva schwieg. (Eva was silent) Eva verstummte. (Eva became silent) Max ist wach. (Max is awake) Max erwacht. (Max wakes up) Er ist Buddhist. (He is a Er wird Buddhist. (He Buddhist) becomes a Buddhist)

There are various diagnostics and criteria identifying the relevant properties, for example, with respect to the temporal structure. Thus usually the (a)-cases allow for durational adverbials, but not the (b)-cases. The present paper will focus on another, well-known aspect, Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

189

190

MANFRED BIERWISCH

viz. the presupposition introduced by the occurrence of BECOME, according to which, for example, Eva verstummte presupposes that she spoke before, while no such requirement comes with Eva was silent. In what follows, I will briefly recapitulate the standard analysis of BECOME, turning then to some of the peculiarities related to its presuppositional aspect. A number of more complex phenomena will then be considered on the basis of this discussion.

Standard Assumptions The most obvious function of BECOME is to characterize so-called inchoative verbs like die, get up, or the examples in (1b) and (2b), including the corresponding causative verbs like kill, denoting the causation of an event expressed by an inchoative verb.1 These changeof-state verbs (or 2 state verbs in the sense of Klein, 1994) denote the transition from a source state s to a target state su. The basic assumptions taking care of these conditions have been formulated, for example, in Dowty (1979) by means of the operator BECOME as indicated in (3), where f is the target state and I, J, K are time intervals, schematically indicated in (4): (3) [ BECOME f ] is true at I if and only if (i) there is an interval J containing the initial bound of I such that :f is true at J, and (ii) there is an interval K containing the final bound of I such that f is true at K. J

(4)

K |

]

[

|

I

There are two problems to be clarified with respect to this account. First, as it stands, I may extend over arbitrary parts of the source as well as the target state, such that The cat died could apply to a situation that includes arbitrary parts of the cat’s lifetime and/or arbitrary periods of the cat’s not being alive anymore. For this reason, 1 It might be noted that in many cases, English inchoative verbs and their corresponding causatives are homophonous (giving rise to the somewhat misleading term ‘‘ergative verb’’), differing, however by the presence of the causative factor and the pertinent argument – as in He opened the shop versus The shop opened. I will have to return to this issue below, noting for the time being, however, that the relevant distinctions and relations carry over from inchoatives to their causative counterparts.

BECOME

AND ITS

PRESUPPOSITIONS

191

Dowty contemplates the extension of (3) by a condition (iii) that there is no interval Iu that meets condition (i) and (ii), i.e. I must be the smallest interval that includes the bound of J and K. Dowty does not suggest to adopt condition (iii) though, because it leads to difficulties in view of the second problem to be clarified, which has to do with the interval Iv between the source and the target state. If one relies on strictly two-valued logic, no interval Iv between J and K is possible, as either f or :f must hold – there is no transition. Together with condition (iii), any change of state would be strictly momentary. The solution Dowty suggests with respect to these two problems is, first to discard condition (iii) in favor of some sort of Gricean maxim, which picks out a short, but non-empty interval appropriate under conditions of encyclopedic or common sense knowledge, and second to acknowledge intervening time-intervals with undecided (or non-two-valued) truth conditions.2 On this account, verbs like die or leave are correctly construed as specifying neither the duration of nor the properties holding at the interval between the initial state :f and the target state f, nor exactly when :f and f begin or end. As a matter of fact, what is fixed by an utterance like (5) is merely the transition (at some time before the utterance and at the same day) from Bill’s being at some contextually given location to his not being there, without any specification whether this took place by walking or driving, by bike or by plane, whether it was a rapid or slow event, etc.: (5) Bill left today. Using notational conventions of an event-based semantics with events e and states s as sorts of eventualities in the sense of Bach (1986), the Semantic Form of (5) could be represented as in (6), where T(e) and T(u) indicate the time of the event and the time of utterance, respectively, and v is a contextually specified place parameter: oT ðuÞ ^ ½TðeÞ  day ^ T ðuÞ  day (6) 9e ½½TðeÞ |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} jPastj jTodayj ^ e : ½become :½loc½bill at v |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} jBill leave vj

2 As Dowty shows, the choice of intervals can be made sufficiently precise by means of an interval-semantics, which I need not pursue here, as the details of time structure are not my primary concern. The undecided truth-value of the interval between :f and f need not concern us here either, as it is the status of I rather than Iv, that we have to deal with.

192

MANFRED BIERWISCH

In the sense of definition (3), the actual event e, instantiating the proposition |Bill leave v | could be characterized as in (7), where  indicates the temporal overlap mentioned before: (7) If e, s, and su are eventualities with T(e) ¼ I, T(s) ¼ J, T(su) ¼ K, where T(e)  T(s), T(e)  T(su), and T(s)oT(su), then Bill leave denotes an event e such that: e : ½s : loc : ½bill at v ^ s0 : :½loc½bill at v |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} source state

target state

This leads to the following general conclusions: A change-of-state or BECOME-event e need (and frequently does) not have any other characterization than that given by its bounds. In other words, the nature of the transition determined by BECOME is characterized exclusively by the boundaries it connects. Verbs like leave, get, arrive, die do in fact provide no other specification of the transition between two states. (ii) There is, moreover, no specification of the time interval Iv between the boundaries of T(s) and T(su). Iv might be empty, as in instantaneous transitions like the atom lost a particle, or it may have substantial extension as in the towel got dry where for a certain time span Iv the towel is not dry, but not as wet as before either: It is just not defined. (iii) The specification of the transition by means of its boundaries noted in (i) is in fact restricted to the target state, the source state being completely derivable as the negation of the target state.3 (Since the target state in (7) is a negated state itself, the source state must, of course, be a positive condition.) Hence the defining information in a BECOME-event is the specification of the target state. (i)

While these generalizations can be read off more or less directly from the characterization in (3) and the illustration in (7), the following observations bring in additional facts: (iv) The restrictions noted in (i) and (iii) must not be construed as excluding substantial specification of the transition – or even as denying the transition as the very core of the BECOME-event. There are in fact verbs that characteristically specify the transition in various ways. German verbs of dying like ertrinken (be drowned), ersticken 3 Apparent counterexamples like melt, where the source state is more specific than just the negation of the target state will be taken up below.

BECOME

AND ITS

PRESUPPOSITIONS

193

(choke), verhungern (starve), verdursten (die with thirst) illustrate the point, to which I will return from a different perspective. What is to be noted here is the fact that qualifications of this sort are not due to the change-of-state conditions expressed by BECOME. (v) It is furthermore well known that the transition identified by BECOME can be subject to temporal localization and delimitation, as in (8) and (9), but not – with two qualifications to be discussed immediately – to durational qualification, as shown in (10): (8)

a. b.

The shop opened at noon. Das Gescha¨ft ¨offnete um zwo¨lf.

(9)

a. b.

The cat died in/within two hours. Die Katze starb in/innerhalb von zwei Stunden.

(10) a. b.

*The cat died (for) two hours. *Die Katze ist zwei Stunden (lang) ertrunken.

(vi) The different status of source and target state noted in (ii) has further aspects, one of which is directly related to the qualifications just mentioned: Durative adverbials combining with change-ofstate verbs, as in (11) and (12), can be accommodated by shifting the specification they express from the event to the target state: (11) a. b.

The shop closed for two hours. Der Laden hat zwei Stunden geschlossen.

(12) a. b.

Eve came to London for some days. Eve kam (fu ¨r) einige Tage nach London.

Cases like (11) cannot be interpreted as denoting a process of two hours, nor, of course, a source state of two hours, but only as determining the duration of the resulting state. This emphasizes its prominence mentioned in (iii) from another perspective. Before I get to the main aspect of the asymmetry to be pursued here, I will mention the other qualification hinted at in (v), which takes care of durative adverbials modifying change-of-state verbs in cases like (13) and (14): (13) a. b.

For several hours, visitors entered the hall. Mehrere Stunden lang betraten Besucher das Haus.

(14) a. b.

Captives broke out of this prison for quite a while. Die ganze Zeit sind Ha¨ftlinge aus diesem Gefa¨ngnis ausgebrochen.

194

MANFRED BIERWISCH

Here the durative adverbial applies to a process that is made up by the iterated change, induced, among other things, by the plurality of the Theme. We will have to return to this problem as we proceed. For the time being, it should be noted that the durative adverbial does not specify the change of state per se, determined by BECOME, but either its result or a kind of process constituted by its repetition. To sum up the main points of the received analysis: Change-ofstate- or BECOME-events necessarily involve a source state s, a target state su, and a transition e, where s precedes su and is contained in the complement of su, and e overlaps with s and su. The substantial information of the change of state is given by the specification of the target state su.

Assertion, Presupposition, and Implication Turning now to the presuppositional aspect of BECOME, we notice a three-way distinction between e, s, and su in (7) – or equivalently I, J, and K in (4) – in terms of assertion (or more generally: condition), presupposition (or precondition)4, and implication (or consequence). Evidence for this claim comes from the usual criteria, to be reviewed briefly. First, the presuppositional character of the initial or source state can be seen from the fact that asserting, denying, or questioning the transition all require the source state to be met, as illustrated in (15a) to (15c), all of which presuppose that Bill was in his office before: (15) a. Bill left his office. b. Bill didn’t leave his office. c. Did Bill leave his office? The presupposition in question is supposed to hold also if the question (c) is negated by (b). It can only be suspended by an explicit denial, usually called ‘‘presupposition protest’’, as in (16a), 4 A terminological remark might be in place. The phenomena usually lumped together as presuppositions have been treated under various perspectives. A systematic account has been developed a.o. in Karttunen and Peters (1979), where ‘‘conventional implicature’’ is used instead of ‘‘presupposition’’. To the extent that this is more than just a terminological difference, I will be concerned here with presupposition in the sense more recently discussed, for example in Kamp (2001) and the references given there. In what follows, I will also adopt the notational conventions proposed in Kamp (2001) to represent presuppositions.

BECOME

AND ITS

PRESUPPOSITIONS

195

while (16b) is out, as it explicitly violates the conditions for a change of state: (16) a. b.

Bill didn’t leave his office, as he wasn’t in at all. # Bill left his office, but he wasn’t in at all.

Second, the resulting or target state is a (necessary) consequence, implied by the transition, as indicated by the contradiction in (17a). But it does not imply a transition, just as the presupposition does not require a change to take place. In other words, the target state may hold, even if no previous change takes place, as borne out by the acceptability of (17b): (17) a. b.

#

Bill left his office, but he was not out of it afterwards. Bill was not in his office, in fact he was out all day (hence he didn’t leave it).

Finally, the condition proper, the change from source to target, is satisfied if and only if the presupposition, the consequence, and the transition hold. In other words, if either the presupposition or the consequence is not met, no change is possible and the condition fails, as (16b) and (17a) show. The negation of the change, on the other hand, is true if the consequence is false, while the presupposition holds – according to default conditions. In other words, the negation of BECOME is equivalent to the assertion that the source state persists. Hence (18a) holds if and only if (18b) holds: (18) (a) (b)

Bill didn’t leave his office this morning. Bill remained in his office this morning.

If, on the other hand, by way of presupposition protest the source state is negated, the negation of the change can only be true if the target state holds without relevant change – the normal interpretation of (16a). In other words, the absence of a change requires either the source or the target state to persist, the former being the default case that can be switched to the second only by means of (explicit or implicit) presupposition protest. According to standard definition, a proposition f presupposes c, if and only if f implies c and :f also implies c. In order to account for the observations just discussed, we might provisionally modify the definition in the way indicated in (19): (19) f presupposes c ¼ def f is true only if c is true and :f is true only if c is true, except c is explicitly rejected.

196

MANFRED BIERWISCH

With this proviso, the asymmetry between the presupposed source state and the implied target state can be accommodated by the following modification of (3): (20) [ BECOME f ] is true at I if and only if (i) there is an interval K containing the final bound of I with f implied to be true at K, (ii) there is an interval J containing the initial bound of I with :f presupposed to hold at J. This characterization has obvious consequences for the information to be specified for lexical entries as well as other constructions involving BECOME: Once the target state is determined, the presupposed source state is automatically defined and need therefore not be specified as lexical information. In other words, while lexically induced presuppositions must somehow be specified as a characteristic property of, for example, factive verbs like grasp or regret (as opposed to nonfactive verbs like claim or assume)5 or particles like too, also6, and again7, the presupposition induced by the inchoativity of verbs like come, leave, enter, die, open, etc. are fully predictable and need not lexically be indicated. The same holds for causative verbs such as bring, kill, or transitive open specifying the causation of a change of state: In order to bring about the change in question, the source state to be changed must be met in the first place. For the same reason, resultative constructions like (21a) are automatically associated with a presupposed source state indicated by (21b), which is not a presupposition of its parts, as can be seen from (21c) and (21d): (21) a. b. 5

Max wiped the table clean (#although it was clean). The table was not clean.

The presuppositional character of factive verbs has first been observed by Kiparsky and Kiparsky (1970) and was subsequently discussed in a large number of different approaches. The point to be noted here is the fact that it is a matter of idiosyncratic, lexical information whether a factive presupposition is associated with a given verb or not, while no such idiosyncracy holds for the initial state of an inchoative verb. 6 See Reis and Rosengren (1997) for a recent survey of the pertinent facts. Even though the paper is primarily concerned with the problem of how the particle determines its focus, it necessarily deals implicitly with the lexical basis of the presupposition the particle induces. 7 There is an extensive discussion about this particle, summarized, for example in Dowty (1979); Kamp and RoXdeutscher (1994); von Stechow (1996), which focuses, however, on the scope assigned to the particle, from which the presupposition it induces is then derived. I will take up one aspect of this particle in the Section on The Interaction of BECOME and AGAIN.

BECOME

AND ITS

PRESUPPOSITIONS

197

c. Max wiped the table (although it was clean). d. The table was clean (#and it was not clean). If no additional conditions interfere, wipe is just an activity without particular presuppositions or specified result. And the adjective clean does not make any presupposition either.8 The causative verb clean on the other hand, appearing, for example, in Max cleaned the table does, of course, automatically induce the presupposition indicated in (21b), due to the component BECOME it contains. The way in which causative verbs like clean and resultative constructions like wipe clean acquire the semantic components CAUSE and BECOME with their interpretive consequences is systematically discussed in Wunderlich (2000) and Bierwisch (2002). In order to spell out the presuppositional aspect of BECOME more explicitly, including related elements with their consequences, some remarks on notational conventions seem to be in order.

Lexical Representation of Presuppositions In Kamp (2001), a systematic distinction is made between representation, computation, and justification of presuppositions. To begin with, an explicit account of their representation is needed, which then allows to clarify the computation of presuppositions, i.e. the way in which they arise on the basis of lexical information and compositional semantics. The justification of a presupposition is then concerned with the problem of how it is either satisfied within the given context, or how the context is to be modified in order to accommodate the presuppositional requirement. According to a revealing observation developed in van der Sandt (1992) and adopted in Kamp (2001), justification of a presupposition by previous or inferred conditions is in essential respects identical to the relation of anaphoric elements to their antecedent. In this section, I will sketch minimal assumptions about the representation of lexical presuppositions, which in turn requires a general format for the representation of lexical information. I will then turn to the 8 For adjectives, the role of presuppositions is in general less obvious and more controversial than for verbs. Thus whether, for example, a type crossing like red idea exemplifies a presupposition violation of the adjective, which requires reference to a physical object, might be a matter of dispute. In any case, presuppositions of adjectives are not excluded in principle, and elements like silent are naturally construed as presupposing animacy. On this background, it isn’t just a triviality that clean has no presupposition, as it applies to physical objects, processes, or even abstract entities like clean measures.

198

MANFRED BIERWISCH

computation of presuppositions as far as they depend on BECOME and related elements. To begin with, we have to recognize that a lexical entry E must provide at least three types of information: First an array of phonetic features determining the contribution of E to the Phonetic Form (PF) of the expressions it occurs in; second a complex structure representing the Semantic Form (SF) of E determining the contribution of E to the semantic interpretation (or meaning) of the expressions it occurs in; and third a specification of the Grammatical Form (GF) defining the combinatorial requirements that guarantee the correspondence between PF and SF for complex expressions build up from E. More specifically, GF has two components sometimes called Categorization Cat and Subcategorization or Argument Structure (AS), where Cat is a structured set of morpho-syntactic features defining the grammatical properties of E as a head, while AS specifies the requirements E imposes on its co-constituents. Each of these components and their interaction can be elaborated in various ways. For obvious reasons, I have little to say here about PF. As to Cat, I will rely on familiar conventions about morphological and lexical categorization by means of binary features. The representation of AS and SF is based on standard assumptions about type systems with lambda abstraction. More specifically, SF is a functor-argument-structure made up from constants and variables, where variables are bound, substituted, or interpreted according to conditions of formal or semantic and situational context, while constants are subject to (more or less invariant) conceptual interpretation.9 Of particular interest in the present context is the binding of variables by lambda-abstractors that constitute Argument Positions in AS, relating the variables in question to the conditions imposed by the complements a lexical head combines with. To that effect, argument positions are usually associated with morpho-syntactic features defining requirements on the categorization of the constituent that is to saturate the position.10 Thus AS participates in the morphosyntactic as well as the semantic information of E and constitutes in 9

To make this interpretation explicit is an extremely complex and ramified task that has to cope with rather different types of problems that can only occasionally be touched here. One type of interpretation, however, is indicated in (20) for the functor BECOME. 10 The most familiar conditions in this respect are Case features distinguishing subject, direct, and oblique object. Other conditions identify finite and infinite complement clauses, or various types of predicates of copulative constructions. There are important generalizations controlling this aspect of AS, which I cannot pursue here. For a more systematic exposition of these matters, which are not in the focus of the present discussion, see for example, Bierwisch (1997); Wunderlich (2000), and references given there.

BECOME

AND ITS

PRESUPPOSITIONS

199

a way the interface between its syntactic and semantic structure. A simplified example illustrating the ingredients just mentioned would be the entry (22) for the adjective open: (22)

The one-place predicate OPEN is to be understood as a short-hand for a fairly complex condition representing free access or passage to or by the argument represented by the variable x. This variable, bound by the entry’s only argument position lx, is to be specified either by the subject of a copulative construction with open as its predicate (as in the window was open) or by the head noun of an attributive construction with open as its modifier (as in an open window). We get closer to issues specifically related to our present interests, if we consider entries that account for open as an inchoative and as a causative verb, related of course to the adjective just discussed. The pertinent entries are given in (23a) and (23b), respectively: (23) a. /open/ [ þ V,  N] lx le [e: [BECOME [OPEN X]]] b. /open/ [ þ V,  N] lx ly le [e: [[ACT y] [CAUSE [BECOME [OPEN X]]]]]

A number of comments are in place here. First, within SF a variable ‘‘e’’ for eventualities is assumed, which is – along the lines familiar from DRT-representations – associated by the colon ‘‘:’’ with a proposition it instantiates. Formally, then, the colon is a relational operator that turns a proposition and an individual into a proposition.11 Second, the event variable just mentioned is bound by the position ‘‘le’’ in AS, which differs from other argument positions in systematic ways, essentially due to the categorization of verbs by the feature combination [ þ V,  N] in Cat. In other words, reference to 11

In Bierwisch (1997) and elsewhere I have used an operator INST indicating the instantiation of a proposition by an eventuality in much the same way. An alternative proposal with ultimately the same effect is made in Wunderlich (1997), where CAUSE, BECOME (and various other semantic elements) are treated as propositional operators that are relativized to eventualities, incorporating, in a way, the effect of the colon represented here as a separate operator. The term ‘‘eventuality’’ is used in the sense proposed by Bach (1986), covering both events and states. Notationally, I will adopt the widespread practice using ‘‘e’’ as a variable for eventualities in general, and ‘‘s’’ as a variable to mark states, whenever the distinction is relevant. It should be noted however, that the logical type of eventualities covers events as well as states and processes as sortal distinctions within the same type.

200

MANFRED BIERWISCH

eventualities, syntactically anchored in a particular argument position, is a category-specific property of verbs. The role of this event position is directly related to the clause-specific functional categories T(ense) and C(omplementizer) in ways that must be left aside for the moment.12 Third, the ordering of lambda-operators in AS, i.e. the hierarchy of syntactic argument positions, is – at least by default – a direct consequence of the position the pertinent variable occupies in SF: earlier, that is lower ranking, argument positions are bound to more deeply embedded variables in SF. For details see, for example, Bierwisch (1997). Finally, given entries like (22) and (23), the well-known relatedness between the adjective and the two verbs open, and more generally the relation between adjectives and corresponding causative and inchoative verbs can be captured in at least two ways. One is to collapse the related entries, using their similarities to reduce the lexical information. I will illustrate this option in two steps. The first step is to integrate the inchoative and the causative verb into one entry that represents two options, depending on the choice of the parts enclosed in heavy parentheses: (24) /open/ [ þ V,  N] lx (bly) le [e: [(b [ACT y] [CAUSE) [BECOME [OPEN X]]]]]

The subscripts of the parentheses indicate that the enclosed parts are either both present or absent. The second step uses the same conventions to integrate (24) with the simple adjective (22), which comes out as the residue if the content of (a . . . ) is deleted:13 (25) /open/ [ þ V,  aN]

lx (a (bly) le [e: [(b[ACT y] [CAUSE)[BECOME) [OPEN X]]]]]

The alternative way to capture the relationship between (22) and (23) is to stick to the simple lexical entry (22) for the adjective, 12 Different, but converging proposals by Higginbotham (1985); Bierwisch and Lang (1989); Kamp and Reyle (1993); Kratzer (1994); Wunderlich (1997); Kamp (2001), to mention just a few, have integrated the verbal event reference with various aspects of argument structure, tense, and adverbial modification. 13 In (25), the subscript a of the parentheses is construed as a variable over þ and  , where ‘‘ þ ’’ indicates the presence and ‘‘  ’’ the absence of the content. These values are then also used to relate the content of SF and AS to the features in Cat, such that þ N shows up, if a is  , hence the content of the parentheses is empty, and  N otherwise. These notational devices have been proposed in Chomsky and Halle (1968), in order to express generalizations in phonological rules. It has since been argued that this use of value-variables allows all sorts of spurious or even absurd generalizations, such that it has been discarded in phonology. Similar objections seem to hold with respect to lexical information. Hence the proposal formulated in (25) is presumably to be discarded in favor for (26). See Bierwisch (1997) for some discussion of the problem.

BECOME

AND ITS

PRESUPPOSITIONS

201

generating the pertinent inchoative and causative verbs by two more general templates (26a) and (26b), which yield (23a) and (23b), respectively, if applied to (22), such that the variable ‘‘P’’ is substituted by OPEN: (26) a. /+/ [ þ V,  N] lP lx le [e: [BECOME [P X]]] b. /+/ [ þ V,  N] lP lx ly le [e: [[ACT y] [CAUSE [BECOME [P X]]]]]

We will notice below that (26a) is a phonologically empty variant of the inchoative copula become, and (26b) is a kind of empty variant of make or cause to be. One of the problems created by (26) depends on the lexical constraints, idiosyncratically delimiting the set of adjectives to which these templates apply: while both (26a) and (26b) apply, for example, to open, dry, clear, fit, only causativization by (26b) holds for clean or wet, and neither applies to long, short, red and many other adjectives.14 It might be added, by the way, that (26b) is very similar to a template that incorporates adjectives not into causative verbs, but rather into resultative constructions like he pushed the door open, where the matrix verb specifies the activity indicated by ACT in (26b).15 With this sketch of lexical information in mind, I will now turn to the question how the representation of presuppositions will enter the picture. The general orientation I will follow in this respect is given by the proposals developed in Kamp (2001). According to these proposals, which I will adopt as far as lexical information is concerned, the semantic representation of a linguistic expression consists of at least two parts, the descriptive part D and the presupposition P, for which I will use the following notation: (27) [{P} D] where D consists of the Semantic Form (SF) as discussed so far, and P is a (possibly empty) set of conditions based on the same elements as SF. 14 Causative formation as characterized by (26b) does in fact apply to long, short, or strong, but not by means of phonetically vacuous templates: lengthen, shorten, strengthen introduce the suffix –en, in addition to previous nominalization in the case of long and strong. Proposals concerning the intricate interaction of derivational suffixes and the lexical items they select are discussed in Bierwisch (1990). 15 More technically, the resultative template proposed by Wunderlich (2000) would look as follows:

(i) /+/ [ þ V,  N] lP lx lV ly le [e: [[V y eu] [CAUSE [BECOME [P X]]]]] Without going into the details of distinguishing strong and weak resultative constructions discussed in Wunderlich (2000), I will merely point out that BECOME enters this construction with all relevant consequences in rather similar ways as it shows up in causatives derived by word formation.

202

MANFRED BIERWISCH

We have to add further comments, as we proceed.16 The two parts of a lexical entry are then subject to compositional integration into the representation of complex expressions. For D the compositional integration is essentially a consequence of saturating the positions in AS by way of functional application (followed by lambda conversion) in one of two ways: If it is the AS of a syntactic head, its positions are saturated by the appropriate complements; if it is the AS of a modifier or adjunct, its designated argument position is unified with a position in the AS of the head it modifies. As to the presupposition P, the conditions it contains are integrated into the presuppositional part P 0 of the constituent it enters by syntactic composition. This integration involves either local justification, if a condition in P is met by the compositionally derived SF, or it is unified with the conditions coming from its sister constituents.17 Let me illustrate the consequences of (27) by a simple example. The inchoative verb melt, for which we might assume an entry like (28) as a first approximation, is a somewhat special case in several respects. (28) /melt/

[ þ V,  N ] lx le [e: [BECOME [LIQUID X]]]

First, the change of state denoted by melt may (and usually does) involve a fairly extended period between the presupposed and the target state. Moreover, this intermediate period is not just undefined with respect to its properties, as in, for example, die, open, or disappear. 16 In Kamp (2001) and related work, the descriptive part as well as the conditions in the presupposition are represented in DRS-format. Although the conversion between the SF- and DRS-format is in practically all respects straightforward, I will stick here to the SF-representation as summarized in Bierwisch (1997) for two interrelated reasons: First, while a DRS is constituted by an essentially unstructured set of propositions, SF combines constants and variables according to their types into an integrated hierarchy. For reasons discussed in Bierwisch (1997), this hierarchy corresponds to a strictly binary branching structure. Second, the hierarchy of positions in the Argument Structure (AS), which relates variables in SF to syntactic constituents specifying their value, depends crucially on just the SF-structure in question. There is no problem to impose this type of structure on the set of propositions of a DRS, but then we arrive at the SF organization adopted here. 17 This is an extremely provisional description of what happens to D and P in the construction of complex expressions. Some of the additional requirements to be observed will be filled in as we proceed. For a slightly more detailed exposition of the saturation of positions in AS see Bierwisch (1997) and references given there. A by far more scrutinized account of the compositional construction of presuppositions is developed in Kamp (2001), where also additional distinctions within P are made, to which I have to return.

BECOME

AND ITS

PRESUPPOSITIONS

203

It rather is a clear state of liquidity of part of the melting object. Nevertheless, melt is clearly a change of state, as borne out by (29): (29) My ice cream melted {(within) in a few minutes/?for half an hour} It is not essential for the moment whether melt should therefore be classified as an achievement or an accomplishment in the standard sense. Second, and that is the crucial point at the moment, melt does not simply denote the transition from non-liquid to liquid, as required from a standard inchoative verb, but more specifically from solid to liquid. Let us suppose that this more specific condition is to be represented as in (30), giving rise to the lexical presupposition as indicated: (30) /melt/

[ þ V,  N]

lx le ½fs : ½solid xg ½e : ½become ½liquid |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} P

D

In other words, the event e of x’s becoming liquid presupposes a state s of x’s being solid. Notice that ‘‘x’’ is bound by the argument position ‘‘lx’’ in both the presupposition and the descriptive part. This is clearly not sufficient, though. In order to specify the relevant presupposition, the state s must meet the condition discussed in (7) above, i.e. it must overlap the event e (and precede the target state su). Suppose we consider the temporal relation of e and su as part of the conditions associated with the constant BECOME that might be expressed as in (31): (31) (e: [BECOME p] is true at t) implies (su: [p] is true at some tu with tu  t, tu  tv, totv). In other words, the target state su overlaps with e, but also with some tv after e, hence su must include the final part of e. (31) at the same time guarantees that the truth of the target state p is implied by the change. With this general condition on BECOME in mind, we might complete the entry for melt as in (32), where, ‘s  e’ is short for ‘T(s)  T(e)’: (32) /melt/ [ þ V,  N]

lx le [{s

 e ^ s: [SOLID x]} [e: [BECOME [LIQUID x]]]]

As the state characterized by [LIQUID x] overlaps with the end of e, [SOLID x] can only precede it, if it is still to overlap e. Notice that both e and x in the presuppositional part are bound by the respective argument positions. Hence whatever happens to these variables in SF will also apply to their occurrence in the presuppositional part. There

204

MANFRED BIERWISCH

are two further problems to be noted. First, the properties identified by SOLID and LIQUID must be construed as relying on what might be called common sense physics, specifying general knowledge about aggregate states of familiar substances, including the possible transitions. Hence the fact that heat is involved in liquefying solid substances might be considered as supplied by encyclopedic knowledge activated by the transition in question. Otherwise, an additional condition which specifies the increase of temperature of x must be added to the target state. Second, as it stands, the transition from solid to liquid applies to the object x as a whole – contrary to fact, as already noted. It might in fact be part of the common sense knowledge just mentioned that a change of aggregate state applies to the substance, rather than shape or function of objects, turning them automatically into mass objects, such that the change gets partitioned according to the continuous or mass character of the object relative to the change of aggregate state.18 Instead of adding further complications to (32), the main purpose of which was simply to illustrate the representation of lexical presuppositions, I will leave it at that, returning to the issue under a different perspective below. One final point is to be made here. The presupposition in (32) introduces the variable ‘‘s’’, which does not occur in SF, and it does not constitute a discourse referent to be taken up elsewhere. Therefore no particular care must be taken of this implicit variable. The more general case, however, needs the representation and binding of discourse referents. Kamp (2001) elaborates the standard assumptions about discourse referents in important ways, providing in particular an additional component BC of Binding Conditions providing the different options and restrictions according to which the referential variables of nominal and verbal constituents exert their referential capacity. I will try to avoid these complexities as far as possible, focusing on the problem of computing lexical presuppositions. 18 There are at least two options to be considered here, if the issue is not shifted from semantics to extralinguistic common sense knowledge. One is to represent the ‘grinding’ of the object and the emerging sub-events explicitly in the lexical information, such that (32) would have to be replaced by something like (i):

(i) /melt/ lx le [{’xi (ei [xiDx 4 eiDe 4 si  ei 4 si: [SOLID xi]]} [ei: [BECOME [LIQUID xi]]]] The other option would be to rely on a general template which as a kind of ‘‘universal grinder’’, accounts for the fact that in (ii) the ice cube is treated as a substance rather than a cube thereby turning (32) into something like (i). (ii) The ice cube melted slowly.

BECOME

AND ITS

PRESUPPOSITIONS

205

Computing the Presupposition of BECOME As already noted, melt differs from inchoative verbs in general, as its presupposed state is more specific than just the negation of the target state. In the general case, however, the presupposed source state is fully predictable. The overall strategy of restricting lexical entries to the idiosyncratic information would therefore require the computation of the predictable presuppositions by general principles or at least language-specific templates that apply across the board. Given the discussion of (32) and the general observations about BECOME, it is obvious what for an inchoative verb like open the result of this computation should look like: The entry (23a) – repeated here as (33a) – must be expanded into (33b): (33) a. /open/ [þ V, N] lx le [e: [BECOME [OPEN x]]] b. /open/ [þ V, N] lx le [{s  e 4 s: :[OPEN x]} [e: [BECOME [OPEN x]]]]

It is just a notational task to formulate an operation that would have this effect: (34) [e: [BECOME [p]]] .[{s

 e 4 s: :[p]} [e: [BECOME [p]]]]

Instead of postulating an operation that expands the left into the right-hand side of (34), which would then require to somehow block iteration, it seems more appropriate to consider (34) as a constraint on well-formed lexical entries, which supplies the presupposition if an entry that meets the left-hand side enters the derivation of an actual expression. Taken together, this constraint and the condition (31) would then provide the full interpretation of BECOME. With this proviso, standard inchoative verbs like open, die, disappear, etc. can be lexically represented in the simple form without presupposition, as in (33a) or (35): (35)

/disappear/

[ þ V,  N]

(ly) lx le [e: [BECOME [:[[PRESENT AT y]x]]]]

Two remarks are to be added here. First, the target state is characterized by a negated condition. Hence the constraint (34) will supply – as its negation – a positive condition. Second, the condition of the target state specifies the absence of the theme x from some location y, which may be left completely implicit, however. This possibility is indicated by the optional argument position binding the variable in question. If the location is made explicit, though, it appears as a directional PP, as in (36). This is provisionally indicated by

206

MANFRED BIERWISCH

associating the optional position ‘‘ly’’ with the morpho-syntactic feature [ þ Dir].19 (36) a. b.

The target will disappear (from the screen). The snow disappeared (from most places).

Obviously, the assumptions discussed so far must hold also with respect to what might be called the inchoative copula, i.e. the verb that merely adds the change-of-state operator to the predicative of the copula. Examples like (37) illustrate the point: (37) a. His brother was famous. The book will be a scandal. She is an excellent teacher. Mary was sick. They were there in time.

b. His brother became famous. The book will become a scandal. She becomes an excellent teacher. Mary got sick. They got there in time.

There are non-trivial conditions distinguishing the use of become and get, which I must ignore here, treating become as the canonical case, which essentially provides BECOME with the accessories of a fully fledged lexical entry:20 (38) /become/

[ þ V,  N]

lP lx le [e: [BECOME [P x]]] [ þ N]

The feature [ þ N] associated with the argument position ‘‘lP’’ is a provisional way to indicate the selectional conditions for the 19 This is only a provisional hint. The actual problems to be faced here syntactically as well as semantically are rather complex and must be left aside here. Notice however, that the value of ‘‘y’’, whether explicitly or implicitly specified, figures in both the target state and the presupposition. 20 This is not the place to deal with the remarkable complexity of the various readings of get, only one of which is a close relative of become. I would nevertheless like to hint at the obvious similarity that the relation between causative and inchoative get shares with that between transitive and intransitive open, melt, dry, etc. In other words, the relation between constructions like (i) and (ii) should be treated parallel to the two verbs open represented in (24). Hence, either by means of the Causative template or as a complex lexical entry, we would have two readings of get as indicated in (iii):

(i) (ii) (iii)

He got things right (B He opened the shop) Things got right (B The shop opened) /get/ [ þ V,  N] lP lx (bly) le [e: [(b[ACT y] [CAUSE)[BECOME [P x]]]]]

for discussion of related properties of get see, for example Haegeman (1985).

BECOME

AND ITS

PRESUPPOSITIONS

207

predicative that combines with become. I assume here, that lP must be saturated by a nominal constituent, i.e. an adjective as in become pregnant, or a noun as in become president. For the sake of illustration, suppose that (39) is a simplified entry for the adjective famous, which (38), completed according to the constraint (34), combines with to form the VP become famous, represented in (40):21 (39) /famous/ [ þ N, þ V] (40) /become famous/ [ þ V,  N]

lz [FAMOUS z] lx le [{ s  e 4 s: :[ FAMOUS x]} [e: [BECOME [FAMOUS x]]]]

I would next like to show how these assumptions can account for what might be considered as the dual counterpart of become, preferably realized by the verb remain. The correspondence between the inchoative copula and its counterpart may be illustrated by (41a) and (41b), which have almost identical meanings, provided ill is construed as equivalent to not healthy:22 (41) a. The kids remained ill. b. The kids didn’t become healthy. c. The kids were ill. Notice first, that if (41a) is true, the simple copula clause (41c) must also be true, while the reverse does not hold: The truth of (41c) does not imply (41a), which asserts the continuation of the state in question. In this respect, the state expressed by the copula clause behaves like the target state of an inchoative construction. The next point to be noted is that (41a) does not simply assert the continuation of sickness, as would, for example, The kids were still sick, it more specifically denies the change leading from the source to a different target state. This is what makes it equivalent to (41b), where the denial of the change to the complement state is expressed

21

More explicitly, (40) derives in three steps. Supplying first (38) with the presupposition by means of (34) gives (i), which then combines with (39), yielding (ii), where by lambda conversion ‘‘lP’’ is deleted and ‘‘P’’ replaced by ‘‘lz [FAMOUS z]’’. Finally, lambda conversion turns (ii) into (40): (i) lP lx le [{s  e 4 s::[P x]} [e: [BECOME [P x]]]] (ii) lx le [{s  e 4 s: :[lz [FAMOUS z] x]} [e: [BECOME [lz [FAMOUS z] x]]]] 22 The correspondence between become and remain is less clear-cut than, for example the relation between werden and bleiben in German, which lacks competitors like get and keep. The main points, however, seem to be as clear as they are in German and can reasonably be discussed with respect to remain as the paradigm case.

208

MANFRED BIERWISCH

explicitly.23 Finally, both (41a) and (41b) share the same presupposition expressed by the copula clause (41c), provided the relevant temporal ordering is assumed. More generally, remain presupposes the predicated state p, and it denies the transition to :p. This comes out automatically, if we set up the lexical entry (42), whose SF differs from that of become just by denying the transition: (42) /remain/

[ þ V,  N]

lP lx le [e::[BECOME :[P x]]]

A couple of remarks are to be added here. First, (42) is incomplete in several respects. It doesn’t take care, for example, of cases like nothing of the book remained, where P is syntactically empty, but must semantically be interpreted as a kind of existential predication. This in turn has to do with the status of the argument position lP and the morpho-syntactic constraints to be associated with it. They are similar, but obviously not identical to those associated with the corresponding position of become. I will leave aside these issues for the time being, turning to a second remark: We clearly must expect the constraint (34) to apply in order to get (43), specifying the required presupposition, viz. the negation of the target state: (43)

/remain/

[ þ V,  N]

lP lx le [{s

 e4s:[Px]} [e: :[BECOME:[P x]]]]

In order to derive (43) from (42) by means of (34), we must however, deal with the fact that BECOME is in the scope of negation in (42), but not in (34) – or in the entry (38) for become, for that matter. Intuitively, one would like to claim that (34), as it takes care of the presupposition to be triggered, by the very nature of presuppositions holds for both the positive and the negative condition the presupposition is associated with. Hence (34) would apply to [BECOME p], irrespective of the negation prefixed in (42). Although this claim seems to me right in spirit, it is insufficient for two reasons. First, the consideration just mentioned applies to the external negation, but not to the negation within the scope of BECOME, which clearly must be respected, because otherwise we would derive the wrong presupposition. Hence the fact that the external (but not the internal) negation is to be ignored must somehow be indicated in (34). Second, this 23 It might be noted that literally speaking the negation of the change implies the negation of the target state only as a consequence of the presupposed source state whose change is denied. Hence if the presupposition is rejected – as in The kids didn’t become healthy, they simply were not sick at all – the result of a negated change would not be the negated target state.

BECOME

AND ITS

PRESUPPOSITIONS

209

requirement seems to be part of a more general issue to be taken up with respect to causative verbs. Finally, it is the occurrence of the negation ‘‘inside’’ and ‘‘outside’’ of BECOME in (42), that makes remain the dual counterpart of become.24 As a consequence of this, the presupposed state, being the negation of a negative state, comes out as a positive condition. For the sake of illustration, (44) represents the verb phrase remain ill, assuming that the SF of ill may be abbreviated as [:[ HEALTHY x ]], which leaves the target state [ HEALTHY x ] by canceling the double negation out. (44) /remain ill/

[ þ V,  N]

lx le [{ s  e 4 s: [:[HEALTHY x]]} [e: :[BECOME [HEALTHY x]]]]

It is easy to figure out that on the basis of these assumptions the SF of not become healthy must come out practically identical to that in (44). In much the same way, the equivalence between predications with remain and negated inchoatives as illustrated in (45) and (46) can be accounted for: (45) a. The shop remains closed. b. He remained in Paris for weeks. c. The dog remained alive. (46) a. The shop doesn’t open. b. He didn’t leave Paris for weeks. c. The dog didn’t die. 24

We might deal with this duality as a relation between constants of SF rather than lexical entries, by setting up a postulate like (i), which then would naturally lead us to replace (42) by the entry (ii): (i) REMAIN p 2 : [BECOME :[p]] (ii) /remain/ [ þ V,  N] lP lx le [e: [REMAIN [P x]]] I will not adopt this alternative analysis for two reasons. First, we need the postulate (i) only to introduce an additional primitive element of SF, all the properties of which are captured by the combination :BECOME :. Furthermore, we would not only have to set up (iii) – in addition to (34) – to account for the presupposition induced by REMAIN, we would also need (iv) to account for its implication, as discussed before, as (34) and (31) would not apply to REMAIN: (iii) [e: [REMAIN [p]] . [{s  e 4 s: [p]} [e: [REMAIN [p]]]]] (iv) (e: [REMAIN p] is true at t) implies (su: [p] is true at some tu with tu tuotvu).

 t, tu  tv,

210

MANFRED BIERWISCH

Examples like these bring up an intriguing problem, once we switch the negation: (47) a. The shop doesn’t remain closed tonight. b. The shop opens tonight. c. The shop is open tonight. (48) a The kids didn’t remain ill. b. The kids became healthy. c. The kids weren’t ill. Most of the comments apply as before. Consider, for example, (48), which corresponds to (41), except that positive and negative predications are exchanged, preserving the presupposition – the kids being ill – but exchanging the target states, as expected. The problem to be noted concerns the state- and event-properties: While negated events like (41b) behave like a state in crucial respects, as shown in (49), a negated non-event, if that is what remain expresses, does not become an event, as (50) indicates.25 (49) a. The kids didn’t become healthy for two months. b. The shop doesn’t open for two hours. #

(50) a. b.

#

The kids didn’t remain ill in two months. The shop doesn’t remain closed in two hours.

Clearly, (50) cannot mean that the kids became healthy in two months, or that the shop opens in two hours. What this means is that the clausal negation and the negation within the SF of remain do not cancel out each other in cases like (50). A preliminary consequence to be drawn from this observation is to distinguish between (a) [ :[e: p]] 25

Things are fairly complicated here, because judgments may rely on the possibility mentioned above, to let durational adverbials qualify the target state, as in he came in for a few minutes, such that the acceptability of sentences like (49) can be construed as qualifying either the negated event or the negated result, which amounts in a way to the same. Somewhat different interpretive options blur the judgment with respect to temporal limitations, as in (i) and (ii), which might be construed in analogy to constructions like with this medicine, he slept in a few minutes, where slept is interpreted as fell asleep: (i) (ii)

(?)

The kids didn’t become healthy in two months. The shop doesn’t open in two hours.

(?)

As the temporal structure of events is not our main concern, I will not pursue the issue any further.

BECOME

AND ITS

PRESUPPOSITIONS

211

and (b) [e: :[p]], where (a) could result from clausal negation, while (b) is a lexical property. Although there are further problems to be taken into consideration, I will assume for the time being that the entry for remain is at least on the right track in this respect. A final comment is to be made with respect to causative verbs like those in (51), which share the presupposed state indicated in (53) with their inchoative counterparts in (52): (51) a. b. c. d.

Fred opened the shop the next day. The sun dried the sheets quickly. The book made the author famous. The police killed the ferocious dog.

(52) a. b. c. d.

The The The The

shop opened the next day. sheets dried quickly. author became famous. ferocious dog died.

(53) a. b. c. d.

The The The The

shop was not open. sheets were wet. author was not famous. ferocious dog was alive.

Before turning to the computation of presuppositions for causative verbs, two side-issues might be mentioned, which need not concern us here. First, causative verbs (and causative constructions in general) mostly, but not necessarily involve inchoativity, which is the source of the presupposition we are looking at. But causation may be responsible for states, rather than events. (54) illustrates causation without inchoativity: (54) a. b.

The new engine turns the wheel very steadily. The fire brigade held the rope tight.

Here, no change of state is asserted or negated, hence no presupposed state is to be computed. Second, causatives may introduce further presuppositions, due to the required type of agent or the sort of causation. (55) illustrates some possibilities: (55) a. b.

They strangled the night-watchman. Two hooligans stabbed the foreigner.

Whatever presuppositions strangle, stab and other verbs might add to the conditions of killing – they need not concern us here, except

212

MANFRED BIERWISCH

that we must make sure that they do not interfere with the construction of the presupposition induced by BECOME, to which we will turn now. The basic schema of causative verbs has already been sketched with respect to the ambiguity of open in (23b) above, which I repeat here for the sake of discussion: (23) b. /open/ [ þ V,  N]

lx ly le [e: [[ACT y] [[CAUSE [BECOME [OPEN X]]]]]

What we need to provide is the presupposition triggered by the inchoative open incorporated in (23b). This should, of course, in principle be accomplished by the constraint (34). To this effect, two problems must be clarified, however. The first of these problems was already mentioned with respect to negation showing up in remain. The generalization hinted at in that connection must deal with the fact that the presupposition of BECOME is to be computed locally, i.e. without reference to the operators taking BECOME as argument. The elements to be ignored would then include not only the negation, as discussed earlier, but also the causative operator – or more correctly CAUSE and the specification of the causing event. To this effect, (34) would have to be modified as shown in (34u), indicating that BECOME triggers its presupposition, even if it is subordinate to other operators:  (34u) ½e : ½. . . ½become ½p ) ½ s  e ^ s : :½p ½e : ½. . . ½become ½p The locality intended by this formulation is subject to crucial conditions governing the computation of the presuppositional component P for arbitrary complex expressions.26 These conditions must, among others, guarantee the clause headed by a given verb as the domain gathering the pertinent presuppositions. Thus the embedded clause (John) to have opened the window in (56a) and (b) must have the presupposition the window was closed, although this becomes a presupposition of the matrix clause only within the scope of the factive verb regret, but not of the non-factive verb believe: (56) a. b.

Johni believed [ei to have opened the window] Johni regretted [ei to have opened the window]

26 There is a large literature on the so-called projection problem of presuppositions dealing with these conditions. A first systematic integration of the conditions in question has been proposed in Karttunen and Peters (1979). The problem is taken up in Kamp (2001) from a somewhat different perspective, which I will adopt here as background.

BECOME

AND ITS

PRESUPPOSITIONS

213

This would fall out naturally, if we assume that presuppositions are unified with respect to the event variable, which defines in a way the boundary of the domain in (34u). This gets us to the second, more intricate problem. Notice that the relation between causative and inchoative verbs explicitly exploited in the notational variant (24) shifts the event reference represented by ‘‘e’’ from the change of state to the causation of the change. This is intuitively correct, as far as the verb’s event reference is concerned, but it requires clarification of the temporal relation between the causation and the caused change, since the presupposition must be temporally located in this respect. As discussed above, the time structure of a change of state is not a trivial issue, the change being defined only by the intervals it overlaps. An even more intricate problem is the temporal location of the event represented as e: [p CAUSE q] with respect to the time of p and q. Consider, for example, a simple case like Fred opened the door. In the simplest case, Fred’s activity and the position of the door have the same location in time. But suppose Fred opens the door pushing a button that releases the door, for reasons of security, with a delay of three minutes. Then the time of p precedes that of q, without even a temporal overlap between p and q.27 There are two strategies to deal with these problems in view of (34u). The first option is to rely on the fact that configurations of the type [p [CAUSE [BECOME q]]] are always to be construed as direct causation, not allowing for any explicitly indicated intermediate events. With this proviso, the source state of the change [BECOME q] will just as well be a presupposition for the change in question as for its causation. With this construal, (34u) would account for the presupposition of both the inchoative and the causative verbs. The second option takes into account the different events e and eu, instantiating [p [CAUSE q]], and q, respectively, with a corresponding reformulation of (34u). As this reformulation would be reasonable only if it takes into account further considerations, which go beyond the scope of the present concern, I will adopt the first option, assuming that (34’) correctly computes the presupposition associated with BECOME.

27 The problem has been discussed, among others, in Dowty (1979), where three possibilities are considered: the time of e: [ p CAUSE q ] coincides with (a) the time of q, (b) the time of p, and (c) the smallest interval that overlaps with the time of p and the time of q. Dowty considers (c) as the most plausible assumption in most cases.

214

MANFRED BIERWISCH

These considerations carry over to the relatively small group of verbs where causativity combines with the negation of inchoativity. A case in point is keep in one of its readings, as exemplified in (57): (57) a. b.

Max kept the kids running. Mary didn’t keep these things in good shape.

The presupposition in (a) is that the kids were running, so that Max caused them to continue, while (b) presupposes, that things were in good shape and denies that Mary caused the continuation of this state. (58) is an entry from which these conditions would follow:28 (58) /keep/

[ þ V,  N]

lP lx ly le [e: [[ACT y] [CAUSE :[BECOME :[P x]]]]] [X]

The morpho-syntactic condition indicated by [X] has to specify the properties characterizing the different predicatives keep allows for. Thus in keep them running ‘‘P’’ is replaced by the SF of them running, which eventually gives (59): (59) /keep them running/ [ þ V,  N]

ly le [ x|{s  e 4 s [RUN x]} [e: [[ACT y] [CAUSE :[BECOME :[RUN x]]]]]]

The underlined variable ‘‘x’’ is a provisional indication of the discourse referent introduced by the pronoun they (or them, for that matter), which binds the argument x of RUN.29 28 The various other readings of keep are, of course, not unrelated to the one considered here. They are thus subject to similar considerations like the entries for open, get, etc. discussed earlier. Thus the relation between (57a) and the intransitive counterpart The kids kept running could be captured by a complex entry like (i), using the notational devices mentioned above:

lP lx (aly) le [e: [(a[ACT y] [CAUSE) :[BECOME :[P x]]]]] [Y] I will not go into those issues, being concerned here primarily with the computation of presuppositions. 29 The notational proposal to represent discourse referents as a component of SF, separated by |, is adopted from Maienborn (2002).

(i) /keep/

[ þ V,  N]

BECOME

AND ITS

PRESUPPOSITIONS

215

BECOME, Events, and Processes In this section, I will take up the subtle, but intricate difference illustrated in (60) and (61), in order to show how it can be accommodated by the analysis of BECOME proposed so far. (60) a. The heap became six feet high {within two days/#for quite a while}. b. The heap became higher than six feet {within two days/#for quite a while}. c. The heap became higher (and higher) {#within two days/for quite a while}. (61) a. Hans wurde {in kurzer Zeit/#eine Weile lang} so groX wie Paul. ({In a short time/#for quite a while}, Hans became taller than Paul.) b. Hans wurde {in kurzer Zeit/#eine Weile lang} gro¨Xer als Paul. ({In a short time/#for quite a while}, Hans became taller than Paul.) c. Hans wurde {#in kurzer Zeit/eine Weile lang} (immer) gro¨Xer. ({#In a short time/for quite a while} Hans became taller (and taller).) What is at issue is the fact that the (a)- and (b)-sentences express changes with a well defined target state in the sense discussed so far, whereas the (c)-sentences describe processes not terminating in a final result state, as borne out by the mutually exclusive temporal adverbials they naturally combine with. Correspondingly, the (a)- and (b)sentences make the usual presupposition – the heap was not six feet high and not higher than six feet in (60a) and (60b), respectively, Hans was not as tall as and not taller than his brother in (61a) and (61b), respectively – while it seems to be fairly unclear what happens to the presupposition in the (61c) sentences. I will try to clarify these problems in three steps. First, as the issue clearly has to do with the semantics of comparatives, I will begin with a sketch of the relevant ingredients of positive and comparative adjectives. Relying on proposals discussed in Bierwisch (1989), the elements minimally needed for an adjective like high and its antonym low are indicated in (62), where VERT abbreviates a function that assigns an object x its vertical extension (or position), i.e. its height, which is then related to the sum [v þ d] or the difference

216

MANFRED BIERWISCH

[v  d], indicating that the height exceeds or falls short of some value v by some difference d:30 (62) a. /high/ b. /low/

[ þ N, þ V] (ld) lx [[VERT x] Z[v þ d]] [ þ N, þ V] (ld) lx [[VERT x] r [v  d]]

The amount of d might be specified by a measure phrase, as in (60a), or it might be left implicit, as in the heap was high. This is indicated by the optionality of the position ‘‘ld’’. If d is left implicit, the value of v must be something like the expected standard or normal case N, while the specification of d, for example, by the value of six feet, requires d to be measured from the ground, such that the value of v must be 0.31 In other words, v is a kind of parameter, the two values of which are N and 0, chosen according to general principles of interpretation. With these assumptions, we get the VP-representations in (64), if (63) abbreviates the entry for the copula be, discussed a.o. in Bierwisch (1997), Maienborn (2002): (63) /be/ [ þ V,  N]

lP lx ls [s: [P x]]

(64) a. /be six feet high/ [ þ V,  N] lx ls [s: [[VERT x] Z [0 þ 6 FEET]] b. /become six feet high/ [ þ V,  N] lx le [{s  e 4 s :[[VERT x] Z [0 þ 6 FEET]]} [e: [BECOME [[VERT x] Z [0 þ 6 FEET]]]]] It might be worth noticing that in (64b) the presupposition induced by BECOME is correctly specified as just the negation of the target state: Eventually the heap became six feet high presupposes that the heap was six feet high was false before the change, without any 30

The value of [VERT x] is not simply equal to [v þ d], but must cover or even exceed it. This must be assumed for a number of reasons, one being the fact that negating [[VERT x]Z [v þ d]] does not allow [VERT x] to be greater than [v þ d]. Hence [[VERT x] Z [v þ d]] means [VERT x] is not less than [v þ d], with [[VERT x] ¼ [v þ d]] as default interpretation. The same consideration holds for the converse relation r in the antonym low, where the value of [VERT x] is less than or equal to [v  d]. 31 This explains, by the way, why *two feet low is anomalous: Since the negative adjective requires the value of d to be subtracted from v, no height would result if v is 0. This does not hold, as we will see immediately, for the well-formed two feet lower, since the comparative provides a value for v that is necessarily different from 0. It might be noted at this point that there is general agreement with respect to the ingredients of this analysis, although they have been formulated in a number of different ways. For surveys see von Stechow (1984) and Bierwisch (1989).

BECOME

AND ITS

PRESUPPOSITIONS

217

specification of the heap’s height (or even existence) – except that it could not be six feet or more. The second step is to supply the comparative, whose essence is to provide an explicit possibility to specify the value of v as a standard of comparison, which is the six feet in (60b), and (the height of) Paul in (61b). Putting aside the problems involved in morphological rules deriving higher than from high, the essential point can be expressed by the following entries for comparative adjectives, which provide an additional argument position, making the variable v available for syntactically specified values: (65) a. /high þ er/ [ þ N, þ V] b. /low þ er/ [ þ N, þ V]

(ld) lv lx [[VERT x] Z [v þ d]] (ld) lv lx [[VERT x] r [v  d]]

On the basis of these entries, VPs like (66) can be constructed, where the optional position ‘‘ld’’ is omitted, hence d is unspecified, while v is replaced by the SF of six feet: (66) a. b.

/be higher than six feet/ [ þ V,  N] /be lower than six feet/ [ þ V,  N]

lx ls [s: [[VERT x] Z [6 FEET þ d]]] lx ls [s: [[VERT x] r [6 FEET  d]]]

Notice that in (66a) – differing from (64a) – 6 FEET is the value of v, rather than the difference variable d, which may be additionally specified, as shown in ten inches higher than six feet or twelve inches lower than ten feet. With these independently motivated ingredients the representation of the VP of (60b) seems to be straightforward: (67) /become higher than six feet/

[ þ V,  N]

lx le [ v|{s  e 4 s: :[[VERT x] Z [6 FEET þ d]]} [e: [BECOME [[VERT x] Z [6 FEET þ d]]]]]

A remark is necessary here concerning the implicit variable ‘‘d’’, which I have construed in (67) as a kind of parameter, the value of which is automatically copied by the constraint (34u). This seems to be correct in view of the fact that ‘‘d’’ can be explicitly specified in constructions like become ten inches higher that six feet, where ‘‘d’’ has the value of ten inches. With this proviso, (67) correctly specifies the shift to x’s exceeding the height of six feet by d, presupposing that it did not before the change. Even though this seems to be the appropriate treatment of d in cases like (60b), we have to consider further aspects, as we proceed. The third step has to provide an account of the process expressed by the (c)-sentences in (60) and (61). One point to notice here is that constructions like these are ambiguous between the process reading in question and an event reading, which is possible only if the optional

218

MANFRED BIERWISCH

repetition higher and higher in (60) and the optional adverbial immer (always) in (61) are omitted. This event reading differs from the (b)sentences by its lack of an explicitly specified value for v, relying on some contextually given value to compare with. Under this interpretation, (60c) is in a way elliptical, leaving the argument position ‘‘lv’’ unsaturated. The VP that accounts for this reading of (60c) can best be represented as in (68), where ‘‘v’’ indicates a discourse referent in the sense mentioned earlier. Thus ‘‘v’’ indicates a contextually specified discourse referent, which supplies the lacking than-complement (say the height of the wall nearby), and it correspondingly binds also the occurrence of ‘‘v’’ in the presupposition: (68) /become higher/ [ þ V,  N] lx le [ v|{s  e4 s::[[VERT x]Z [v þ d]]} [e: [BECOME [[VERT x]Z [v þ d]]]]]

It should furthermore be noted in passing that this reading is compatible with a measure phrase specifying the value of ‘‘d’’, as in the heap became ten inches higher (than the wall), which suggests, that not only v, but also ‘‘d’’ be copied in the negated condition of the presupposition. The crucial point now is the treatment of the process reading. This reading characteristically excludes any explicit specification of the values for either ‘‘v’’ or ‘‘d’’, as (69) shows: (69) a. The heap became (*two feet) higher and higher (*than the wall). b. Hans wurde (*10 cm) immer gro¨Xer (*als Paul). (Hans became (*10 cm) taller and taller (*than Paul).) The actual impact of (69) is that the process reading does not only prevent explicitly specified values for either ‘‘v’’ or ‘‘d’’, it induces in fact a completely different regime over these variables. The intuitive notion that suggests itself is that the process is to be construed as a continuous sequence of changes each of which adds some amount d to the value v reached before. This means with respect to ‘‘v’’, that it must not refer to any discourse referent of its own, but merely to a value defined by the previous state of x. With respect to ‘‘d’’, its value can only be determined by the successive states of x and the value of [VERT x] along these states. In other words, the process reading of become þ Comparative takes an earlier state of x as defining the value of ‘‘v’’. This allows neither ‘‘v’’ nor ‘‘d’’ to be specified by values external to the process of x’s continuous changing. In order to make this notion more explicit, let vi be the value of [VERT x] at some time ti,

BECOME

AND ITS

PRESUPPOSITIONS

219

where ti precedes the time tui of the target state su of the event e under description. Suppose furthermore that we represent by ‘‘di’’ the difference between the value of [VERT x] at ti and tui, such that the value of [VERT x] reaches [vi þ di] at tui. With this assumption we get (70a) as a possible representation for the incremental steps constituting the process in question, with (34u) computing the standard presupposition spelled out in (70b): (70) a. /become higher/ [ þ V,  N] b. /become higher/ [ þ V,  N]

lx le [ e: [BECOME [[VERT x] Z[vi þ di]]]] lx le [{s  e4 s: : [[VERT x]Z [vi þ di]]} [e: [BECOME [[VERT x]Z[vi þ di]]]]]

The presupposition in (70b) requires [VERT x] to be below [vi þ di] before the change instantiated by e takes place. This condition, although apparently correct, leaves a problem to be clarified, because it is crucially less specific than the assumption already made, namely that [[VERT x] ¼ vi] holds at ti. Now the condition :[[VERT x]Z[vi þ di]] (or its formal equivalent [[VERT x]o[vi þ di]]) converges with this condition to exactly the degree to which di approximates 0. In other words, ti with the condition [[VERT x] ¼ vi] becomes a more specific condition imposed on the presupposed source state of e, if we take d ¼ 0 in the presupposition. With this consideration in mind, we may replace (70b) by the more restrictive (70c): (70) c. /become higher/ [ þ V,  N]

lx le [{s  e4 s:[ [VERT x] ¼ vi]} e: [BECOME [[VERT x]Z[vi þ di]]]]]

Notice that no specific condition on the time ti fixing vi is needed: It simply precedes the final state of e, according to the conditions holding at both states. What should be observed, however, is the specific regime that this option requires for the variables ‘‘vi’’ and ‘‘di’’. We already noted that they are not available for syntactically specified values, and they are not open to contextual interpretation either, although they are under strict, albeit implicit, control. The variable v in expressions like (62), for which the comparative would create a special argument position, now obviates any recourse to either syntactic or contextual specification. The same holds for d in (62), whose counterpart in (70) is moreover subject to its own specific conditions, indicated by the value di. As di was introduced to represent the difference between the extent of [VERT x] at ti and tui, its actual value is dependent on vi and t. As a matter of fact, the value of d under this construal becomes a function of vi and t; it ceases to be an independent parameter, in case it is not explicitly specified. In order to see to what extent these assumptions – and

220

MANFRED BIERWISCH

representations like (70) that rely on them – provide an appropriate account of the process character of getting higher (and higher), two problems need to be clarified. First, as already mentioned, become higher in the relevant reading of (60c) is supposed to express a continuous change, while (70) seems to represent a single event instead of a homogeneous process. Second, as (60c) is neither elliptical nor open for explicit specification, the origin of vi that provides the value of comparison needs to be clarified. These problems, which are not independent of each other, shall be discussed in turn. The basic intuition about states and processes has been made explicit, for example, in Bach (1986): Each part of a process e must meet the same condition as e as a whole. Complying with this requirement, each part of x’s getting higher is itself an instance of x getting higher. This intuition can be captured in two ways. The first is to explicitly turn the eventuality e referred to in cases like (70) into a (potentially infinite) set of sub-eventualities all subject to the condition expressed in (70), a proposal sketched a.o. in Bierwisch (2000). The necessary ‘‘grinding’’ of e can be accomplished by a template like (71), which distributes the condition P from an argument z to its (im)proper parts, thus warranting the decisive property of processes (as well as mass terms): (71) lP lz ½8zk ½zk  z½Pzk  A well-known case in point is the mass interpretation of count nouns like chicken in examples like There was chicken all over the plate.32 Applying this template to (70a), we derive (72):33 (72) /become higher/ [ þ V,  N] lx le [’zk [zk D e] [zk: [BECOME [[VERT x]Z [vi þ di]]]]] This leaves us with the task to relate the variable ‘‘vi’’ indicating the height at the previous stage to the sub-event zk. Intuitively, one 32

This is not quite correct as an account of the mass-interpretation available for nouns like chicken, lamb, egg, etc. The mass noun property is rather a consequence of construing chicken by what has been called conceptual shift in the sense of meat from chicken. This concept is then already subject to the condition (71). What is to be noted here is merely, that (71) is not an ad hoc stipulation. 33 Technically, (71) combines with (70a) by functional composition, with lP being saturated by the predicate le [e: [BECOME [[VERT x]Z [vi þ di]]]]. Then lambda conversion yields (72), where ‘‘e’’ replaces ‘‘z’’ and the argument position ‘‘lx’’ is inherited from (70a).

BECOME

AND ITS

PRESUPPOSITIONS

221

would like to say that vi is the value of [VERT x] resulting from the previous sub-event zk  1. A moment’s reflection shows that this is already inherent in the stipulation by which vi was introduced, according to which ti has to precede tui, and ti is moreover the time of the source state of e, according to the reasoning underlying (70c). Hence without further stipulation, we get (73) with the presupposition spelled out as discussed with respect to (70c) where iok and tiotk for arbitrary i and k. (73) /become higher/ [ þ V,  N] lx le [’zk [zk De] sk  zk4sk :[[VERT, x] ¼ vi]]} [zk : [BECOME [[VERT, x] Z[vi þ di]]]]]

What (73) in fact expresses is that e consists of subevents zk, each of which presupposes a state where the height of x is vi and ends in a state where it is vi þ di, which then becomes the presupposition of a possible subsequent event zku. This turns e into an eventuality that meets exactly the conditions of a process.34 The interesting point is that the presupposition automatically triggered by BECOME takes care of the internal ordering among the subevents collected in e.35 This consideration directly leads to the second possibility to account for the homogeneity in the process of getting higher (and higher). It consists in the proposal to simply take (70c) as it stands, relying on the following consideration. The homogeneity that holds for mass terms, states, and processes alike is usually not explicitly represented in the semantic representation, but taken as an intrinsic characteristic that comes into play if the pertinent properties are at

34 It has sometimes been objected, for example in Steinitz (1999), against this line of reasoning that processes like getting higher are not sequences of events but rather continuous changes. In a similar vein, Steinitz argues, Jackendoff (1996) rejects the view that motion can be characterized by a finite set of subevents. ‘‘A finite sequence of subevents necessarily has as specified beginning and ending, so it cannot encode the absence of endpoints.’’ (p. 316). The misunderstanding comes from assuming a finite sequence of subevents. The collection of subevents relied on in (71) is neither discrete nor finite, just like conceptually an amount of water is neither a discrete nor a finite set of subquantities of water. What the condition in (71) expresses is merely the requirement that whatever zk might be chosen, it has the same property as z and any other part of it. This is exactly what Bach (1986) assumes for processes. 35 It might be noted that the optional adverbial immer in German and the optional repetition of the comparative higher and higher in English and German indicate somehow the iteration introduced by (71). Although this looks plausible in principle and might be correct etymologically, it is not an effect of compositional interpretation and has to be left as a remark about plausibility.

222

MANFRED BIERWISCH

issue. Hence entries like (74) rather than (75) are taken as canonical for the mass- and process-terms water and sleep, respectively: (74)

a. /water/ b. /sleep/

[ þ N,  V] [  N, þ V]

lz [WATER z] lx le [e:[SLEEP x]]

(75)

a. /water/ b. /sleep/

[ þ N,  V] [  N, þ V]

lz[’zj [zj D z] [WATER z] lx le [’zj [zj D e] [zj: [SLEEP x]

The conditions made explicit in (75) do in fact hold in (74): They are inherent conditions of the predicates WATER and SLEEP, respectively.36 We might, in a similar vein, suppose that the process character of getting higher is a matter of the predicate applying to e in (70), where the condition by which it differs from the (elliptical) event interpretation in (68) is just the variable ‘‘vi’’ which is not bound by an (implicit) discourse referent, like the ‘‘v’’ in (68), but related to a previous state of x, such that no event-external target state gets defined. Viewed in this way, the property defined in (70) is inherently a process, just like the properties defined by SLEEP or MOVE are inherently processes. In other words, the event represented in (68) and the process represented in (70) differ just as much as a lamb and lamb – or whisky and a whisky, for that matter, the difference formally represented by the discourse referent v in (68) and the unbound parameter vi with the value of di depending on it in (70). This gets us to the second problem to be clarified with respect to (70), namely the source and the status of the value vi as well as di, which need to be treated separately. As to vi, a straightforward way to clarify the issue seems to be the assumption that the position ‘‘lv’’ of the comparative is removed from the argument structure in entries like (65), leaving ‘‘v’’ as a parameter, for which ‘‘vi’’ is just a notational variant to make its character explicit. By this assumption, (70) would not be elliptical, as (68) indeed is, and it ceases to represent an event that ends with a distinct target state. This sort of reasoning, however, runs into serious difficulties: Even though there are different proposals as to how the comparative is to be represented, there is clear consensus that relating to an entity to be compared with is the distinguished property of comparatives. In terms of the analysis assumed here, it is the very position ‘‘lv’’ in (65) that 36 There are in fact templates having the inverse effect of the ‘‘grinder’’ (71) deriving count nouns like two beers or more than one whisky with different ways to support the individuation. Note also the comments in footnote 32 on shifted count nouns.

BECOME

AND ITS

PRESUPPOSITIONS

223

represents this property. Hence removing it from the entry of the comparative deprives the comparative of its core characteristic. Moreover, the notational contrast between ‘‘vi’’ and ‘‘v’’ is not just a hint to the reader but has systematic consequences, since ‘‘vi’’ is not available for the values 0 and N, the options for ‘‘v’’ in the positive. Hence ‘‘vi’’ (or whatever notation one might choose) must have a systematic status by itself. In short, then, ‘‘vi’’ must be different from ‘‘v’’, presumably related to the comparative morpheme in one way or the other, and it is not available for binding by an argument position or an (implicit) discourse referent. For the time being, I will leave it at that, briefly returning to the problem below.37 I will assume, however, that eventually an account should be available which reduces this problem to independently motivated assumptions. Although the status of di requires an account by its own, it is clearly not independent of vi, hence I will not make an ad hoc proposal for this element either. To sum up the considerations about the contrast illustrated in (60) and (61): If BECOME combines with a comparative not providing a value of comparison, the result denotes a process rather than an event. The reason is the lack of a proper target state: The property of x defined by [[DIM x]Z[vi þ di]] – where [DIM x] stands for any dimensional characteristic of x – is only fixed with respect to the change-internal value vi. In this sense, the parameter vi turns the event marked by BECOME into a continuous process, where the presupposition provided by (34u) is restricted in a way which guarantees the coherence between the parts of the process, providing each possible stage with its presupposed predecessor. As a side-effect, the collection of values for ‘‘di’’ maps the growths (or reduction) of [DIM x] onto parts of the time interval of e, whatever partitioning of the process e is considered. Additional motivation for this analysis comes from the fact that verbs like fall, rise, grow, increase, reduce, change, and quite a few others, denoting transition processes which Fabricius-Hansen (2001) calls counterdirectional, can most naturally be construed as incorporating comparative constructions of the sort just discussed, as examples like (76) and their close synonyms (77) show: (76) a. b. c. d. 37

The price of gas rises. The interest rate fell rapidly. His debts increased. The tree grew, but only slowly.

See footnote 39.

224 (77) a. b. c. d.

MANFRED BIERWISCH The price of gas becomes higher. The interest rate rapidly became lower. His debts got larger. The tree became taller, but only slowly.

Assuming that fall and rise indicate movements on a vertical scale, the entry for rise can be given as (78).38 (78) /rise/

[ þ V,  N]

lx le [e: [BECOME [[VERT x]Z[vi þ di]]]]

This entry triggers a presupposition in much the same way in which (70a) fosters the presupposition spelled out in (70c). There is one nontrivial point to be added here, however. (78) is intended to account for cases like (76a), which do not come with a specification of the amount of rising. But such a specification is possible, as shown in (79): (79) a. The price of gas rises at least ten cents. b. The river has risen two feet. Cases like these require two amendments. First, like dimensional adjectives and their comparatives – see (62) and (65), respectively – rise and fall admit an optional argument position, by which the value of ‘‘d’’ can syntactically be specified by a Measure Phrase. This leads to the following modification of (78): (80) /rise/ [ þ V,  N]

(ld) lx le [e: [BECOME [[VERT x]Z[v þ d]]]]

Second, in order to accommodate a specification of ‘‘d’’, which (80) makes possible, ‘‘v’’ must provide a value this specification can be added to (or subtracted from). This is, according to the assumptions made above, an automatic consequence of replacing ‘‘di,’’ which is completely dependent on the occurrence of ‘‘vi’’, by the variable ‘‘d’’, which is bound by the argument position ‘‘ld’’. The value of ‘‘v’’ can only be 0 or some contextually specified amount to compare with. In cases like (79), this must be due to some discourse referent v taken from the previous context. In other words, it must not be a ‘‘processinternal’’ value ‘‘vi’’, but rather an externally given target v. For this 38 It might be noted that I ignore here (and throughout) the alternation between vertical extension and vertical position. As a matter of fact, high is ambiguous between extension and position in high window, while high position is restricted to location, and high mountain to extension. In the same way, rise alternates between position in the sun rises and extension in the river rises and is presumably ambiguous in the ground rises. Hence VERT is underspecified in this respect, sensitive to conditions not to be explored here.

BECOME

AND ITS

PRESUPPOSITIONS

225

reason, I have switched back in (80) to ‘‘v’’ instead of ‘‘vi’’, supposing that ‘‘v’’, if it is not bound by some discourse referent v, allows for the value 0, N, or vi, depending on general conditions of semantic interpretation.39 With this proviso, the VP of (79b) will be something like (81): (81) /rise two feet/ [ þ V,  N] lx le [ v | e: [BECOME [[VERT x]Z[v þ 2

FEET]]]]

One might observe, incidentally, that the presupposition ‘‘s: :[[VERT x] Z [v þ 2 FEET]]’’, which (34u) triggers in (81), is implied by the narrower condition ‘‘s: [[VERT x] ¼ v]’’. Notice, moreover, that rise as a lexical entry with the optional argument ‘‘ld’’ is parallel to its close synonym become higher as discussed with respect to (60c), the two readings of which are made explicit in (76) versus (78). One consequence of an explicit value assigned to ‘‘d’’ is that, for example, rise two feet is an event (or more specifically an accomplishment according to the Vendler-terminology), and not a process, since only the final part of the event satisfies the condition of having added two feet, much like became two feet higher or became higher than six feet.40 To summarize these considerations: the occurrence of ‘‘vi’’ (or whatever represents the process character of rise, become higher, etc.) is incompatible with the assignment of process-external values to either ‘‘v’’ or ‘‘d’’. In this sense, vi and di warrant the homogeneity of the continuous change expressed by rise, fall, and their equivalents.

The Interaction of BECOME and AGAIN A final consideration will take up one of the issues repeatedly discussed with respect to BECOME and illustrated by the ambiguity of cases like (82): (82) At six the shop closed again. 39 Two points of the otherwise fairly complex conditions should be sorted out here. First, vi must now be considered as one of the values available for ‘‘v’’, if ‘‘v’’ is not bound by the argument position ‘‘lv’’. Second, for the value vi to show up, the variable ‘‘d’’ must not be assigned any specific value. In other words, if ‘‘d’’ is specified by a measure phrase or any other complement, ‘‘v’’ cannot be construed as vi. For discussion of the conditions governing the choice of 0 and N for ‘‘v’’, see Bierwisch (1989). 40 Notice that the discourse referent v in (81) binds also the variable ‘‘v’’ in the presupposition generated by BECOME, as shown in (i):

(i) /rise two feet/ [ þ V,  N] lx le [v| {s  e4s: : [[VERT x]Z[v þ 2 [e: [BECOME [[VERT x]Z[v þ 2 FEET]]]]]

FEET]]}

226

MANFRED BIERWISCH

Under the so-called restitutive reading, (82) presupposes that the shop was closed before the event in question, under the repetitive reading, it presupposes a previous event of closing, hence a previous period of the shop’s being open. The restitutive reading is preferred under nuclear stress on close, while nuclear stress on again forces the repetitive reading. The standard analysis relates this difference in one way or the other to the scope of the particle again.41 Thus adding the particle again to a VP like close in (82) introduces the presupposition that either the event as a whole (repetitive) or merely the state resulting from it (restitutive) occurred already before. The semantics of again must therefore proliferate the semantic structure f of the constituent it combines with, adding the presupposition that f or the result of f occurred before. In order to capture this alternation, I will define an operator RES that suppresses the transition, retaining merely the eventual resulting state: (83) s: [RES [BECOME p]] ¼ def [s: p] By means of this operator, we can express the two readings induced by again by the entry (84), where the categorization [Part] is completely provisional, indicating that again is a particle adjoined to a [ þ V]-constituent of as its syntactic head: (84) /again/

[Part]

lP lx le [{(eu[euoe4eu: (RES)[P x]]} e: [P x]] [ þ V]

The condition ‘‘euoe’’ indicates that the event eu and hence its result precedes e. The optionality of RES accounts for the two readings of again: If RES shows up, we get the restitutive case, if it is dropped, we get the repetitive reading.42 Applying the two variants of this entry to 41

For an overview of the various attempts to account for these phenomena see the references given in footnote 7, and also Bierwisch (2000), and Fabricius-Hansen (2001). The following discussion will be restricted to the semantic effect that again has on expressions with BECOME. I must refrain from dealing with problems of nuclear stress, position of again, and the dependencies among them. 42 This optionality is not the correct solution, though, as the restitutive reading is restricted to again without nuclear stress, which is again restricted syntactically. A first step to account for these additional conditions might be expressed as in (i), using the index-mechanism mentioned earlier: (i) /again/ [Part, aFocus] lP lx le [{(eu[euoe4eu: (bRES) [P x]]} e: [P x]] [ þ V] Condition: a ¼ plus implies b ¼ minus This is only a first approximation for reasons mentioned in footnote 41. In particular, (i) does not spell out the syntactic and prosodic properties connected to [7Focus], regulating stress and surface position.

BECOME

AND ITS

PRESUPPOSITIONS

227

the intransitive verb close in (85) – the straightforward antonym of open in (33) with the BECOME-presupposition supplied – yields (86) for the restitutive and (87) for the repetitive reading, respectively. Let us first look at (86), where RES picks out the target state :[OPEN x] and adds it to the presupposition component, fixing it as preceding the event e. As the state s: [OPEN x] induced by BECOME and su: :[OPEN x] induced by again cannot be simultaneous, the letter can only precede the former: (85) /close/ [ þ V,  N] lx le [{s  e4s: [OPEN x]} e: [BECOME:[OPEN x]]] (86) /close again/ [ þ V,  N] lx le [{suoe4s  e4su: :[OPEN x]4s: [OPEN x]]} e: [BECOME :[OPEN x]]]

Turning to the repetitive reading, we notice that first again adds the complete previous event eu: [BECOME :[OPEN x]] to the presupposition component, which then triggers its own presupposition, indicating that eu also starts with the condition su: [OPEN x]: (87) /close again/ [ þ V,  N]

lx le [{euoe4su  eu4s  e4su: [OPEN x]4 eu: [BECOME :[OPEN x]]4s: [OPEN x]]} e: [BECOME :[OPEN x]]]

Notice that the correct sequence of states OPEN - :OPEN - OPEN - :OPEN with the first three states belonging to the presupposition, while the last one marks the eventual result, automatically derives from the entries for again and close plus the template (34u). Let us finally turn to the intricate cases of BECOME that alternate between event and process. For the sake of illustration, I will restrict the discussion to the cases in (88), where underscoring indicates focus stress: (88) a. The river rises slowly. b. The river has risen two feet. c. The river has risen two feet again. d. The river has risen two feet again. / Again the river has risen two feet. e. The river rises again. f. The river rises again. / Again the river is rising. To begin with, (88a) allows for an event- and a process-reading, as already discussed. For the sake of reference, l repeat as (89) the entry given in (80), supplied with the presupposition induced by BECOME, from which rise two feet derives as already shown in (81), where the process reading is blocked by the specification of d, which

228

MANFRED BIERWISCH

requires v to be contextually specified, as indicated by the discourse referent v: (89) /rise/

[ þ V,  N]

(81) /rise two feet/

[ þ V,  N]

(ld) lx le [{s  e4s: :[[VERT x]Z[v þ d]]} e: [BECOME [[VERT x]Z[v þ d]]]] (lx le [ v{s  e4s::[[VERT x]Z[v þ 2 FEET]]} e: [BECOME [[VERT x]Z[v þ 2 FEET]]]])

If this is combined with again, we get the two readings, as expected. Intuitively, the preferred interpretation of (88c) indicates the restitution of an earlier sate, where the river had a height that has now been reached again through rising by two feet. Formally, that comes out as (90), an automatic consequence of combining again with (81): (90) /rise two feet again/ [ þ V,  N]

lx le [ v|{suoe4s  e4su:[[VERT x]Z [v þ 2 FEET]]4s: :[[VERT x]Z[v þ 2 FEET]]} e: [BECOME [[VERT x]Z[v þ 2 FEET]]]]

Notice that, much like for restitutive close again represented in (86), three states are identified: the presupposed height of v þ 2, picked out by the result-operator RES of again, then less than v þ 2, presupposed by BECOME, and finally the target v þ 2, all with respect to a contextually fixed value v of ‘‘v’’. Looking next at the repetitive interpretation, which is the only possibility for (88d),43 where no RES reduces the presupposition of again to the previous occurrence of the target state, we get (91), with four successive states, as in (87) for close again, with necessarily two values v and vu to be compared with.44 (91) /rise two feet again/ [ þ V,  N] lx le [ v, vu|{euoe4su  eu4s  e4su::[[VERT x]Z[vu þ 2 FEET]]4 eu:[BECOME [[VERT x]Z[vu þ 2 FEET]]] 4s: :[[VERT x]Z[v þ 2 FEET]]} e: [BECOME [[VERT x]Z[v þ 2 FEET]]]]

43

Whether the repetitive interpretation is also available for (88c), can be left open here. Clearly the restitutive reading is preferred for (c) and definitely blocked for (d). 44 Taking the more restricted presupposition discussed above as a kind of default interpretation for comparative-like representations, we get (i) with a more perspicuous sequence of the four states in question: (i) lx le [ v, vu|euoe4su  eu4s  e4su : [[VERT x] ¼ vu]4 eu: [BECOME [[VERT x] Z vu þ 2 FEET]]]4 s: [[VERT x] ¼ v]} [e: [BECOME [[VERT x]r[v þ 2 FEET]]]]]

BECOME

AND ITS

PRESUPPOSITIONS

229

What (88d) expresses, does not necessarily mean that the two events add up to the river’s rising by four feet. As a matter of fact, e and eu are independent of each other, with falling or rising or not changing water in between. Let us finally turn to the processes in (88e) and (88f), where once again (88f) unambiguously denotes a repeated process, i.e. it asserts a process of rising, and it presupposes a previous process of the same type. Relying on the process reading of rise given in (78) with the presupposition added, we get (92) as the effect of combining it with repetitive again, i.e. the reading without the RES-operator: [ þ V,  N] (92) /rise again/ lx le [{euoe4su  eu4s  e4su: [[VERT x] ¼ viu]4 eu: [BECOME [[VERT x]Z[viu þ di]]] 4s: [VERT x] ¼ vi]} [e: [BECOME [[VERT x]Z[v þ d]]]]] As required, we get two independent processes, with the water rising, falling or simply not changing in between. The most intriguing case is (88e), which has readings that are compatible with at least the following conditions: (93) a. b.

The river rises again, after it didn’t change for a while. The river rises again, after it was going down before.

In (93a), the presupposition triggered by again is most naturally met by a previous process of rising, followed by an interval with no increase of water. This means that (93a) is an instance of repetitive interpretation of again. In other words, the VP rise again in (93a) must be that already spelled out in (92). A less prevalent reading of (93a) might however be compatible with a situation where the steady period was preceded by the river’s going down. This would be a reading that is explicitly required in (93a) and must hence be accounted for anyway, even though it cannot be represented by (92), which relies on the repetitive again. As a matter of fact, the presupposition of this reading must be compatible with the inverse process required by repetitive again, as explicitly stated in (93b). The presupposition that (88e) requires independently of the conditions added in (93) could presumably be paraphrased most plausibly as it was already higher before. This seems to be the minimal condition in order to justify the requirement of again – something that could be repeated in the process of rising, but is still compatible with various conditions in between, such as rising, falling, or steady height of waters, except that some falling is logically necessary. Otherwise the river cannot get back from the earlier height. Could this

230

MANFRED BIERWISCH

condition be derived by means of the restitutive reading of again? (94) shows the VP resulting from this possibility: (94) /rise again/ [ þ V,  N]

lx le [{suoe4s  e4su: [[VERT x]Z[vi þ di]]4 s: [[VERT x] ¼ vi]]} [e: [BECOME [[VERT x]Z[vi þ di]]]]]

What we’ve got here is the specification of a target and two presupposed states, just as in (90) for rise two feet again. The presupposition su: [[VERT x] Z [vi þ di]] is the automatic result of applying the restitutive reading of again, and especially its operator RES, to the process reading [e: [BECOME [[VERT x] Z [vi þ di]]]] of rise, which in turn triggers the second presupposition s: [[VERT x] ¼ vi]]. Differing from (90), however, the target state as well as the presupposed s: [[VERT x] ¼ vi] in (94) are not states external to the process, but by definition part of and dependent on the process, exactly as in (70c) for become higher. Now, this comment applies equally to the presupposed state that again computes from the target of the process: It is not a proper, external state that gets repeated, but a processinternal target. In any case the value one might chose for vi is surpassed by the presupposition vi þ di. Hence before the presupposition that initiates the actual rising is met, the inverse process must take place, which makes (94) compatible with the condition spelled out in (93b). It seems that this analysis accounts neatly for the apparently vague intuition about the interpretation of (80e) The river rises again, which doesn’t seem to be restitutive in a clear sense at first glance, as it does not presuppose a definite state nor a process to return to. But it presupposes a state holding before the actual rising of the river and correctly described by a bare comparative. This however is precisely what the restitutive reading of again applied to the process reading of ‘‘counterdirectionals’’ predicts. If this is correct, it supports the various assumptions introduced independently in a non-trivial way. The analysis furthermore seems to extend as expected to a wide range of related phenomena. Thus comparatives with BECOME create the same process readings, even if they are not originally dimensional, as in Things became better every day, the road got bumpier (and bumpier), the situation got more (and more) out of control, the weather became less friendly again.45 Similarly for other verbs ‘‘incorporating’’ BECOME 45 For some discussion of comparatives based on evaluative and other non-dimensional adjectives see Bierwisch (1989), where it is claimed that the comparative of a nondimensional predication [P x] introduces a template Q, such that Q [P x] maps x w.r.t. P on a scale of comparison. Technicalities are to be left aside here.

BECOME

AND ITS

PRESUPPOSITIONS

231

plus comparative as in the colors gradually faded in the sun, the road widened again, (due to the rain) slowly the grass grew again.

Loose Ends Although the properties of BECOME have been explored extensively in various frameworks, it is still worthwhile to take a closer look at its presuppositional structure. For one thing, even though the change to a target state starting from a source that implies the negation of the target is certainly a definitional condition of BECOME, it is not a triviality that the source state is a presupposition. It is one of the consequences of this fact, that in the very common cases of caused change as in kill, clean, and the causative variants of all ergative verbs like break, open, etc. it is the BECOME-component, rather than the causative operator CAUSE, that triggers the presupposition. This is borne out, among other things, by the fact that causative verbs without the BECOME-component, like turn as in the engine turned the wheel steadily do not have a presupposition of the relevant type. Interesting consequences of the presupposition triggered by BECOME show up if BECOME interacts with other elements affecting the relevant conditions. The two cases we have explored in more detail are the different effects of comparatives and the alternative readings of again, the semantic contribution of which consists in nothing but an additional presupposition. While the relevant representations have been spelled out in some detail, their systematic computation is in need of further clarification in at least three respects. The first aspect concerns the nature of implicit values of comparison, which show up in comparatives and result in the process character in combination with BECOME. Notationally, ‘‘vi’’ is intended to express the process-internal specification of the value, but so far there is no independent mechanism to specify and interpret the value. The stipulation seems to provide what is necessary. What we do want, however, is a way to relate this stipulation to a more general account of values for hidden variables. What comes to mind is the regime for implicit variables of complex tenses or the conditions regulating deictic specifications in locative prepositions. But this is merely a loose association with no substantive content to rely on. To this problem, the second aspect in need of clarification is related. It concerns the way in which the standard presupposition triggered by BECOME is restricted to the special condition [DIM x ¼ vi] instead of the more general [DIM xovi þ di]. While it is always possible

232

MANFRED BIERWISCH

to satisfy a presupposition by a more specific condition, the restriction in this case should be available for independent reasons. These reasons have to do with the special status of di, by which DIM x differs from vi. Intuitively, the value of ‘‘di’’ increases from 0 during each of the subeventualities that can be chosen from the process in question, although no change of di needs to be explicitly represented. Here too, more general mechanisms, rather than an appropriate stipulation, should eventually explain the source of the representations that seem to be intuitively correct. Finally, I have considered again as lexically ambiguous between a restitutive reading with the operator RES and a repetitive reading without it.46 RES is intrinsically related to BECOME, distinguishing between two types of presupposition.47 What needs clarification, as already mentioned in footnote 42, is the way in which the choice between repetitive and restitutive interpretation is related to the location of nuclear stress and the possible syntactic position of the particle. Similar, albeit different relations show up with particles like also and still. Whether these different conditions and effects can be reduced to general mechanisms, remains to be seen.

References Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9:1–16. Bierwisch, M. 1989. The semantics of gradation. In B. Manfred and L. Ewald, eds., Dimensional Adjectives, pp. 71–261. Berlin: Springer-Verlag. Bierwisch, M. 1990. Event nominalization: Proposals and problems. Acta Linguistica Hungarica 40:19–84. Bierwisch, M. 1997. Lexical information from a minimalist point of view. In C. Wilder, H.-M. Ga¨rtner, and M. Bierwisch, eds., The Role of 46

It has been argued by Dowty (1979), and more recently by von Stechow (1996), that the alternative interpretation of again can be reduced to one lexical reading, the difference being due to again having scope over BECOME or only the resulting state, the latter corresponding to the effect of RES. There are various problems with this proposal, however, some of which are discussed in Kamp and RoXdeutscher (1994) and Bierwisch (2000). 47 It must be added that the distinction between the two readings of again show up also if no BECOME is involved, as in he was sick again (repetitive) versus he was sick again (‘‘restitutive’’), where the interval between the earlier state and its restitution must be warranted in a different way. For further discussion, which must be left aside here, see Bierwisch (2000). It might be noted, though, that the two readings are subject to the same conditions that also govern the choice between repetition and restitution in the context of BECOME.

BECOME

AND ITS

PRESUPPOSITIONS

233

Economy Principles in Linguistic Theory, pp. 227–266. Berlin: AkademieVerlag. Bierwisch, M. 2000. Wieder-Ga¨nger. In J. Dolling and T. Pechmann, eds., Prosodie, Struktur, Interpretation, Linguistische Arbeitsberichte 74, pp. 67–82. Leipzig: Institut fu ¨r Linguistik, Universita¨t Leipzig. Bierwisch, M. 2002. A case for CAUSE. In I. Kaufmann and B. Stiebels, eds., More than Words, pp. 299–325. Berlin: Akademie Verlag. Bierwisch, M. and E. Lang. 1989. Dimensional Adjectives. Berlin: SpringerVerlag. Chomsky, N. and M. Halle. 1968. The Sound Pattern of English. New York: Harper and Row. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Fabricius-Hansen, C. 2001. Wi(e)der’’ and ‘‘Again(st). In C. Fery and W. Sternefeld, eds., Audiatur Vox Sapientiae, pp. 101–130. Berlin: Akademie Verlag. Haegeman, L. 1985. The GET-passive and Burzio’s generalization. Lingua 66:53–77. Higginbotham, J. 1985. On semantics. Linguistic Inquiry 16:547–593. Jackendoff, R. S. 1996. The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14:305–354. Kamp, H. 2001. The importance of presupposition. In C. Rohrer, A. RoXdeutscher, and H. Kamp, eds., Linguistic Form and its Computation. Stanford: Center for the Study of Language and Information. Kamp, H. and U. Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer Academic Press. Kamp, H. and A. RoXdeutscher. 1994. Remarks on lexical structure and DRSconstruction. Theoretical Linguistics 20:97–164. Karttunen, L. and S. Peters. 1979. Conventional implicature. In C.-K. Oh and D. A. Dinneen, eds., Presupposition, Syntax and Semantics 11, pp. 1–56. New York: Academic Press. Kiparsky, P. and C. Kiparsky. 1970. Fact. In M. Bierwisch and K.-E. Heidolph, eds., Progress in Linguistics, pp. 143–173. The Hague: Mouton. Klein, W. 1994. Time in Language. London: Routledge. Kratzer, A. 1994. The Event Argument and the Semantics of Voice. MS thesis. University of Massachusetts, Amherst. Maienborn, C. 2002. Die logische Form von Kopula-Sa ¨tzen. Berlin: Akademie Verlag. Reis, M. and I. Rosengren. 1997. A modular approach to the grammar of additive particles: The case of German Auch. Journal of Semantics 14:237–309. Steinitz, R. 1999. Die Kopula werden und die Situationstypen. Zeitschrift fu ¨r Sprachwis-senschaft 18:121–151.

234

MANFRED BIERWISCH

van der Sandt, R. 1992. Presupposition projection as anaphora resolution. Journal of Semantics 9:333–377. von Stechow, A. 1984. Comparing semantic theories of comparison. Journal of Semantics 3:1–77. von Stechow, A. 1996. The different readings of wieder ‘‘again’’: A structural account. Journal of Semantics 13:87–138. Wunderlich, D. 1997. Cause and the structure of verbs. Linguistic Inquiry 28:27–68. Wunderlich, D. 2000. Predicate composition and argument extension as general options. In B. Stiebels and D. Wunderlich, eds., Lexicon in Focus, pp. 247–270. Berlin: Akademie Verlag.

8

Only If an Indefinite is Topical, the Pronoun Picks it Up PAUL DEKKER

Abstract If and only are elegant and inspiring particles with substantial logical and linguistic roles. With this paper we hope to contribute to the understanding of the two, and to that of their interaction. Some of our (dynamic semantic) intuitions about the interpretation of these particles are tested on a small corpus of only if-donkey sentences found on the Internet.

Possibly the interpretation of the lexical items ‘‘if ’’ and ‘‘only’’ is the subject disputed most in the largest number of language-oriented disciplines, such as logic, natural language semantics, cognitive psychology, and artificial intelligence. In this paper I want to focus on the logical and linguistic aspects of these items. I do not aim to present a brand new theory about or new insights on the two subjects. My aim, rather, is to try and improve upon existing analyses and where possible help with integrating them. Some standard formal and logical analyses of ‘‘if ’’ and ‘‘only’’ are very attractive from a theoretical point of view, but their compatibility with more empirically oriented findings about the use of ‘‘if ’’ and ‘‘only’’ in natural language can be questioned. With this paper I want to support the view that no such incompatibilities exist, that we witness two ways Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

235

236

PAUL DEKKER

of approaching the same phenomenon, and that the findings in one perspective may contribute to those obtained in the other. As a starting point I take a system of dynamic semantics, actually my mother tongue, formally speaking. I will point out that a rigid dynamic semantic conception of meaning as such stands in the way of a more general, flexible, notion of interpretation, including that of if-clauses (Section on Dynamic Semantics (Second Generation)). I will next study a more empirically motivated interpretation of if-clauses as domain restrictors and argue that this alternative interpretation does not really pay off (Section on If-Clauses as Domain Restrictors?). In the Section on Inverse and Only If Donkey Sentences I sketch an analysis of ordinary, inverse, and only if-donkey sentences on the basis of (i) a classical notion of implication, (ii) a dynamic semantic notion of information, and (iii) an independently needed notion of information structure. In the Section Only if a Internet Search these more or less theoretical observations are tested on a number of only if-donkey sentences which I collected on the Internet. My basic intuitions seem to be supported by this little corpus, although they are not sufficient to account for all of the data. The Section on Causal and Other Dependencies gives a glimpse into the reasons why I think the basic interpretation of ‘‘if ’’ and ‘‘only’’ should not be taken to account for these additional data – mainly because these can be attributed to (the interplay with) other modal and pragmatic aspects of the conditionals at issue.

Dynamic Semantics (Second Generation) One of the major achievements of the dynamic semantic paradigm is its uniform semantic account of anaphoric relationships across conjunctions and implications. Let us focus in on a dynamic semantics like that of dynamic predicate logic (Groenendijk and Stokhof, 1991, DPL), by and large inspired by Discourse Representation Theory (Kamp, 1984; Kamp and Reyle, 1993, DRT) and file change semantics (Heim, 1982, FCS). The characteristic (distinctive) feature of DPL is that it renders (1), ‘Egli’s theorem’, semantically valid: (1) ð9xf ^ cðxÞÞ39xðf ^ cðxÞÞ By means of the supporting dynamic semantics, a formal and systematic analysis has thus been provided of a couple of sentences that have troubled semantic theory for ages. The equivalence in (1)

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

237

reflects the intuitive, truth-conditional, equivalence of the following two examples: (2) A man is walking in the park. He is whistling. (3) A man who is walking in the park is whistling. The core (dynamic semantic) idea behind the equivalence in (1) has always been that (i) meaning is some kind of update potential and (ii) indefinites ‘introduce’ some kind of discourse referents, which are available for being picked up by subsequent anaphoric pronouns. Quite an appealing feature of DPL (and of DRT and FCS, if you want) is that it renders (4) valid, as a mere consequence of (1): (4) ð9xf ! cðxÞÞ3:ð9xf ^ :cðxÞÞ3ð1Þ :9xðf ^ :cðxÞÞ38xðf ! cðxÞÞ Thus, DPL (like related systems) is able to deal, successfully, with Geach’s notorious donkey-sentence (5) which has standardly been taken to involve universal quantification over farmer-donkey pairs: (5) If a farmer owns a donkey (s)he beats it. Every farmer beats every donkey he or she owns. Example (5) has it that if you are presented with a (any) farmer who owns a (any) donkey, you will find that the first beats the latter. From a formal point of view DPL is very well behaved: its semantics is elegant and transparent, and it comes with a nice deduction system (Veltman, 2001; cf. also van Eijck, 2001). However, as a model for the interpretation of anaphoric relationships in natural language it is rather limited and, worse, too rigid to be suitably extended. The reason is that a notion of update is hardwired into the system’s notion of conjunction (and that of implication) and that anaphoric pronouns are associated with possible referents which cannot but have the properties attributed to their antecedents.1 This is problematic, as can be seen from Strawson’s famous example of pronominal contradiction: (6) A: A man just fell off the cliff. (7) B: He didn’t fall, he was pushed. In this example person B denies the referent to have the property previously ascribed to him by A. Besides, B even might continue with: (8) B: Besides, it was a woman. 1

This is also a characteristic feature of an E-type interpretation of pronouns.

238

PAUL DEKKER

In defense of an update (or E-type) approach one might claim that Strawson’s example is a special case, involving an intended referent which is (or, rather: was) demonstratively present. As argued elsewhere (e.g., in Dekker, 2002) we do not think demonstrative presence is a prerequisite for cross speaker anaphora such as we might witness here, and, besides, it may be fairly obvious that someone who overhears Strawson’s protagonists may very well understand what they claim without having access to the individual A and B intended to refer to (and which may even be the product of their imagination). Rigid update systems of interpretation are also deemed to fail in response to minor variants of (5) such as: (9) A farmer may beat a donkey if (s)he owns it. (10) Only if a farmer owns a donkey may (s)he beat it. In example (9) the clause ‘‘a farmer beats a donkey’’ is semantically (conditionally) dependent on the phrase ‘‘he owns it’’, whereas the latter is structurally (anaphorically) dependent on the first. In a onedimensional update system of interpretation this constitutes a paradox of interpretation. Essentially the same goes for example (10). The only if clause describes a necessary condition for a situation as described by the main clause to obtain. But what would such a situation be, according to the update analysis of pronouns? That (s)he, a farmer who owns a donkey, may beat it, the donkey (s)he owns? But if this is indeed the situation at issue, (10) would be vacuously true – which, as a matter of fact, the sentence is not.2 The need for a potentially more flexible treatment of anaphoric relationships has given rise to what may be termed ‘second generation’ systems of dynamic semantics. These systems crucially employ the notion of information developed in the first generation (Heimian information sets, say, the semantic correlate of discourse representation structures). The crucial difference with the first generation is that sentence meanings are not taken to be updates of these information sets, but information sets themselves. By means of a dynamic notion of conjunction, which is derived from a classical notion of conjunction as intersection, anaphoric relationships between indefinites and anaphoric pronouns can be established in a flexible way. The system underlying the remainder of this paper is such a system ‘predicate logic with dynamic conjunction’ (PLDC, 2

Notice that an E-type pronoun approach faces essentially the same problem.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

239

Dekker, 2002).3 This system is by and large inspired by Stalnaker (1978) and Stalnaker (1998) and close in basic spirit to that of Zeevat (1989), van der Does (1996) and van Rooy (1997). Passing over all technical details, Egli’s theorem (1) is also valid in PLA, as is the donkey equivalence (4). However, since the dynamics resides in a specific form of conjunction, and since it is not inherently encoded in the meanings of the sentences themselves, it is easy to come up with other forms of merging information sets. In particular, the information sets associated with Strawson’s examples (6) and (7) can be construed as being (intended to be) about one individual, without this requiring us to merge their contents. The nice thing about a system like PLDC then is, that, on the one hand, it is more flexible in easily allowing other forms of merging information, whereas, on the other hand, it comprises DPL as a special case. This indeed implies that the nice logical properties of DPL can be preserved, be it that they are qualified as logical properties in a restricted domain of application: that of monologue, or indeed of dialog without conflict and revision. Having given a very rough sketch of the basic machinery, let us turn back to ‘‘if ’’ and ‘‘only’’. In this paper we will assume basic interpretations of ‘‘if ’’ and ‘‘only’’ in the spirit of Philo, Frege, Quine, and Geach (but also Peirce and Grice) on the one hand, and a variety of authors like Horn, Rooth, de Mey, and Hendriks (among many many others) on the other. The basic idea is to associate ‘‘if ’’ with - (and, more generally, with D), and ‘‘only’’ with + (and, in specific cases, with ’).4 Simple motivating examples are the following: (11) Who qualify?  PostDocs do. (PDQ)  Only PostDocs do. (P+Q) (12) Will Francis get the job?  If he satisfies the prerequisites. (p-q)  Only if he satisfies the prerequisites. (p’q) 3 For the purposes of exposition it is the most transparent one to use here. When it actually comes to the study and motivation of the underlying semantics, we however prefer the system called ‘predicate logic with anaphora’ (PLA). For a technical introduction to that system, cf. Dekker (2002); for more philosophical motivation, cf. Dekker (2004). The central ideas and properties of the systems PLDC and PLA are by and large the same. 4 Notice that these four connectives are interdefinable, e.g., + ¼ (lPlQ QDP), ¼ (lplq (lx p) D (lx q)), D ¼ (lPlQ’x(Q(x)’P(x))), etc.

240

PAUL DEKKER

A more general interpretation is required in the following setting: (13) When does a student qualify?  If (s)he satisfies the prerequisites. (pDq)  Only if (s)he satisfies the prerequisites. (p+q) Before we take a further look at the inverse and only if donkey sentences (9) and (10), let us first consider an alternative idea about the meaning of if-clauses, which has gained some popularity in the literature.

If-Clauses as Domain Restrictors? Quite a few authors have observed that if-clauses also figure as ‘domain restrictors’ (Lewis, 1975; Heim, 1982; Kratzer, 1991; von Fintel, 1998), and it has even been argued that domain restriction is their pivotal role, cf. The history of the conditional is the story of a syntactic mistake. There is no two-place if . . . then connective in natural language. If-clauses are devices for restricting the domains of various operators. (Kratzer, 1991: 656)

Such a conclusion is by and large motivated by sentences with adverbial quantifiers, in which the if-clause indeed provides the restriction of an adverbial quantifier, and in which the main clause figures as its nuclear scope: (14) If a friend has problems Joe sometimes/usually/always/never helps him out. Intuitively these examples seem to express that Joe helps in {some/most/all/no} cases in which a friend has problems. It has also been argued that if-clauses may serve to impose further constraints on the restriction of adnominal quantifiers. Consider the following two examples due to Higginbotham (1986):5 (15) Every student will succeed if he works hard. (16) No student will succeed if he goofs off. Example (15) can be given an intuitively correct interpretation, if it is taken to state that for every student the following holds: if he works 5

Examples such as these are also discussed in Bosch (1983).

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

241

hard he will succeed. However, a similar analysis of example (16) would give rather disastrous results. Suppose example (16) is rendered as stating that for no student this holds: if he goofs off he will succeed. If we read the latter implication as a material one, then the sentence would turn out to state that every student goofs off and no student succeeds, which is way too strong. Alternatively, if the sentence is taken to state that for no student there is a rule-governed connection between goofing off and success, then this is way too weak. Rather, example (16) seems to state that no student who goofs off will succeed, i.e., that goofing off implies failure. Intuitively acceptable readings of both (15) and (16) can be obtained if the two if-clauses are interpreted as constraining the quantifier’s restriction. Example (15) says that every student who works hard succeeds and example (16) that no student who goofs off succeeds. Something essentially similar holds of two examples which Kai von Fintel has attributed to Irene Heim: (17) Few people like New York if they didn’t grow up there. (18) Most letters are answered if they are shorter than 5 pages. Intuitively, (17) states that few people who didn’t grow up in New York like New York, and example (18) can be taken to state that most letters which are shorter than five pages are answered. Finally, the following examples can be interpreted in a similar way: (19) Students qualify if they work hard. Example (19) intuitively says that hard working students qualify. The case for a domain restriction analysis of if thus appears to be rather strong. Indeed: Thus there are good reasons to adopt the restrictor analysis for noun-phrase indicatives and not just for sentences involving adverbial quantifiers. (von Fintel, 1998: 212)

The case is not that simple, though, as the following discussion may serve to show. First, consider some slight modifications of the students’ success stories: (20) Derek succeeds if he works hard. There seems to be nothing to restrict in (20), and certainly it does not say that among those who work hard, Derek succeeds. Rather, the

242

PAUL DEKKER

interpretation of (20) must involve some implication, that hard work will guarantee success for Derek. (21) Only Derek succeeds if he works hard. (22) Only students succeed if they work hard. A domain restriction reading of (21), if any, and (22) would entail that all who succeed work hard, Derek, or some of these students. Instead, on the wide scope only reading the sentences seem to state something different.6 The sentences then can be taken to state that if someone succeeds who has worked hard, it is Derek, or a student, respectively. The domain restriction analysis goes even more drastically astray when we consider examples with indefinites and numeral determiners. (23) Some students succeed if they work hard. (24) Exactly three students succeed if they work hard. Clearly, (23) and (24) do not serve to state that some/exactly three students who work hard succeed. These sentences instead say something about the number of students for which hard work is a guarantee for success. An implication analysis of these sentences thus seems to be most appropriate, provided that the domain of quantification is restricted to those students for which (to the speaker’s knowledge) working hard is still an option.7 To conclude this negative part, let us finally point out another rather severe complication for a domain restriction analysis. Consider: (25) Every student accepts any position if (s)he qualifies for it. (26) No postdoc disregards a job opportunity if (s)he qualifies for it. If if-clauses constrain the domains of quantifiers, then they must be taken to constrain two domains simultaneously in (25) and (26). This is not impossible, but it certainly transcends the capacities of the 6 If the if-clause of (21) is given wide scope, it says that if Derek works hard, then only he will succeed. So it is part of his unchummy ambitions then to become the one and only who succeeds. 7 As von Fintel (1998) observes, example (23) indeed corresponds to the examples figuring in what has been called ‘‘Peirce’s puzzle’’. But sure enough, Peirce’s puzzle is definitely not solved by a domain restriction analysis of if-clauses, but by a combined semantic pragmatic (Gricean) analysis of utterances of conditionals, completely similar to the one offered by Peirce himself. (See Dekker (2001) for discussion.)

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

243

envisaged techniques. Instead, it seems that for these examples, as for the examples (20)–(24), some implication analysis of a main-clause if-subordinate clause structure is appropriate. But then, again, what about (15)–(19)? Actually, an implication analysis of these sentences does not fare that bad at all. First observe that an analysis of if A as lq p-q is appropriate for (15) (and 25) and (19). For if any hard working student qualifies, then, and only then, it is true for any student that (s)he qualifies if (s)he works hard. Schematically: (27) ðS \ WH Þ  Q iff S  ðWH [ QÞ With respect to the examples (16), and (17) (and (22)) observe that the respective verb phrases figure in a downward entailing context. In Dekker (1993, Ch. 3, 1999) I have argued that external modification of material in a downward entailing context is subject to a dualization operator *, which guarantees preservation of monotonicity properties. Such a logical operator turns a conditionalizer lq p-q into a conjoiner lq pXq: (28) (lq p-q)* ¼ lq :(p-:q) ¼ lq pXq If used in the interpretation of example (16) the example will be read as stating that no student will goof off and succeed, (17) as stating that few people didn’t grow up in New York and like the city, and (22) as stating that only students both work hard and succeed. Surely, there are some modal and aspectual properties which are thus not accounted for (and which we will come back to below), but truth-conditionally these paraphrases are equivalent with those obtained on the domain restriction reading. This leaves us with example (18): (18) Most letters are answered if they are shorter than 5 pages. Also for this example an implicative reading is certainly not inappropriate. Example (18) can be taken to say that most letters are subject to the rule or principle that if a letter is shorter than 5 pages, then it gets answered.8 This reading is slightly different from the one in 8 As before, the suggested connection between main and subordinate clause seems to be stronger than the one given by a material implication analysis, but this is something which can be accounted for in pragmatic terms, as Grice (1989) has already argued for. An essentially similar account can be given of the fact that quantification in (18) seems to be restricted to a domain of letters which are still under consideration for being answered.

244

PAUL DEKKER

which the if-clause figures as a domain restrictor. For instance, suppose 5 letters are still under consideration, 3 are shorter than 5 pages and one of the three is answered. In that case 3 out of 5 letters (the answered letter and the 2 longer ones) obey the principle so the sentence would be true under the implicative reading. In the same case only 1 out of 3 letters which are shorter than 5 pages are answered; so the sentence would be false under the restrictive reading. I leave it to the general public to judge the issue. So far we have argued that indeed there is nothing obviously wrong with an implicative interpretation of if-clauses, and that a domain restrictive reading is not generally tenable. If-clauses do not restrict adnominal quantifiers. But what about the adverbial quantifiers, which made up the clearest motivation for domain restriction analyses? Here we have to point out that, if, indeed, if-clauses impose conditions on the domains of quantification of adverbial quantifiers, this does not mean they are not lq p-q, basically. Suppose p is an (intentionally closed) proposition, a condition on worlds, situations, assignments, sequences of individuals, or what have you. Suppose, moreover, we analyze an if A-clause as 4lq(3p-3q), abbreviated as R. The condition p then can be retrieved from R in the following way: (29) E(R) ¼ df

4

’q(3R(q)-3q)

With R as specified, this gives us 4’q((3p-3q)-3q)3 ’q( p33q). Indeed, the worlds (situations, . . . ) in which p is true in disjunction with any proposition, including a false one, is precisely the set of worlds (situations, . . . ) in which p is true.9 As this result is perfectly general, we think we can be satisfied here with not choosing between analyzing if-clauses as restrictive or as implicative.10 Notice, again, that, since we can retrieve p from lq p-q, the meaning of only if p (lq p’q) can be defined as a function on the meaning of if p. To conclude this section, let us sum up the results. We have argued that if-clauses do not generally restrict adnominal quantifiers, that they may restrict that of adverbial (or adsentential) quantifiers, 4

3

9 Notice that we could have done the same without the 4’s and 3’s, but the present formulation may be more insightful. 10 If if-clauses are analyzed as (incomplete) conditionals, there is indeed an arrow present in the interpretation of the full sentences, contrary to the conclusion of Kratzer (1991) quoted above. However, the point, entirely consistent with her conclusion, is that - doesn’t really act as a two-place operator then. For, pushing the point a bit, such sentences can be analyzed as O(p-)(q), with O the main (explicit or implicit) sentential operator. The whole issue, thus, trivializes.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

245

but that they can do so also on their lq p-q interpretation. Finally, we have observed some presuppositional restrictions upon domains of quantification, which will be taken more seriously in the next section.

Inverse and Only If Donkey Sentences Let us now turn back to inverse and only if-donkey sentences. In von Fintel (1994) it has already been observed that phonological structure (with associated information structure) is highly relevant for the interpretation of inverse and only if-donkey sentences. Consider: (30) A farmer may [beat]F a donkey if (s)he owns it. (31) Only if a farmer [owns]F a donkey may (s)he beat it. These examples seem to be fine. They appear to be about farmers and donkeys, and either the question when a farmer is allowed to beat a donkey, or the question what a farmer may do to a donkey he owns. Of all pairs of a farmer and a donkey (30) says that if it is an own-pair, it is also a may-beat-pair, and (31) that only if it is an own-pair it is a may-beat-pair. The farmers and donkeys, in other words, seem to have escaped the conditional structures, and to live in their domain of quantification. A proper interpretation of the pronouns thus seems to be possible precisely because they relate to an antecedent which, semantically, is not dependent on the clause which the pronoun figures in. The whole conditional structures, with the pronouns in them, are dependent on the domains of quantification evoked by ‘‘a farmer’’ and ‘‘a donkey’’. Such an analysis can be motivated further by examples where some of the indefinites figure in a focalized constituent: (32) ?A farmer may [beat a donkey]F if (s)he owns it. (33) ?Only if a farmer [owns a donkey]F may (s)he beat it. These sentences are problematic, semantically speaking. They appear to be about farmers and to address the question when such a farmer is allowed to beat a donkey. But, if this is so, then it is unclear what the pronoun ‘‘it’’ stands for. It is clear that the pronoun ‘‘he’’ in these examples, as in the previous ones, stands for a, or any, farmer. But where ‘‘it’’ in the previous examples clearly refers to a or any donkey, it seems to be unresolved here. ‘‘Only if a farmer [owns a donkey]F may he beat what?’’ seems to be a natural response to (33). Again, a reply with ‘‘a, or any, donkey he owns’’ would render the

246

PAUL DEKKER

example trivial. The examples are odd, we claim, because the donkeys are focalized, and they are therefore unable to escape the conditional structure. Putting it the other way around, indefinites may escape certain semantic structures, and figure as antecedents of subsequent anaphoric pronouns only if they are non-focal (non-novel, or ‘‘topical’’ as we will also say). This observation is not at all new by the way (cf. Gawron, 1996; Aloni et al., 1999; Krifka, 2001), but (30)–(33) provide striking examples of the phenomenon. Before we turn to a (sketch of an) analysis, consider the two replies in example (13) again. They could as well have been formulated as follows: (34) A student qualifies if (s)he satisfies the prerequisites. If a student satisfies the prerequisites (s)he qualifies. (35) A student qualifies only if (s)he satisfies the prerequisites. Only if a student satisfies the prerequisites (s)he qualifies. As a reply to the question in (13) it is clear that these statements are about the students and about the question whether or in which cases they qualify. Notice that it is immaterial whether the indefinite is in the main or in the subordinate clause in (34) and (35). Thus, the semantic contribution of these topical phrases indeed seems to escape from the structures in which they figure. Let us finally point at a parallel mechanism which can be seen to be at work in the following examples, the first attributed to Marco Bikker by Helen de Hoop, the second from Regine Eckardt: (36) Most boys were rejected because of their height. (37) Almost all tickets were sold at checker 4. (38) Only boys who were rejected because of their height may file a complaint. Example (36) can be taken to quantify over the boys who were rejected. Similarly, (37) can be understood as an assertion about the tickets which were sold, and (38) as an assertion about boys who were rejected (and who may file a complaint). As in the previous examples, non-focal material figuring at one place in the structures – in the verb phrase in (36)–(37) and in a restrictive relative clause in (38) – escapes from there and somehow projects itself into a domain of quantification.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

247

In the remainder of this section we will see how this behavior can be modeled in a straightforward way using techniques dealing with presupposition and information structure familiar from the literature. In line with most, if not all, theories of presupposition and information structure we assume that information conveyed by means of natural language can be divided into a presupposition or ground part, and an assertion or focal part. More specifically, we will assume some compositional method of building such structured pieces of information in the style of Karttunen and Peters (1979), using Heimian information sets to model first order information.11 In all of the examples which we have discussed in this section, presuppositional-like material restricts a domain of quantification. In principle, this type of domain restriction can be effectuated in two ways. Presuppositions may impose additional constraints on the explicitly specified restrictions of quantifiers, as on the domain restriction analysis of if-clauses, or as constraining the contextually given domain of quantification (cf. Westersta ˚hl, 1984).12 It seems the latter option has to be preferred, intuitively, but, arguably, also empirically. First observe that contextually restricted domains of quantification should be kept distinct from the first argument of a nonconservative term like only (examples like these are discussed by

11

What we here envisage is, arguably, the semantic correlate of the very well-behaved structural theory of van der Sandt (1989) and Geurts (1999). Indeed, using Heimian information sets, or discourse representation structures like Geurts and van der Sandt do, the so-called ‘binding problem’, which has been deemed so disastrous for Karttunen and Peters’ analysis, automatically disappears. As Karttunen and Peters (1979) themselves observed their analysis has a problem with the following example: (39) Someone managed to succeed George V on the throne of England. Upon Karttunen and Peters’ analysis the person who is presupposed to have had difficulties with succeeding George V can be different from the one who is asserted to have done so eventually. But as soon as one can render the semantics of anaphoric relationships in order, this problem dissolves completely. For instance, example (39) can be said to presuppose that someone has tried hard to succeed George V, and to assert that that person eventually did succeed him. 12 If QD (A) (B) is a quantified construction evaluated relative to a domain D, and if C is a presupposition of A or B, then the whole can, in principle, be interpreted in one of two ways: as QD(C-A) (B) or as QD-C (A)(B). Distinguishing cases involve (readings of) determiners which do not satisfy extension or conservativity.

248

PAUL DEKKER

Ja¨ger, 1996, and by Herman Hendriks, in many presentations): (40) Which Athenians are wise? (41) Only sophists are wise. (OA(S)(W)3(A-S)+(A-W)) As a reply to (40), (41) can be used to state, not that all wise men are Athenian sophists, but, rather, that only (Athenian) sophists are wise Athenians. Similarly, example (38) has a reading which entails nothing about girls or about boys who were not rejected. This reading can be obtained if the non-focal condition of being a boy who is rejected is projected, from the restriction of only into the domain of quantification. We take it then, that such presuppositions, thus, restrict the domain of quantification and do not constrain the restriction of the determiner. Let us now try and see how the Bikker-de Hoop-Eckardt type of examples can be handled. Presuppositional restriction of domains of quantification can be represented, roughly, as: (42) GQC ðAÞðhB; DiÞ3GQC ðhB; AiÞðDÞ3GQC \B ðAÞðDÞ Backgrounded material B in the nuclear scope or in the restriction of a quantifier GQ can simply be used to constrain the (possibly presupposed) domain of quantification C. For (37) and (38) this gives us the intended readings: (37) Almost all tickets were sold at checker 4. AAC ðTÞðhS; SAT4iÞ3AAC \S ðTÞðSAT4Þ (38) Only boys who were rejected because of their height may file a complaint. OC ðhB \ R; RBOTH iÞðMFC Þ3OC \B\R ðB \ RBOTH ÞðMFC Þ Example (37) is evaluated against a background domain of quantification consisting of the tickets which were sold (in the present context C). Example (38) relates to the boys (in the context) who were rejected, and it is taken to state that only those among them who were rejected because of their height may file a complaint. Essentially the same techniques can be used to model what is going on in inverse and only if-donkey sentences, now we have the means to deal with anaphoric relationships in a sophisticated way. In the following examples it is important to realize that context

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

249

sets, like information sets, may contain information about tuples of individuals: (30) A farmer may [beat]F a donkey if (s)he owns it. II C ðh9xFðxÞ ^ 9yDðyÞ; MBðxÞðyÞiÞðOðxÞðyÞÞ3 OðxÞðyÞÞÞ 8ðx; yÞC ððFðxÞ ^ DðyÞÞ ! ðMBðxÞðyÞ Example (30) is true iff for all pairs of farmers x and donkeys y in the context this holds: x may beat y if x owns y. (31) Only if a farmer [owns]F a donkey may (s)he beat it. OI C ðh9xFðxÞ ^ 9yDðyÞ; OðxÞðyÞiÞðMBðxÞðyÞÞ3 MBðxÞðyÞÞÞ 8ðx; yÞC ððFðxÞ ^ DðyÞÞ ! ðOðxÞðyÞ Example (31) is true iff for all of the same type of pairs: only if x owns y, x may beat y. Interestingly, the very same type of analysis seems to be fit for ordinary donkey-sentences: (5)

If a farmer [owns]F a donkey (s)he beats it. I C ðh9xFðxÞ ^ 9yDðyÞ; OðxÞðyÞiÞðBðxÞðyÞÞ3 BðxÞðyÞÞÞ 8ðx; yÞC ððFðxÞ ^ DðyÞÞ ! ðOðxÞðyÞ

Such a donkey sentence is rendered true iff for all of the same type of pairs above: if x owns y then x beats y. This is interesting, because it shows we get a strong reading of donkey sentences, not just from a dynamic semantic interpretation of -, but from a basically classical interpretation of - and independently motivated principles of interpreting information (anaphoric) structure. In short, all key examples motivating a dynamic semantics can be handled using a static semantics, an independently motivated notion of information structure, a flexible combinatorics, and, not unimportantly, a dynamic semantic notion of information.13 In this section we have sketched an analysis of inverse and only if-donkey sentences in terms of their information structure, and based on the independently motivated idea that topical or non-focal material projects into domains of quantification. The analysis generalizes to ordinary donkey sentences and allows us to do without a dynamic or DRT-style analysis of - and stick to a classical one. 13 We here have to add a note on so-called asymmetric quantification. If indefinites are non-topical, as in:

(43) If a man [had a dime]F, he threw it in the parking meter. our analysis does not apply. However, there are independent reasons to assume that the pronoun ‘‘it’’ must act like some kind of an E-type pronoun in these contexts (cf. van Rooy, 1997 for relevant discussion).

250

PAUL DEKKER

Only if a Internet Search Some semantic/linguistic discussions tend to get so sophisticated that they reach a realm of subtle expert examples and intuitions which do not raise any sensible reaction (support, rejection) from ordinary language users. For this reason I have tried to lift our armchair philosophy to the level of mouse-arm linguistics. I was seriously interested in the question if only if-donkey sentences are actually used, and, besides, I must admit, it was easy enough to look for sentences like this with an ordinary search engine. I was pretty much surprised to find an interesting lot of these examples, and in this section I will comment upon them from the perspectives sketched in the previous sections. I hope it needs no comments that my findings in this section are statistically not at all significant. I have not collected a sufficiently significant corpus of examples, and my evaluation of each of the examples is not tested upon judgments of others. I did my search with the search engine Altavista, 2000 October 10, 11.35, looking for þ‘‘Only if a’’. Altavista reported: (44) ‘‘1968 pages found’’, ‘‘word count: Only if a: 320’’ and presented me 200 hits. One-third of these were misfires, so I ended up with a corpus of 133 examples.14 Not all examples were really understandable as such, e.g.: (45) June 12, 2000 – What should Ollie do? 47% Yes, in every case 48% No, never 5% Only if a repeat offender [128] (1) Like I said, the humble question which I started out with was: 

are only if-donkey sentences actually used?

and the answer is unconditionally positive. In more than half of the examples an indefinite in the if-clause is anaphorically related to material in the main clause, directly (as in (46)) or more indirectly (as in (47)): (46) Only if a greenish beryl is found without either a yellow or blue tone will it be called a ‘‘green’’ beryl. [6] 14

An overview of the examples, in context, and with address, can be found at



http://www.wins.uva.nl/Bpdekker/SALT/

References to the examples in that file are given here in square brackets: ‘[nn]’.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

251

(47) Only if a project takes an extremely excessive amount of time and work will the fee be any more than that. [1] (2) In almost all of these examples the indefinite(s) can be seen to restrict the domain of quantification. Sometimes the indefinites are partly topical (as in (48)) or functional upon the things that actually are quantified over (as in (49)): (48) Only if a ‘‘plain vanilla’’ ISO-9660 CD were being burned would these messages be of any interest, and such a CD would be useless for Linux installation anyway. [47] (49) Only if a second smoke detector trip occurs within a short time, typically 60 seconds, or if the detector fails to reset, an alarm is activated. [34] Only seven examples are, arguably, one-case only if-conditionals, such as (50): (50) Only if a miracle takes place will the far reaching goals it set at the start of the decade be reached. [18] (3) Generally only if-clauses express necessary conditions for the condition in the main clause. There is only one obvious exception in our corpus: (51) Only if a trick contains no trump, it is won by the highest card of the suit led to it. [17] Strictly speaking example (51) is false. However, the context in which (51) occurs might speak in its favor.15 (4) According to the analysis sketched above an indefinite in an only if-clause must be non-focal (topical, non-novel) in order to allow an anaphoric element in the main clause to pick up its possible values. The indefinite must contribute to the domain of quantification. This prediction is by and large confirmed.16 15 There can be, and actually has been, discussion about the proper evaluation of example (51). I thank Ede Zimmermann for some pertinent comments. 16 Only in seven cases the indefinite does not itself restrict the domain of quantification, but in six of these cases the indefinite is partly topical or functional upon the material that is quantified over, as, e.g., in (52):

(52) Only if a land worthy of economic exploitation was found would it be colonized. [124] Only (53) seems to present a real counterexample: (53) Only if a leader sees you he’ll ask you if your in a server he’s in. [49] But then, indeed, it seems hard to make sense of this mysterious example. The context of (53) does not help.

252

PAUL DEKKER

(5) Upon our analysis it ought to be possible to reformulate only if A B-sentences systematically as B, only if A. Indeed this turns out to be generally possible. The possibility to swap the two clauses is only blocked, systematically, if the A-clause contains an indefinite and the B-clause a related pronoun or anaphoric definite. Anaphoric relationships can thus be seen to systematically require an ante-cedent, and not to be satisfied with a post-cedent. For, indeed, if we swap the only if A- and the B-clause in these sentences, and at the same time exchange antecedents and anaphors, the results turn out fine again. Try for yourself, for instance, with the following examples: (54) Only if a student changes residence to living at home or vice versa will the student need to contact the Financial Aid Office. [12] (55) Only if a child feels right can he think right. [27] (56) Only if a judge were actually biased should her decision be reversed. [80] The fact that indefinites (deemed topical) can be so easily moved from the subordinate clause to the main clause may serve as a further indication that their semantic contribution is not local, and that, as we claim, it is projected into the domain of quantification.17 (6) In line with observations in Geis (1973), McCawley (1981), and Horn (1996) (among others), a sentence Only if A B must be (up to) equivalent with a contraposed sentence If not A not B. The examples in our corpus support this thesis, but with one systematic qualification. Indefinites in the subordinate only if A-clause which are deemed topical naturally escape the negation in the if not A-clause. Try for yourself, with, e.g.: (58) Only if a student’s behavior is disruptive or coercive should it be prohibited. [21] (59) Only if a U.S. exporter sells to the Nigerian government through an agent is there a registration fee requirement. [66]

17 Swapping the two clauses of only if-sentences with indefinites that are partly topical requires some more subtle adjustments. Example (52), for instance, would have to be reformulated as:

(57) A land found would be colonized only if it was worthy of economic exploitation.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

253

Non-topical indefinites, however, do not do so: (60) Only if a common framework is created, can innovative tax concepts be successful. [93] (61) Only if a king’s son would agree to marry me as a frog could the spell be broken. [105] These observations thus give further support to the thesis that the semantic contribution of topical indefinites is not local.18 (7) Just for the record we mention that our corpus contains seven examples with a conjunctive if-clause such as (62), and four with a disjunctive one (like (49) above). (62) Only if a student has participated and turned in work for more than half of the class will an ‘‘I’’ be given. [53] All of these examples seem to behave as can be expected. That is, they all seem to validate the following Boolean scheme: (63) Only if A and/or Au B 3 If not A or/and not Au, not B But even though they are generally well behaved, it can be difficult to choose the right Boolean – as can be seen from the condisjunction: (64) Only if a couple posted bond, obtained a license, and or presented a certificate to the county official would their marriage be recorded. [22] Resuming our results so far we find that (i) only if-clauses indeed appear to state necessary conditions for possible states or events reported in the main clauses; (ii) topical (or non-novel) indefinites in these clauses project their semantic contribution into the relevant domains of quantification; and (iii) pronouns in these structures may be anaphoric upon antecedent indefinites in them if, and only if, these indefinites are topical. Up to this point (our interpretation of) the examples in our corpus support the views we have exposed above. 18 There are also some interesting interactions between modals in the main B-clause and the second negation in the contraposed reformulation. Unfortunately, there is no room to go into this issue here.

254

PAUL DEKKER

Causal and Other Dependencies Upon our basic analysis, ‘‘if ’’ and ‘‘only if ’’ are each other’s (logical) converses, and they report regularities at best, no dependencies, and they are therefore deemed to be each other’s converses. However, it has often been observed that conditionals are used to express dependencies between eventualities reported in main and subordinate clauses, and that this hampers a reformulation of only if A B into A if B, and vice versa. Indeed ‘‘if ’’ and ‘‘only if ’’ are not at all ideal converses in natural language. A stunning counterexample comes from McCawley (1974) and McCawley (1981): (65) If butter is heated, it melts. (66) *Butter is heated, only if it melts. Whereas (65) seems a perfectly natural thing to say, and to be true, probably, its converse (66) sounds very odd, if indeed understandable at all. The same holds for the majority of examples in our little corpus. Only if A B and A if B do not seem to express exactly the same states of affairs, except in only a few cases: (67) Only if a unit is flawless is it ready for shipping. [59] (68) Only if a file is available at both sites the mirror sites are listed. [91] (69) Only if a great saint occurs, do we permit [a relaxation] on his memorial. [129] (Note: the pronoun ‘‘his’’ in (69) is not anaphoric upon ‘‘a great saint’’.) There are a few points to be made here. First, the observations about McCawley’s (65)–(66) can be explained by a principle stated in Talmy (1978) and Reinhart (1984) according to which causal consequences of events reported in a main clause are not allowed to figure in subordinate clauses.19 A consequential interpretation of (65) is thus blocked in (66). Instead, (66) can be taken to suggest that, for

19 I am indebted to Antje RoXdeutscher for supplying me with the reference to Reinhart’s work. Talmy states a principle like this as a possible universal of natural language.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

255

butter, melting is a necessary condition for being heated, which, indeed, is an odd thing to state.20 Second, the Talmy/Reinhart principle only serves as an explanation of the non-convertibility of Only if A B into A if B if these sentences report a ‘deeper’ connection between the material described in the main and the subordinate clauses than just a regularity or correlation. This point can be strengthened with an example from von Fintel (1997). A reformulation of only if with if is possible if must is added (and will is dropped): (73) We will celebrate only if John wins. (74) ?If we will celebrate, John wins the race. (75) If we celebrate, then John must have won the race. According to (73) John’s success is a prerequisite for a celebration, and this dependency blocks a reformulation into (74) – which even suggests that a possible celebration would speed up John sufficiently. But such a consequential relation can be canceled by adding an epistemic must, indicating that a possible celebration (as referred to with (75)) constitutes sufficient evidence for conclusions about the outcomes of the race. Notice, however, that there is a subtle difference between (73) and (75). Whereas the first indeed seems to state a genuine dependency between possible events described by the main and the subordinate clause (a prerequisite, for instance), the second at best states a consequence from the dependency reported by the first, and not a dependency itself. A completely similar pattern can be observed with quite a few examples in our corpus, as you may try for yourself on: (76) Only if a youngster is playing along with senior cricketers, he will develop in confidence and do well. [114] 20 There are more constructions which may occur happily in a main clause, but not in subordinate clauses. Questions, demands, epistemic modals and probabilities can be conditionalized, whereas they are not allowed to figure in if-clauses. Consider:

(70) Do not invite the Smiths, if you don’t like gossip. *Only if do not invite the Smiths, you don’t like gossip. (71) If I agree, then what’s in it for me? *I agree, only if what’s in it for me? (72) If Mary is home, then maybe/probably Bill is around as well. *Mary is home, only if maybe/probably Bill is around (as well).

256

PAUL DEKKER

(77) Only if a system also talks his language, a user will be convinced by machine-found proofs and feel his understanding of the topic improved. [123] A major question at this point then is the following. If (only) if-sentences are used to report something over and above mere regularities, is this a uniform feature, and should it be taken to belong to the meanings of ‘‘(only) if ’’? Unfortunately, there is no room here to really go into this question, but we will not withhold from the reader the impression we have got from the examples in our corpus. These are, first, that there is a great variety of modals21 in the main clauses of our examples, and that these contribute to the expression of a great variety of dependencies. Many examples are concerned with regularities, necessities and possibilities of all kinds, personal, conventional, logical, moral, juridic, causal, . . . . Quite a few others are concerned with permission, obligation, advice and instruction, including that of the workings of machines and politicians. And although quite a lot of this variety can be attributed to the fact that conditionals may relate to a variety of ‘‘modal bases’’ (Kratzer, 1981), it does not seem that they use these modal bases in a uniform way. I am thus tempted to conclude, fully in the spirit of Grice (1989), that there must be a uniform, basic meaning of ‘‘if ’’ and ‘‘only’’, and that, as a matter of fact, this must be the weakest one, the one we have also employed in this paper. Additional aspects of meaning should be explained as the result of a combination of factors: on the one hand, the basic semantics of ‘‘only’’, ‘‘if ’’ and of modal and temporal operators in the main clause of a conditional, and, on the other, relevant pragmatic features, such as information structure, implicit questions under discussion, indexical features, besides general pragmatic principles and constraints like those of Talmy and Reinhart.

Conclusion This paper started out with the observation that inverse and only if-donkey sentences pose a problem for a classical dynamic semantic framework. We have sketched an account of these sentences in what we call a ‘‘second generation’’ framework, which crucially employs the 21 There are 28 examples which come without a modal or temporal operator, 34 with will, 20 with would, 17 with can, 3 with could, 9 with should, and 22 with modalities stated otherwise.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

257

notion of information developed in the first, but with a more classical and more flexible combinatorics. The main idea has been that donkey indefinites in inverse, only if and ordinary donkey sentences are topical and are projected into the domains of quantification of (im- or explicit) adverbial or adsentential quantifiers. The effects of this have been modeled on the basis of a classical analysis of ‘‘if ’’ and ‘‘only’’ and a Karttunen and Peters-style treatment of information structure. In passing we have argued that domain restriction effects of ifclauses are real, that these effects can, but need not, be attributed to the semantic contribution of if, and that if-clauses do not restrict nominal quantifiers (although their presuppositions may do so). We have next tested our findings on a corpus of only if-donkey sentences drawn from the Internet. Our intuitions about these examples indeed constitute motivation for the idea that topical indefinites do not semantically contribute to their local context, and that only these can be antecedents for anaphoric pronouns. We also find strong motivation for the idea that if- and only if-sentences report something over and above the statement of a mere regularity, but we have not found sufficient reason to attribute this additional aspect of meaning to the meaning of ‘‘if ’’ and ‘‘only’’. (Of course, not being able to find such a reason may be a shortcoming of the author.) Just to conclude, let me repeat that our little corpus can be inspected at: 

http://www.wins.uva.nl/Bpdekker/SALT/

There you can also find some of the minimal context of our final number of the show: (78) Only if a building is selected, and a name is picked randomly from a list of the persons in that building, can we use the fact that the name denotes a person on the 2nd floor to infer that the probability that it is a 100-persons house is less than 75%. (In fact, it’s 8%.) [13]

References Aloni, M., D. Beaver, and B. Clark. 1999. Focus and topic sensitive operators. In P. Dekker, ed., Proceedings of the 12th Amsterdam Colloquium, pp. 55– 60. Amsterdam: ILLC/Department of Philosophy.

258

PAUL DEKKER

Bosch, P. 1983. Agreement and Anaphora. London: Academic Press. Dekker, P. 1993. Transsentential Meditations. Ups and Downs in Dynamic Semantics. Ph.D. thesis, ILLC/Department of Philosophy, University of Amsterdam, Amsterdam. Dekker, P. 1999. Scopes in discourse. Language and Computation 1(1):7–32. Dekker, P. 2001. Dynamics and pragmatics of ‘‘Peirce’s Puzzle’’. Journal of Semantics 18(3):211–241. Dekker, P. 2002. Meaning and use of indefinite expressions. Journal of Logic, Language and Information 11:141–192. Dekker, P. 2004. Grounding dynamic semantics. In A. Bezuidenhout and M. Reimer, eds., Descriptions and Beyond, pp. 484–502. Oxford: Oxford University Press. Gawron, J. M. 1996. Quantification, quantificational domains and dynamic logic. In S. Lappin, ed., The Handbook of Contemporary Semantic Theory, pp. 247–267. Oxford: Blackwell. Geis, M. 1973. If and unless. In B. Kachru, R. Lees, Y. Malkiel, A. Petrangeli, and S. Saporta, eds., Issues in Linguistics: Papers in Honor of Henry and Rene´e Kahane, pp. 231–253. Urbana, IL: University of Illinois Press. Geurts, B. 1999. Presuppositions and Pronouns. Oxford: Elsevier. Grice, H. P. 1989. Indicative conditionals. In P. Grice, ed., Studies in the Way of Words. Cambridge, MA: Harvard University Press. Groenendijk, J. and M. Stokhof. 1991. Dynamic predicate logic. Linguistics and Philosophy 14(1):39–100. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. thesis, University of Massachusetts, Amherst. Published in 1988 by Garland, New York. Higginbotham, J. 1986. Linguistic theory and Davidson’s program in semantics. In E. Lepore, ed., Truth and Interpretation. Perspectives on the Philosophy of Donald Davidson, pp. 29–48. Oxford: Blackwell. Horn, L. R. 1996. Exclusive company: Only and the dynamics of vertical inference. Journal of Semantics 13(1):1–40. Ja¨ger, G. 1996. Only updates. On the dynamics of the focus particle only. In M. Stokhof and P. Dekker, eds., Proceedings of the Tenth Amsterdam Colloquium, pp. 387–405. Amsterdam: ILLC, University of Amsterdam. Kamp, H. 1984. A theory of truth and semantic representation. In J. Groenendijk, T. Janssen, and M. Stokhof, eds., Truth, Interpretation and Information, pp. 1–41. Dordrecht: Foris. Kamp, H. and U. Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer. Karttunen, L. and S. Peters. 1979. Conventional implicature. In C.-K. Oh and D. A. Dinneen, eds., Syntax and Semantics 11 – Presupposition, pp. 1–56. New York: Academic Press. Kratzer, A. 1981. The notional category of modality. In H.-J. Eikmeyer and H. Rieser, eds., Words, Worlds and Contexts, pp. 38–74. Berlin: de Gruyter.

ONLY IF

AN

INDEFINITE

IS

TOPICAL,

THE

PRONOUN PICKS

IT

UP

259

Kratzer, A. 1991. Conditionals. In A. von Stechow and D. Wunderlich, eds., Semantik: ein internationales Handbuch der zeitgeno ¨ssischen Forschung, pp. 651–656. Berlin: Walter de Gruyter. Krifka, M. 2001. Non-novel indefinites in adverbial quantification. In C. Condoravdi and G. Renardel de Lavalette, eds., Logical Perspectives on Language and Information. Stanford, CA: CSLI Publications. Lewis, D. 1975. Adverbs of quantification. In E. L. Keenan, ed., Formal Semantics, pp. 3–15. Cambridge, UK: Cambridge University Press. McCawley, J. D. 1974. If and only if. Linguistic Inquiry 5:632–635. McCawley, J. D. 1981. Everything that Linguists have Always Wanted to Know about Logic* *but were Ashamed to Ask. Oxford: Basil Blackwell. Reinhart, T. 1984. Principles of Gestalt perception in the temporal organization of narrative texts. Linguistics 22:779–809. Stalnaker, R. 1978. Assertion. In P. Cole, ed., Syntax and Semantics 9 – Pragmatics, pp. 315–332. New York: Academic Press. Stalnaker, R. 1998. On the representation of context. Journal of Logic, Language and Information 7:3–19. Talmy, L. 1978. Figure and ground in complex sentences. In J. H. Greenberg, C. A. Ferguson, and E. A. Moravcsik, eds., Universals of Human Language, pp. 625–649. Stanford, CA: Stanford University Press. van der Does, J. 1996. An E-type logic. In J. Seligman and D. Westersta ˚hl, eds., Logic, Language and Computation, pp. 555–570. Stanford, CA: CSLI. van der Sandt, R. A. 1989. Presupposition and discourse structure. In R. Bartsch, J. van Benthem, and P. van Emde Boas, eds., Semantics and Contextual Expression, pp. 267–294. Dordrecht: Foris. van Eijck, J. 2001. Incremental dynamics. Journal of Logic, Language and Information 10:319–351. van Rooy, R. 1997. Descriptive pronouns in dynamic semantics. In P. Dekker, M. Stokhof, and Y. Venema, eds., Proceedings of the Eleventh Amsterdam Colloquium, pp. 265–270. Amsterdam: ILLC, University of Amsterdam. Veltman, F. 2001. How Natural is Natural Deduction for DPL? Technical report, ILLC, Amsterdam. von Fintel, K. 1994. Restrictions on Quantifier Domains. Ph.D. thesis, University of Massachusetts, Amherst. von Fintel, K. 1997. Bare plurals, bare conditionals, and only. Journal of Semantics 14:1–56. von Fintel, K. 1998. Quantifiers and if-clauses. The Philosophical Quarterly 48:209–214. Westersta ˚hl, D. 1984. Determiners and context sets. In J. van Benthem and A. ter Meulen, eds., Generalized Quantifiers in Natural Language, pp. 45–71. Dordrecht: Foris. Zeevat, H. 1989. A compositional approach to discourse representation theory. Linguistics and Philosophy 12:95–131.

9

Good and Bad Transitive Arguments MICHAEL MORREAU

Abstract Some treatments of counterfactual conditionals validate transitive arguments. They must explain why some of these arguments have no force. I shall argue that they do not succeed in this. Other treatments make transitive arguments invalid and face the complementary problem of explaining why some of these arguments are quite compelling. I shall solve this problem by distinguishing between contextually good and bad arguments. Good arguments are those that are completed by contextually acceptable assumptions that exclude far-fetched possibilities. Bad arguments are those whose completers are contextually unacceptable.

Transitive Arguments Consider arguments that string together counterfactual conditionals. I am explaining the workings of a car. I say first: (1) If I were to turn the ignition key, I would engage the starter motor. Then I say: (2) If I were to engage the starter motor, the engine would start. You may infer: (3) If I were to turn the ignition key, the engine would start. Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

261

MICHAEL MORREAU

262

Indeed, this conclusion really does seem pretty inescapable. Letting W be a counterfactual conditional connective, conjoining say ‘‘I turn the ignition key’’ and ‘‘I engage the starter motor’’ to make (1), this inference is an instance of Transitivity: (4)

A4B; B4C A4C

How best to account for the force of transitive arguments? It has long been recognized that counterfactual conditionals are not material conditionals. Counterfactuals are properly, indeed typically used to make nontrivial claims when their antecedents are false; but the corresponding material conditionals are then trivially true (see Goodman, 1947). Some have thought that counterfactuals might be strict conditionals, though: material conditionals prefixed by operators expressing necessity. Certainly this will account for the force of arguments such as that from (1) and (2) to (3). Transitive arguments are valid if counterfactuals are strict conditionals. The main problem with this idea is that there appear to be counterexamples to transitivity. Suppose someone asserts: (5) If J. Edgar Hoover had been a Communist, he would have been a traitor to the U.S.A. Now, plausibly: (6) If Hoover had been born a Russian, he would have been a Communist. (6) expresses a matter that we believe or can easily accept for the sake of the argument. But it does not appear that we may string together these two sentences to infer: (7) If J. Edgar Hoover had been born a Russian, he would have been a traitor to the U.S.A. Presumably (7) is false. Some say transitivity is valid and argue that apparent counterexamples are pragmatically unsound. According to Jonathan Lowe, for instance, an argument like the one from (5) and (6) to (7) is something like a fallacy of equivocation. He argues that while there are conversational contexts in which (5) is acceptable, and other contexts in which (6) is acceptable, there are no contexts in which both premises are acceptable (Lowe, 1990).

GOOD

AND

BAD TRANSITIVE ARGUMENTS

263

In order to evaluate a proposal such as Lowe’s it is necessary to know more about contexts. He has not said very much about them but others have said more. David Lewis once entertained the idea that counterfactuals are strict conditionals whose strictness is fixed by the very local context: the conditional antecedent. Lewis thought this idea defeatist, insofar as it ‘‘consigns to the wastebasket of contextually resolved vagueness something much more amenable to systematic analysis than most of the rest of the mess in that wastebasket’’ (see Lewis, 1973: 13). But let us take the idea a little further. In line with Lewis’s ideas about counterfactuals, let us say that asserting a counterfactual has the effect of setting a contextual parameter, the worlds-accessibility function used to evaluate strict conditionals. The assertion of AWC brings it about that a sphere of possible worlds becomes accessible to the world of evaluation, with this world of evaluation at its center. This sphere includes the A-worlds that are most similar to the world of evaluation, along with all worlds that are as similar to the world of evaluation as they are, or more similar to it than they are.1 Finally, let us say that a counterfactual AWC is true, in the world of evaluation, just in case the material conditional A*C holds true at all worlds within this sphere. Interpreting counterfactuals in this way, the argument from (5) and (6) to (7) is indeed a sort of fallacy of equivocation. Asserting (5) sets the accessibility function one way; asserting (6), with its different conditional antecedent, sets it a different way. We do not evaluate the two premises with respect to one and the same worlds-accessibility function. This, I suggest, illustrates Lowe’s claim that this argument equivocates. This proposal will not satisfy anyone who thinks that counterfactuals are strict conditionals, though. It makes just about every argument involving counterfactuals equivocate, as successive counterfactual assertions with different conditional antecedents set the accessibility function first this way then that. And so the question arises whether it is possible to give a more sophisticated treatment that allows the accessibility function to evolve more stably as arguments unfold. Such a treatment might draw a better line separating pragmatically sound and unsound transitive arguments. For it might allow the contextually determinate accessibility function to remain 1 By ‘‘A-worlds’’ I mean worlds in which A holds true. I assume throughout that for any possible world there are always some A-worlds that are most similar to it, in the sense of Lewis’s notion of comparative similarity. This assumption simplifies the presentation. Nothing really depends on it.

264

MICHAEL MORREAU

unchanged in the course of compelling arguments, such as the one from (1) and (2) to (3), while requiring it to change in the case of intuitively unsound arguments, such as the one from (5) and (6) to (7). Kai von Fintel (2001) has suggested a treatment of counterfactuals along these lines. He nicely calls the possibilities that are accessible in any given context the modal horizon. These are the possible worlds that, in this context, are relevant when evaluating modal claims for truth. What is distinctive in his account is that, when evaluating a counterfactual assertion, in addition to the local context, also the global context helps to determine the modal horizon. Which possibilities are relevant depends not only on the counterfactual antecedent, but also on the antecedents of counterfactuals asserted earlier on. To see how this proposal sanctions compelling transitive arguments, it will help to have a sketch of the technical development. A modal horizon, in von Fintel’s treatment, is a worlds-selection function that maps possible worlds onto sets of possible worlds. Asserting a counterfactual AWC, in any given context with modal horizon f, tends to effect a transition to a new context with a wider modal horizon. This wider horizon f|AWC | assigns to any given world w not only the possible worlds that f assigns to w, but in addition some worlds needed to evaluate AWC. Specifically, it adds those worlds that are as similar to w as the most-similar A-worlds. That is, it adds those A-worlds that are most similar to w, along with all other worlds that are as similar to w as they are, or more similar to w than they are. Evaluation proceeds with the wider modal horizon: AWC is true at w, relative to f, if and only if the material conditional A*C is true at each world within f|AWC |.2 If AWC is true in w relative to f we write w f AWC. The innovation of a contextually determinate modal horizon can explain some contextual phenomena that might without it appear puzzling. For one thing, there is the sensitivity of the argument from (5) and (6) to (7) to the order in which the premises are asserted. Notice that even if we initially accept a speaker’s assertion of (5), we may balk at a reiteration of (5) after he has asserted (6); if he asserts (6) first, we may find it impossible to accept his later assertion of (5). It would 2 Here I put aside the more complicated case of nested counterfactuals, considering here only conditionals whose antecedents and consequents are sentences whose interpretation is independent of the modal horizon. Von Fintel suggests a more general truth condition that covers nested cases as well but the complication is not relevant for our present purpose.

GOOD

AND

BAD TRANSITIVE ARGUMENTS

265

appear that the assertion of (6) changes the conversational context in some way that makes (5) unacceptable. Plausibly, the change is in our modal horizon. With the assertion of (6), this widens to include the possibility that Hoover was born a Russian. But to take this possibility into account is just to take into account the possibility that Hoover was a Communist but not a traitor to the U.S.A. Interpreted with the wider modal horizon, (5) is false. To make a logical argument, von Fintel claims (see von Fintel, 2001: 143), is to give an implicit promise that the context will not change in the course of the argument.3 There is a sense in which any given premises can be said to entail a given conclusion just in case truth invariably passes from these premises to this conclusion, whenever the modal horizon remains constant as the argument unfolds: (8) P1, . . . , Pn C if and only if for each w and for each f such that f ¼ f jP 1 j...jP n jjC j : if w f P1, . . . , and w f Pn, then w f C. This notion of entailment – ‘‘Strawson entailment’’ – can be used to explain the force of some transitive arguments. Since transitivity is Strawson valid, von Fintel’s proposal suggests a criterion under which transitive arguments ought to be compelling: a transitive argument, made in a context with modal horizon f, ought to be compelling if: (9) a. b.

It can be assumed that f remains constant throughout the argument; and: The premises are true, relative to f.

Take for instance the initial argument from (1) and (2) to (3). The possibilities relevant to evaluating these sentences are just that I might turn the ignition key and that I might engage the starter motor. Possible worlds in which I do these things can be assumed to be within our modal horizon in the context within which we considered this argument (as they can in just about any ordinary context in which a car is under discussion). Therefore the modal horizon can be assumed to remain constant throughout the inference. Furthermore, we can suppose that, this being an ordinary context in which the car is presumed to function normally, the possibility that I might turn the ignition key without engaging the starter motor, and the possibility that I might engage the starter motor without starting the engine, are 3

See von Fintel (2001: 143).

266

MICHAEL MORREAU

both beyond our modal horizon. The premises, in this context, are true. Accordingly, the argument ought to be compelling. And it is. Take on the other hand the argument from (5) and (6) to (7). Arguably, there is no modal horizon f such that both: (10) a. b.

It can be assumed that f remains constant when (5) and (6) are asserted; and: (5) and (6) are both true, relative to f.

Suppose f remains constant when (6) is asserted. Then f assigns to the actual world a set of worlds, among them some in which Hoover was born a Russian. Now suppose (6) is true, relative to f. Then in these worlds Hoover is a Communist. Furthermore, we can assume, he is in some of these worlds – plausibly, in all of them – not a traitor to the U.S.A.4 Therefore, by the truth condition, (5) is false, relative to f. The upshot is that the argument from (5) and (6) to (7) is pragmatically unsound. We simply cannot assume that the modal horizon is to remain constant throughout this argument. Not if we are to accept that the premises are true. For all its plausibility, though, von Fintel’s proposal seems to overshoot its mark. Not only does it sanction compelling transitive arguments. It also seems to sanction some that, intuitively speaking, have no force at all. In the following section, I shall give an example.

A Counterexample I am explaining why I might start the car without using the ignition key, by ‘‘hotwiring’’ the starter motor directly to the battery. I say: Ordinarily, if I were to turn the ignition key, the engine would start; but this is now not the case because there is something wrong with the engine. The problem is not in the starter-motor circuit. If I were to turn the key, I would engage the starter motor, as usual. The problem is elsewhere. Turning the ignition key would short circuit the fuel pump and prevent the engine from starting. For this reason, if I were to engage the starter motor then that would be by ‘‘hotwiring’’ it, to spare the fuel pump. So if I were to engage the starter motor, the engine would start. 4

I assume that relative to the initial context: If J. Edgar Hoover had been born a Russian, he might not have been a traitor to the U.S.A. is actually true.

GOOD

AND

BAD TRANSITIVE ARGUMENTS

267

When discussing this car – and certainly by the end of this speech – the possibility that I will turn the ignition key can be assumed to lie within our modal horizon. So can the possibility that, one way or another, I will engage the starter motor. All possibilities needed to evaluate (1), (2), and (3) can be assumed to lie within our modal horizon. Furthermore, when evaluated in this context, the premises (1) and (2) are intuitively true. This argument, taken in this context, satisfies the earlier criterion under which transitive arguments ought to be compelling. But this argument would not be at all compelling if made in this context, in the light of the explanation why the conclusion (3) is false. There seems to be a difficulty for von Fintel’s proposal. In the remainder, I shall not try to locate the difficulty. Instead, I shall propose a different way of drawing the line between intuitively compelling and uncompelling transitive arguments. My proposal does not rest on the idea that counterfactuals are strict conditionals, of contextually determinate strictness. It rests on the standard possibleworlds treatment of counterfactuals as developed by Stalnaker (1968) and Lewis (1973). In the following section, I briefly set out this treatment.

Contextually Good and Bad Arguments Standard treatments of counterfactuals are based on the idea that a sentence AWC is true, at any given possible world of evaluation, just in case all A-worlds that are most similar to this world of evaluation are C-worlds. Transitivity is invalid in standard treatments. Counterexamples arise when one hypothesis, expressed, say, by a sentence A, is more ‘‘far-fetched’’ than another, expressed by B. Then it can happen that while some third sentence C holds true at all of the most-similar B-worlds, C does not hold true at all of the most-similar A-worlds, although B does hold true in all of the latter. Standard treatments, then, do not face the problem of saying what is bad about intuitively uncompelling transitive arguments. Rather, they face the complementary problem of explaining what is good about intuitively compelling transitive arguments. I shall propose a solution to this second problem. As I see it, a compelling transitive argument derives its force from a tacit assumption. But this assumption is not that the context will remain constant as the argument unfolds, as von Fintel thinks. Rather, it is an assumption that certain apparently unrelated things will not interfere with one another. In the case of the initial argument from (1) and (2) to (3), for

268

MICHAEL MORREAU

instance, the assumption is that turning the ignition key will not interfere with the causal connection between my engaging the starter motor and my starting the engine. I shall propose that it is because we tend to make such assumptions of independence that we tend to find transitive arguments compelling. Sometimes prior beliefs keep us from making these assumptions. For instance, a prior belief that turning the ignition key will cause the fuel pump to fail will keep us from assuming that turning the ignition key will not interfere with the relevant causal connection. Then transitive arguments have no force. Let us consider assumptions of independence in more detail. A failure of the fuel pump is something that would keep the engine from running. But there are many more things that would not keep it from running. For instance, presumably the stare of a black cat would not. Let R be ‘‘The engine runs,’’ and let B be ‘‘A black cat is staring.’’ Then we can render: (11) The stare of a black cat would not keep the engine from running. as: (12) RW(BWR). Notice that (12) says neither whether a black cat is staring, nor whether the engine runs, since, in a standard treatment, the truth of (12) is compatible with the truth of R and B, and also compatible with their falsity. This is just what is needed, if (12) is to be true to (11). Intuitively, (11) also says neither thing. Sentence (12) is a symbolization of: (13) If the engine were to run, then (even) if a black cat were staring it would run. There might seem to be a difficulty here for the idea that (12) is also a symbolization of (11). Typically (13) will be used with some sort of understanding that the engine does not run and that there is no black cat staring, but (11) will be used without either of these presuppositions; how, then, can (11) and (13) amount to the same thing, (12)? In fact there is no real difficulty. The truth conditions of counterfactuals do not account for their presuppositions, in a standard treatment. A pragmatic explanation can be given instead.5 Accordingly, having adopted a standard treatment, it is natural to say that the relevant 5

As Lewis (1973) points out on p. 1.

GOOD

AND

BAD TRANSITIVE ARGUMENTS

269

difference between (11) and (13) has nothing to do with their truth conditions, which are the same as the truth conditions of (12). It just has to do with how we use these sentences. Presumably the stare of a black cat will not keep the engine from running, but we can imagine how it might do anyway. There might be some complicated unsuspected chain of natural causes and effects linking stares of cats to failures of fuel pumps.6 There might be something even more surprising, say an unknown occult connection. All sorts of things just might be bound up with each other in all sorts of unsuspected ways. We can imagine these possibilities but ordinarily we do not reckon with them. We can just assume that the stares of a black cat will not keep the engine from running. An assumption that one event or state of affairs will not keep another from occurring or obtaining we shall call an assumption of independence. Not all assumptions of independence are acceptable: we cannot accept, say, the assumption that a failure of the fuel pump would not keep the engine from running, because we know very well that it would have just this effect. But we are disposed to make assumptions of independence, and mostly, since by and large they are compatible with prior beliefs and other commitments, they are acceptable. We can without giving the matter any special thought just assume that having a beer will not keep the engine from running, that putting off a vacation will not, and so on. The disposition to assume independence is bound up with ordinary thinking. Now I shall argue that it is this disposition, held in check by contextually determinate beliefs and assumptions, which makes some transitive arguments compelling, but not others. The salient feature of contexts, for our purpose, is that they determine prior beliefs and assumptions. For now we can identify contexts with sets of sentences that express such beliefs and assumptions.7 To add an assumption to a context is to form the obvious set-theoretic union. Simplifying further, we shall say that an assumption of independence is acceptable, in any given context, if the result of adding this assumption to this context is satisfiable.8

6 To see how these things might be linked up, visit http://www.rube-goldberg.com, or see Marzio (1973). 7 In doing so I set aside as tangential to our topic all contextuality in the relation between sentences and what they express. 8 Some sentences are satisfiable if there is some world at which they are true. Assumptions can be unacceptable though satisfiable together with prior beliefs and assumptions. They can be too complicated, too far-fetched, or wanting in some other way. Here I ignore these other failings that assumptions can have.

270

MICHAEL MORREAU

Here is a procedure for evaluating any given argument in any given context. First, add the premises of the argument to the context, as assumptions made for the sake of the argument. Second, add to the resulting context such assumptions of independence as are acceptable in it. Finally, see whether the resulting context entails the conclusion of the argument. If it does, the argument is good, in the initial context. Otherwise it is bad. The inference from (1) and (2) to (3) is good, when evaluated in the context of ordinary beliefs about car engines. Let SM be ‘‘I will engage the starter motor,’’ let S be ‘‘The engine starts,’’ and let I be ‘‘I will turn the ignition key.’’ Following the procedure, take the context of evaluation and add IWSM and SMWS, symbolizations of premises (1) and (2). Sentence (12) is one acceptable assumption of independence, in the resulting context. Here, presumably, is another: (14) (SMWS)W(IW(SMWS)) This sentence says that my turning the ignition key will not keep it from being the case that, if I were to engage the starter motor, the engine would start. That is, my turning the key would not break the causal connection between engaging the starter motor and the engine’s starting. Now (14) completes the inference from (1) and (2) to (3). This is because in any standard treatment the scheme: (15)

A4B; B4C ; ðB4C Þ4ðA4ðB4C ÞÞ A4C

is valid.9 We have seen that the argument from premises (1) and (2) to conclusion (3) is good, in the context of common knowledge. I shall now argue that, in the special context of evaluation that I described in the Section A Counterexample, this argument is bad. Not only is (14) not 9

Some sentences complete an argument if the argument that results when they are added to its premises is valid. To see that this scheme is valid, consider any world w where all three premises are true. I shall show that also the conclusion is true in w. Notice first that w is itself a world that, among BWC-worlds, is most similar to w. This is because the second premise is true in w. Therefore, since the third premise is true in w, by any standard truth condition AW(BWC) is true in w. Now take any A-world n that, among A-worlds, is most similar to w. I shall show that C is true in n. By any standard truth condition, BWC is true in n. Since the first premise AWB is true in w, also B is true in n, so n itself is a B-world that, among B-worlds, is most similar to n. So, by the standard truth condition, C is true in n. The same holds for any such world n. So AWC is true in w. This completes the demonstration that the above scheme is valid.

GOOD

AND

BAD TRANSITIVE ARGUMENTS

271

an acceptable assumption of independence in the relevant context. In addition, as we shall see, there is no other acceptable assumption that would complete this argument. First I shall show that (14) is not an acceptable assumption. That is, I show that it is not satisfiable together with the context of evaluation and the two premises (1) and (2). The context, recall, was one in which I have explained that if I were to turn the ignition key, the fuel pump would fail. We can express this matter: IWF. I have also explained that a failure of the fuel pump will keep the engine from starting no matter what. We can express this using a strict conditional: nec.(F * not S). Here the operator nec. asserts the truth of its scope in all worlds relevant to the evaluation of the other sentences in this example. By the logic of counterfactuals in a standard treatment, these two sentences entail that if I were to turn the ignition key then the engine would not start: IWnot S. So the context of evaluation entails this sentence. Following the evaluation procedure, we now add both premises to the context of evaluation. Among them is (1): IWSM. The result, in any standard logic of counterfactuals, entails: (16) not (IW(SMWS)) (I assume that the context of evaluation entails: not (IW(SMW falsum)). This sentence says that the hypothesis that I might turn the ignition key and engage the starter motor is not completely absurd.) The other premise that we have added to the context of evaluation is (2): SMWS. Now we see that (14) is not satisfiable together with the context of evaluation and the premises. For, in a standard treatment, (2) and (14) together entail: (IW(SMWS)), the negation of (16). What this shows is that (14) is not a contextually acceptable assumption of independence. This sentence will not be added to the context of evaluation in the second step of the procedure for evaluating arguments as good or bad. It is in this context not available to complete the argument from (1) and (2) to (3). Might this argument still be good anyway, being completed by some other assumption that, unlike (14), is acceptable? There is no such assumption. In a standard treatment, (1) and (3) jointly entail (IW(SMWS)). So any sentence that completes the argument from (1) and (2) to (3) is inconsistent with (16). Therefore, no assumption that completes this argument is contextually acceptable. This argument is bad, taken in the special context described in the Section A Counterexample.

272

MICHAEL MORREAU

An almost completely analogous argument shows that the argument from (5) and (6) to (7) is bad, taken in the context of common knowledge. It is common knowledge that in Russia, in Hoover’s time, being a Communist did not make you a traitor; if anything, it made you a patriot. That is, it is common knowledge that it is not the case that had J. Edgar Hoover been born a Russian, he would have been a traitor had he been a Communist. Letting R be ‘‘J. Edgar Hoover was born a Russian,’’ letting C be ‘‘He was a Communist,’’ and letting T be ‘‘He was a traitor,’’ we render this item of common knowledge: (17) not (RW(CWT )) Now the argument that no acceptable assumption completes this argument proceeds as above, with (17) instead of (16). In conclusion, I have argued that it is our tendency to make assumptions of independence, held in check by prior beliefs and assumptions made for the sake of the argument, that makes some transitive arguments compelling. In contexts where the relevant assumptions are contextually acceptable we make these assumptions and infer the conclusions of transitive arguments. In other contexts, where these assumptions conflict with prior beliefs and assumptions, we do neither.

References Goodman, N. 1947. The problem of counterfactual conditionals. The Journal of Philosophy 44:113–128. Lewis, D. 1973. Counterfactuals. Cambridge, MA: Harvard University Press. Lowe, E. J. 1990. Conditionals, context and transitivity. Analysis 50:80–87. Marzio, P. C. 1973. Rube Goldberg, his Life and Work. New York: Harper and Rowe. Stalnaker, R. 1968. A theory of conditionals. In N. Resher, ed., Studies in Logical Theory. Oxford: Blackwell. von Fintel, K. 2001. Counterfactuals in a dynamic context. In M. Kenstowicz, ed., Ken Hale: A Life in Language. Cambridge, MA: MIT Press.

10

Privative Adjectives: Subsective Plus Coercion BARBARA H. PARTEE

Introduction Montague (1970a) presented a semantic treatment of adjectives which he credited to unpublished work done independently by Hans Kamp and Terence Parsons; that work, and similar independent work of Romane Clark, was subsequently published (Clark, 1970; Parsons, 1970; Kamp, 1975). The central claim in that work was that adjective meanings should be analyzed as functions from properties to properties. Among adjective meanings, some might satisfy further constraints such as intersectivity or subsectivity, but no such constraint can be imposed on the class as a whole, the argument goes, because of the existence of adjectives like false, ostensible, alleged. The strategy of ‘‘generalizing to the worst case’’, followed by Montague in order to have a uniform assignment of semantic types to syntactic categories, called for giving all adjectives the type of functions from properties to properties. More restricted subclasses of adjectives, such as the subsective (skillful, good) and intersective (purple, carnivorous) adjectives, might be indicated by the use of meaning postulates. In theories which allow type multiplicity and type-shifting, the intersective adjectives might indeed be assigned the simpler type of one-place predicates; this is now widely assumed.

Presuppositions and Discourse: Essays offered to Hans Kamp. Rainer Ba ¨uerle, Uwe Reyle and Thomas Ede Zimmermann (eds.). Current Research in the Semantics/Pragmatics Interface, Vol. 21. r 2010 by Emerald Group Publishing Limited. All rights reserved.

273

274

BARBARA H. PARTEE

Kamp and Partee (1995) review the more or less standard ‘‘hierarchy’’ of classes of adjectives as a preliminary to arguing that arguments concerning the appropriateness of prototype theory as a part of the account of the semantics of adjective-noun combinations should be restricted to intersective adjectives. The hierarchy ranges from intersective adjectives like carnivorous to privative adjectives like counterfeit, fake, and fictitious. The same article makes some proposals for coercion of adjective meanings in context, driven by certain general constraints, which help to explain a number of kinds of shifts and adjustments that take place when adjective-noun combinations are interpreted in various kinds of contexts. Some problem cases remained, especially the case of stone lion, where it seems that the noun rather than the adjective shifts its meaning when faced with incompatibility of the primary senses of each word. The aim of the present note is to argue that in fact adjective meanings are more constrained than was appreciated either at the time of the work of Montague, Kamp, Parsons, and Clark or at the time of the work of Kamp and Partee. In particular, I will argue that some facts about the possibility of ‘‘NP-splitting’’ in Polish cast serious doubt on the standard hierarchy, and that the data become much more orderly if privative adjectives like counterfeit, fake, and fictitious are reanalyzed as subsective adjectives. Further evidence for that move comes from long-standing puzzles about what to say about sentences like Is that gun real or fake? The revised account requires the possibility of coerced expansion of the denotation of the noun to which such an adjective (as well as adjectives like real, genuine, which were not examined in the earlier-cited literature) is applied. Such coercion can be motivated by treating the constraints on possible adjective meanings as presuppositions that must be satisfied by any use of an adjective; the corresponding coercion may then be seen as a form of presupposition accommodation. The present note is structured as follows. The Section on Adjective Classification briefly reviews the adjective classification familiar since the work of the 1970s as summarized in Kamp and Partee (1995) and in Partee (1995). The Polish NP-splitting data (Nowak, 2000) and the problem they pose for the familiar hierarchy are presented in the Section on Privative Adjectives and Polish NP-Split Phenomena. The Section on Principles of Interpretation and the ‘‘No Privative Adjectives’’ Hypothesis reviews some of the constraints on possible adjective meanings proposed in Kamp and Partee (1995) and proposes further constraints that exclude privative adjectives and

PRIVATIVE ADJECTIVES: SUBSECTIVE PLUS COERCION

275

account for the coercion of the noun meaning in cases that would otherwise come out as privative.

Adjective Classification An adjective like carnivorous is intersective, in that (1) holds for any N. (1) ||carnivorous N|| ¼ ||carnivorous|| \ ||N|| But skillful is not, as shown by the invalid inference pattern in (2), familiar from the work of Kamp, Parsons, Clark, and Montague cited in the introduction. (2) Francis is a skillful surgeon. Francis is a violinist. Therefore Francis is a skillful violinist. [Not valid] Skillful is not intersective, but it is subsective: (3) holds for any N. (3) Subsectivity: ||skillful N|| D ||N|| The adjectives former, alleged, counterfeit are neither intersective nor subsective. (4) (a) (b)

||former senator|| 6¼ ||former|| \ ||senator|| ||former senator|| J ||senator||

Nonsubsective adjectives may either be ‘‘plain’’ nonsubsective (no entailments at all, no meaning postulate needed), or privative, entailing the negation of the noun property. The meaning postulate for privative adjectives is stated informally in (5). (5) ||counterfeit N|| \ ||N|| ¼ | Additional examples of each type are given below. (6)

(i) (ii) (iii)

intersective: sick, carnivorous, blond, rectangular, French. non-intersective but subsective: typical, recent, good, perfect, legendary. a. non-subsective and privative: would-be, past, spurious, imaginary, fictitious, fabricated (in one sense), mythical (maybe debatable); there are prefixes with this property too, like ex-, pseudo-, non-. b. plain non-subsective: potential, alleged, arguable, likely, predicted, putative, questionable, disputed.

276

BARBARA H. PARTEE

The conclusion drawn by Parsons, Kamp, Clark, and Montague was that the simplest general rule for the interpretation of the combination of an adjective with a noun (or common noun phrase: CNP) is the following: Adjectives are functions that map the (intensional) semantic value of the CNP they combine with onto the semantic value of the ADJ þ CNP combination. That is, ‘‘The denotation of an adjective phrase is always a function from properties to properties. (This was one of the proposals advanced by Kamp and Parsons.)’’ (Montague, 1970a: 211 in Montague, 1974). In terms of the type theory of Montague’s Intensional Logic (Montague, 1970b, 1973), where common noun phrases are of type oos,eW,tW, this meant that the most general type for adjectives was taken to be oos,oos,eW,tWW, oos,eW,tWW. On the variant of Bennett (1974), followed in most subsequent work in the Montague grammar tradition, the CNP is of type oe,tW, and adjectives are then of type oos,oe,tWW, oe,tWW. Meaning postulates specify various restrictions on these functions, characterizing various subclasses of adjectives. ‘‘Semantic features’’ may be seen as labels for meaning postulates which give them determinate content. Thus a lexical entry for an intersective adjective like green might contain the ‘‘feature’’ INTERSECTIVE, or þ INTERSECTIVE, which can be taken as labelling a semantic property of the adjective, spelled out by a meaning postulate. Meaning postulates for the subtypes listed above are spelled out in somewhat more formal terms below. The meaning postulates are written with the assumption that the basic type for all adjectives is oos,oe,tWW, oe,tWW. (7) Intersective adjectives: For each intersective adjective meaning ADJu, (Poe,tW&’Qos,oe,tWW’xe[ADJu(Q)(x)2P(x)&3Q(x)] (Alternatively, intersective adjectives (and only those) can be interpreted in type oe,tW. This automatically guarantees their intersectivity and eliminates the need for a meaning postulate. Typeshifting rules of the sort described in Partee (1995) will give them homonyms of type ooe,tW,oe,tWW when needed.) (8) Subsective adjectives: For each subsective adjective meaning ADJu, &’Qos,oe,tWW’xe[ADJu(Q)(x)-3Q(x)] The ‘‘plain’’ nonsubsective adjectives (alleged, possible) have no meaning postulate; this class is ‘‘noncommittal’’: an alleged murderer may or may not be a murderer.

PRIVATIVE ADJECTIVES: SUBSECTIVE PLUS COERCION

277

(9) Privative adjectives: For each privative adjective meaning ADJu, &’Qos,oe,tWW’xe[ADJu(Q)(x)-:[3Q(x)]] The privative adjectives (fake, counterfeit) have a ‘‘negative’’ meaning postulate; a fake gun is not a gun. On this familiar classification, adjectives are seen as forming a hierarchy from intersective to subsective to nonsubsective, with the privative adjectives an extreme case of the nonsubsective adjectives.1 There are of course many questions and disputes when it comes to assigning particular adjectives to particular classes. Kamp (1975) added an important dimension to the discussion in arguing that adjectives like tall, which at first sight seem to be non-intersective, are actually intersective but context-dependent. Kamp’s analysis found linguistic support in Siegel’s analysis of long-form and short-form adjectives in Russian (Siegel, 1976a, b). There has been much further work on the semantics of adjectives in the intervening years, and the context-dependence of interpretation of adjectives is central in the work of Klein (1980) and Kennedy (1997). Among many other debated points, one which has always been troubling, and to which we will return, is the question of whether an adjective or adjectivally used noun like fake or toy is really privative. One nagging problem is the evident tension between the apparent truth of (10a) and the undeniable well-formedness and interpretability of (10b). (10) a. b.

A fake gun is not a gun. Is that gun real or fake?

Privative Adjectives and Polish NP-Split Phenomena Nowak (2000) studied the phenomenon of ‘‘split PPs’’ and ‘‘split NPs’’ in Polish.2 Ignoring PPs for simplicity, and ignoring the topic-focus structure that motivates the splitting, the facts are that an NP 1

Although it has been common to treat these four classes as forming a scale, intersectiveosubsectiveonon-subsectiveoprivative, the meaning postulates do not actually give a linear order. From the meaning postulates, one can make a three-class scale, intersectiveosubsectiveonot-necessarily-subsective. The intersective adjectives are a subset of the subsective adjectives, which are in turn a subset of the unrestricted set of all adjectives. The privative adjectives are another subset of the unrestricted set, disjoint from the set of subsective adjectives. 2 See also Gouskova (2000) and Mehlhorn (2001) for related work on Russian, as well as Junghanns (2001) for a perspective on ‘‘rightward backgrounding’’ in Slavic.

278

BARBARA H. PARTEE

consisting of Adj and N in Polish may be ‘‘split’’, with either the Adj sentence-initial and the N sentence-final, or the N sentence-initial and the Adj sentence-final. Sequences of Adj’s can be sentence-initial; only a single element can be sentence-final. Examples of NP-splits (these are all actually PP-splits, which combine properties of NP-splits with constraints on where the preposition can end up) are given in (11) and (12) below. Sentences (11b) and (12b) are ‘split’ versions of sentences (11a) and (12a), which represent the unmarked word order. In (11b) the preposition and adjective are in sentence-initial position and the bare noun is sentence-final, while in (12b) the preposition and noun are sentence-initial and the adjective is sentence-final. All examples are from Nowak (2000).3 (11) a. Kelnerki rozmawia!y o przystojnym ch!opcu. waitresses talked about handsome-LOC boy-LOC ‘The waitresses talked about a handsome boy.’ b. O przystojnym kelnerki rozmawia!y ch!opcu. boy-LOC about handsome-LOC waitresses talked ‘The waitresses talked about a handsome BOY.’ (12) a. W!amano sie¸ do nowego sklepu. broke-in-NEUT.SG REFL to new-GEN store-GEN ‘Someone broke into the new store.’ b. Do sklepu w!amano sie¸ nowego. to store-GEN broke-in-NEUT.SG REFL new-GEN ‘Someone broke into the NEW store.’ What is of particular interest here is that some adjectives can participate in the splitting construction and some cannot. Do rozleg!ej weszlis´my doliny. to large-GEN enter-PST.1ST.PL valley-GEN ‘We entered a large VALLEY.’ b. Do doliny weszlis´my rozleg!ej. to valley-GEN enter-PST.1ST.PL large-GEN ‘We entered a LARGE valley.’ (14) a. *Z by!ym rozmawia!a prezydentem. with former-INSTR talk-PST.3RD.F.SG president-INSTR ‘She talked with the former PRESIDENT.’ (13) a.

3 Boz˙ena Cetnarowska (pers. comm.) has informed me that the data are less black-andwhite than they appear here. I am not competent to discuss the complexities; fortunately, they do not seem to substantially affect the generalizations made in the text.

PRIVATIVE ADJECTIVES: SUBSECTIVE PLUS COERCION

279

b. *Z prezydentem rozmawia!a by!ym. with president-INSTR talk-PST.3RD.F.SG former-INSTR ‘She talked with the FORMER president.’ Those that CAN split include: (15) a. b.

rozlegley ‘large’ biedny ‘poor’ in the sense of ‘not rich’, not in the sense of ‘pitiful’ c. Polish translations of generous, pretty, healthy, Chinese, talkative (intersective) d. skillful, recent, good, typical (subsective) e. counterfeit, past(?!), spurious, imaginary, fictitious (privative [!])

Those that CANNOT split include: (16) a. b.

biedny ‘poor’ in the sense of ‘pitiful’ Polish translations of alleged, potential, predicted, disputed (non-subsective, non-privative ‘modal’)

Another important fact is that the ones that cannot split also cannot occur predicatively. What is peculiar about this data in the light of the traditional classification outlined in the Section on Adjective Classification is that the NP-split phenomenon does not apply to a ‘‘natural class’’. It is unexpected for the intersective, subsective, and privative adjectives to pattern together, while the non-subsective adjectives that are ‘‘noncommittal’’ (and which can reasonably be characterized as ‘‘modal’’), cannot participate in the NP-split.

Principles of Interpretation and the ‘‘No Privative Adjectives’’ Hypothesis The hypothesis I propose is that Nowak’s data tells us that adjectives fake and imaginary aren’t actually privative, but subsective, and that no adjectives are actually privative. In interpreting a question like (10b) above or a sentence like (17) below, I hypothesize that we actually expand the denotation of ‘fur’ to include both fake and real fur. (17) a. b.

I don’t care whether that fur is fake fur or real fur. I don’t care whether that fur is fake or real.

280

BARBARA H. PARTEE

In fact, even in (10a), it is reasonable to suppose that the first occurrence of gun, modified by fake, is similarly coerced, whereas the second, unmodified, occurrence is not. Normally, in the absence of a modifier like fake or real, all guns are understood to be real guns, as is evident when one asks how many guns the law permits each person to own, for instance. Without the coerced expansion of the denotation of the noun, not only would fake be privative, but the adjective real would always be redundant.4 Kamp and Partee (1995), in discussing the ‘‘recalibration’’ of adjective interpretations in context, introduced a number of principles, including the following ‘‘Non-Vacuity Principle’’. (18) Non-Vacuity Principle (NVP): In any given context, try to interpret any predicate so that both its positive and negative extension are non-empty. (Kamp and Partee 1995: 161) The NVP applies not only to simple predicates but to predicates formed, for instance, by combination of an adjective and a noun: these should be interpreted in such a way that the ADJ þ N combination is a non-vacuous predicate. However, Kamp and Partee (1995) also argued, in part on the basis of clear examples like (19), that in ADJ þ N constructions, one first interprets the noun in the given context (ignoring the adjective), and then ‘‘recalibrates’’ the adjective as necessary. This principle is expressed as the ‘‘Head Primacy Principle’’ in (20). (19) a. b.

giant midget (a midget, but an exceptionally large one) midget giant (a giant, but an exceptionally small one)

(20) The Head Primacy Principle (HPP): In a modifier-head structure, the head is interpreted relative to the context of the whole constituent, and the modifier is interpreted relative to the local context created from the former context by the interpretation of the head.5 (Kamp and Partee 1995: 161) In many cases, the NVP and the HPP cooperate to account for the observed results, including not only the examples in (19), but also the 4

This property of real is noted by Lakoff (1987: 75). ‘‘In the simplest cases, the effect of the interpretation of a head noun on a given context will be to restrict the local domain to the positive extension of the head in the given context.’’ (Kamp and Partee 1995: 161, fn. 23) 5

PRIVATIVE ADJECTIVES: SUBSECTIVE PLUS COERCION

281

fact that the truth of (21b) below is compatible with a non-redundant use of the modifier in (21a). (21) a. b.

This is a sharp knife. Knives are sharp. (Kamp and Partee 1995: 162)

If the HPP is absolute, the proposed shift in the interpretation of the head noun under coercion by a privative adjective like fake or a ‘‘tautologous’’ adjective like real would be impossible. But there are other examples as well that suggest that the HPP probably has to be seen as non-absolute. In particular, there is a large and productive class of ‘‘constitutive material’’ modifiers that occur in examples like stone lion, wooden horse, velveteen rabbit, rubber duck. It is evidently so easy to shift nouns from their literal meaning to a meaning ‘‘representation/ model of . . . ’’ that we hardly notice the shift.6 Without trying to formalize this idea (which might have a natural expression within Optimality Theory), I would suggest the following. We normally try to obey both the HPP and the NVP. But if there is no reasonable way to obey the NVP without shifting the noun outside its normal bounds (as in the case of fake and real), then it may be shifted in such a way as to make the compound predicate obey the NVP. (Since this is always necessary with privative and ‘‘tautologous’’ modifiers, there might even be something in their lexical semantics that particularly indicates the need to shift the head to which they apply.) And if there is an extremely productive and ‘‘easy’’ shift of the noun that would make it easy to satisfy the NVP, as in the case of the ‘‘representations’’ in wooden horse, etc., there too we can override the HPP.7 6

In fact, in the literature on prototype theory, one can observe that many of the reported experiments on judgments of prototypicality are carried out with pictures of objects rather than actual objects, but all of the language of the experiments and of the discussion of the experiments refers to the corresponding objects, not to pictures of objects. And normally we don’t even notice; starting with our children and their picture books, we say things like, ‘‘Where’s the doggy? There’s the doggy!’’ Presumably no normal parent would say ‘‘Where’s the picture of the doggy? There’s the picture of the doggy!’’. In these examples, the shifting of the noun is not dependent on the presence of an adjective. Thanks to Benjamin Lyngfelt (pers. comm.) for the plausible suggestion that in the stone lion cases as well, the head noun may shift because of the context it finds itself in, and the modifier may not be essential. If so, then the stone lion cases do not threaten the Head Primacy Principle. 7 See Franks (1995) and Cooper and Franks (1996) for a possibly over-simplified but interesting concrete proposal for dealing not only with stone lion and fake gun but also iterated examples like fake stone lion.

282

BARBARA H. PARTEE

And I would suggest that no adjectives are privative. ‘‘Normal’’ adjectives are always subsective, and there should be some ways to identify ‘‘modal’’ adjectives as a special subclass, such that only they are not necessarily subsective.8 One surprising piece of data that remains in the Polish facts is that the adjective corresponding to English past does allow NP splitting and can occur in predicate position, while the adjective in (14) glossed as former does not. In English there does not seem to be any difference between past and former with respect to privativity – both are normally regarded as privative, although this is sometimes called into question: if a former/past senator is re-elected after spending some time out of office, is she then both a former/past and current senator, or does she cease being a former/past senator? It would be interesting if it were to turn out that Polish has two different adjectives with very similar meanings but which settle this issue in opposite ways. Or perhaps Polish past is like English fake in allowing a corresponding expansion of the extension of the noun it applies to. There are certainly other unclear cases in English: witness the uncertainty in classifying retired, dead as intersective versus privative. Is a retired professor a professor? Probably yes. Is a retired CEO a CEO? Probably no. Probably the line is not sharp because the extension of nouns is quite ‘adjustable’, as I first became aware in the course of listening to Hans Kamp’s lectures on tense and modality at the Linguistic Institute at the University of Massachusetts in the summer of 1974. I will close with an example which is my reworking, I think, of the kind of case study he was illustrating at one point in the lectures. When thinking about a question like (22a) below, there is in normal circumstances no temptation to include dead poets. Note that it’s not that the predicate poet by itself presupposes that the entities it applies to are alive, since we readily talk about anthologies of works by 18th century poets, and do we don’t usually refer to them as dead poets or former poets or ex-poets (the movie title Dead Poets Society has the feel of an intentionally surprising phrase). And be in Amherst cannot be said to presuppose that the entities it applies to are living animate entities. Yet the combination of poet with an extensional present-tense predicate together carries at least a very strong implicature that we are to count ‘‘live’’ poets. But exactly the opposite is true for question 8 Meredith Landman (2000) suggests an even stronger proposal, hypothesizing that apart from the modal adjectives, all apparent departures from intersectivity result either from context-sensitivity (as with tall) or from implicit additional arguments/parameters (as with skillful, good).

PRIVATIVE ADJECTIVES: SUBSECTIVE PLUS COERCION

283

(22b), since the predicate are buried, discounting ghoulish situations, when combined with a subject noun phrase that (normally) denotes an animate being, normally presupposes that the ‘animate entities’ it applies to are dead. (22) a. b.

How many poets are there in Amherst? How many poets are buried in Amherst?

The conclusion from such examples seems to be that whether the extension of a noun like poet at a given time includes only poets living at that time or both living and dead poets is highly dependent on the rest of the context, and easily shifts. Similar examples can easily be multiplied, and there may well be other phenomena that should be looked at in a similar light.9 If the ‘‘no privatives’’ hypothesis in this section can be maintained, then the classification of adjectives would be much more neatly constrained. Adjectives would still be functions from properties to properties in the most general case, but in harmony with the traditional notion of modifiers, they would normally be constrained to be subsective. We still need to allow for the ‘modal’ adjectives, which are not so constrained; the Polish data would provide fuel for a proposal to consider them syntactically as well as semantically distinct. I have said nothing to help with the problem of how to constrain the nonsubsective adjectives to just the kinds of ‘modal’ adjectives which actually occur and do not allow random nonsubsective functions: this challenge is raised in Heim (1999), and I have no solution to it. But if we can exclude privative adjectives completely, that would be one step in the direction of constraints. Of course more work also needs to be done on the detailed lexical semantics of each of the putatively privative adjectives, since they are far from identical; but that is beyond the scope of this note. 9 Bennett (1974) observed that Montague’s list of intensional verbs contained verbs of two different sorts. The typical intensional verb in Montague’s list was seek, which exhibits all the classic opacity properties. But Montague’s list also included worship and remember, and Bennett noted that an indefinite object with those verbs is always interpreted as ‘‘specific’’, not ‘‘non-specific’’; the only sense in which it is ‘‘intensional’’ is that the object in question need not exist at the world and time of the worshipping or remembering. It might be fruitful to consider these verbs as ones which sometimes coerce expansion of the domains in which their direct objects are interpreted rather than as intensional.

284

BARBARA H. PARTEE Acknowledgment

This short note is dedicated with affection to Hans Kamp on the occasion of his 60th birthday, belatedly, which is only fitting, given the frequency with which both of us have been late with promised texts, but no less warmly. I am grateful to Anita Nowak for acquainting me with the Polish Split-NP facts, and to Meredith Landman for initial stimulating discussion. For valuable comments and discussion I am grateful to Lisa Matthewson and the participants in her UMass Fall 2000 Pro-Seminar on Modifiers, to several classes of students at RGGU in Moscow, to participants of a colloquium in honor of Terry Parsons at Notre Dame in 2003, and to audiences in Leipzig, Copenhagen, Amsterdam, Arizona, Prague, St. Petersburg, Barcelona, La Corun ˜a, the University of Canterbury, NZ, and Go¨teborg. I thank Maria Gouskova, Boz˙ena Cetnarowska, and Boz˙ena Rozwadowska for very helpful discussions of the data. This paper overlaps in part with Partee (2007). This material is based on work supported by the National Science Foundation under Grants No. BCS-9905748 and BCS-0418311 to Barbara Partee and Vladimir Borschev.

References Bennett, M. 1974. Some Extensions of a Montague Fragment of English. Ph.D. dissertation, University of California at Los Angeles; distributed by Indiana University Linguistics Club, Bloomington. Clark, R. L. 1970. Concerning the logic of predicate modifiers. Nou ˆs 4:311–335. Cooper, R. and B. Franks. 1996. The iteration of concept combination in sense generation. In G. W. Cottrell, ed., Proceedings of the 18th Annual Conference of the Cognitive Science Society, pp. 523–528, San Diego, CA. Franks, B. 1995. Sense generation: A ‘‘Quasi-Classical’’ approach to concepts and concept combination. Cognitive Science 19:441–505. Gouskova, M. 2000. The Interaction of Topic, Focus and Constituency in Russian. Unpublished manuscript, Amherst, MA. Heim, I. 1999. Class Handout, Spring Seminar in Semantics. Unpublished manuscript, MIT, Cambridge, MA. Junghanns, U. 2001. On rightward backgrounding. In G. Zybatow, U. Junghanns, G. Mehlhorn, and L. Szucsich, eds., FDSL-3: Current Issues in Formal Slavic Linguistics, pp. 329–343. Frankfurt-Main: Peter Lang. Kamp, H. 1975. Two theories about adjectives. In E. L. Keenan, ed., Formal Semantics of Natural Language, pp. 123–155. Cambridge: Cambridge University Press. Kamp, H. and B. Partee. 1995. Prototype theory and compositionality. Cognition 57:129–191.

PRIVATIVE ADJECTIVES: SUBSECTIVE PLUS COERCION

285

Kennedy, C. 1997. Projecting the Adjective. Ph.D. dissertation, University of California, Santa Cruz, CA. Klein, E. 1980. A semantics for positive and comparative adjectives. Linguistics and Philosophy 4:1–45. Lakoff, G. 1987. Women, Fire, and Dangerous Things. Chicago, IL: The University of Chicago Press. Landman, M. 2000. Adjectival Modification Restricted. Unpublished manuscript, Amherst, MA. Mehlhorn, G. 2001. Information structure of discontinuous constituents in Russian. In G. Zybatow, U. Junghanns, G. Mehlhorn, and L. Szucsich, eds., FDSL-3: Current Issues in Formal Slavic Linguistics, pp. 344–352. Frankfurt-Main: Peter Lang. Montague, R. 1970a. English as a formal language. In B. Visentini, et al. ` e nella Tecnica, pp. 189–224. Reprinted in Montague Linguaggi nella Societa 1974, pp. 188–221. Milan: Edizioni di Comunita`. Montague, R. 1970b. Universal grammar. Theoria 36:373–398. Reprinted in Montague 1974, pp. 222–246 Montague, R. 1973. The proper treatment of quantification in ordinary English. In K. J. J. Hintikka, J. M. E. Moravcsik, and P. Suppes, eds., Approaches to Natural Language, pp. 221–242. Reprinted in Montague 1974, pp. 247–270. Dordrecht: Reidel. Montague, R. 1974. Formal Philosophy. Selected Papers of Richard Montague. Edited and with an introduction by Richmond H. Thomason. New Haven/ London: Yale University Press. Nowak, A. 2000. On Split PPs in Polish. Unpublished Manuscript, Amherst (Generals Paper, UMass.). Parsons, T. 1970. Some problems concerning the logic of grammatical modifiers. Synthe`se 21:320–334. Reprinted in D. Davidson and G. Harman (eds.). 1972, Semantics of Natural Language, pp. 127–141. Dordrecht: Reidel. Partee, B. 1995. Lexical Semantics and Compositionality. In L. Gleitman and M. Liberman, eds., An Invitation to Cognitive Science, 2nd edn. Vol. 1: Language, pp. 311–360. Cambridge: MIT Press. Partee, B. H. 2007. Compositionality and coercion in semantics: The dynamics of adjective meaning. In G. Bouma, I. Kra¨mer, and J. Zwarts, eds., Cognitive Foundations of Interpretation, pp. 145–161. Amsterdam: Royal Netherlands Academy of Arts and Sciences. Siegel, M. 1976a. Capturing the Russian adjective. In B. H. Partee, ed., Montague Grammar, pp. 293–309. New York: Academic Press. Siegel, M. E. A. 1976b. Capturing the Adjective. Ph.D. dissertation, University of Massachusetts, MA.