Discourse, Beliefs and Intentions: Semantic Defaults and Propositional Attitude Ascription (Current Research in the Semantics Pragmatics Interface, 2)

Discourse, Beliefs, and Intentions

Current Research in the Semantics/Pragmatics Interface Series Editors: K.M. Jaszczolt, University of Cambridge, UK K. Turner, University of Brighton, UK

In this series: TURNER, K. (ed.) 1999 GEURTS, B. forthcoming JASZCZOLT, K. M. (ed.) forthcoming

The Semantics/Pragmatics Interface, from Different Points of View Presuppositions and Pronouns ; , The Pragmatics of Propositional Attitude Reports

Related books JASZCZOLT & TURNER KOERNER & ASHER LAMARQUE

MEY

Contrastive Semantics and Pragmatics (2-Volume Set) Concise History of the Language Sciences Concise Encyclopedia of Philosophy of Language Concise Encyclopedia of Pragmatics

Related journals Journal of Pragmatics Editor: Jacob Mey Language & Communication Editors: Roy Harris and Talbot J. Taylor Language Sciences Editor: Nigel Love Lingua Editors: Teun Hoekstra, John Anderson and Neil Smith Free specimen copies of journals available on request For further information on the CRiSPI series and for details of how to submit a proposal go to: www. elsevier. nl/locate/series/crispi

DISCOURSE, BELIEFS, AND INTENTIONS Semantic Defaults and Propositional Attitude Ascription K.M. Jaszczolt University of Cambridge, UK

1999 ELSEVIER

Amsterdam - Lausanne - New York - Oxford - Shannon - Singapore - Tokyo

ELSEVIER SCIENCE Ltd The Boulevard, Langford Lane Kidlington, Oxford OX5 1GB, UK

c

919

Elsevie r

Science

Ltd . lAl right s reserved .

This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier Science Rights & Permissions Departments, PO Box 800, Oxford OXS 1DX. UK; phone: (+44) 1865 843830, fax: (+44) 1865 853333, e-mail: permissions@ elsevier.co.uk. You may also contact Rights & Permissions directly through Elsevier's home page (http://www.elsevier.nl) selecting first 'Customer Support', then 'General Information', then 'Permissions Query Form'. In the USA, users may clear permissions and make payments through the Copyright Clearance Center. Inc.. 222 Rosewood Drive. Danvers, MA 01923, USA; phone: (978) 7508400, fax: (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London WIP OLP. UK; phone: (+44) 171 631 5555: fax: (+44) 171 631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distributions of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions request to: Elsevier Science Rights & Permissions Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made. First edition 1999 British Library Cataloguing in Publication Data A catalogue record from the British Library has been applied for. Library of Congress Cataloging-in-Publication Data Jaszczolt, Katarzyna. Discourse, beliefs and intentions : semantic defaults and propositional attitude ascription / by K.M. Jaszczolt. p. cm. Includes bibliographical references and index. ISBN 0-08-043060-0 1. Semantics. 2. Ambiguity. 3. Pragmatics. I. Title. P325.J337 1999 401'.43—dc21 99-21253 CIP

ISBN: 0 08 043060 0

©The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.

dedykuJe mojej mamie

Current Research in the Semantics/Pragmatics Interface (CRiSPI)

Series Editors: K.M. Jaszczolt, University of Cambridge, UK, and K. Turner, University of Brighton, UK Editorial Advisory Board: N. Asher, USA J. van der Auwera, USA C. Casadio, Italy M. Dascal, Israel B. Fraser, USA T. Fretheim, Norway B. Gillon, Canada P. Gochet, Belgium J Groenendijk, The Netherlands Yueguo Gu, PRC A. Kasher, Israel M. Krifka, USA Susumu Kubo, Japan Chungmin Lee, Korea S. Levinson, The Netherlands T. McEnery, UK F. Nemo, France P. Pelyvlas, Hungary J. Peregrin, Czech Republic A. Ramsay, UK R. van der Sandt, The Netherlands R. Stalnaker, USA M. Stokhof, The Netherlands G. Ward, USA 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.

Contents

Preface

xi

List of Abbreviations and Symbols

xv

Introduction

xvii

1. Semantic Ambiguities and Semantic Underspecification 1.1. Ambiguous Discourse? 1.2. Highlights from the Ambiguity Debates 1.2.1. Logical Form and Propositional Representation 1.2.2. Ambiguity Tests 1.2.3. Ambiguity of i f . . . then! 1.2.4. Ambiguity of Negation? 1.2.5. Ambiguity of and? 1.2.6. Ambiguity of Number Terms? 1.2.7. Ambiguity of Definite Noun Phrases? 1.2.8. Ambiguity of Indefinite Noun Phrases? 1.2.9. Ambiguity of Propositional Attitude Sentences? 1.2.10. Interim Conclusions 1.3. What Is Said 1.4. Towards Delimiting Implicatures 1.4.1. Neo-Griceans and Ineradicable Conventions 1.4.2. Relevance Theory and Frames 1.5. Conclusions

1 2 5 5 9 11 13 18 20 23 24 26 26 27 37 37 42 44

2. Semantic Defaults 2.1. Setting the Scene 2.2. Intentions in Communication 2.3. The Primary Intention Principle and Definite Descriptions 2.4. The Principle of the Parsimony of Levels 2.5. Default Semantics?

47 48 49 51 56 58

viii

Contents

2.6.

2.7. 2.8. 2.9. 2.10.

2.5.1. Default Reasoning 2.5.2. Defeasible Knowledge 2.5.3. Defaults for Definite Descriptions: Interaction of Intentions 2.5.4. Defaults for Indefinite Descriptions Other Applications 2.6.1. On Semantic Representation and Negation Again 2.6.2. And, or, and if 2.6.3. Number Terms Defaults in Dynamic Semantics Interim Conclusions The Doubly-Dynamic Approach: Assumption Recovery or Creation? Conclusions

58 60 62 65 66 66 67 68 70 75 76 85

3. Intentionality and Propositional Attitudes 87 3.1. 'Directedness' of Acts of Consciousness: The Phenomenological Tradition 88 3.2. Intentional Relation and Defaults 92 3.3. Intentionality and Voluntariness 94 3.4. Intentions and Intentionality 96 3.5. De Re, De Dicto, and the Horizons 99 3.6. Speech Acts: Double Level of Intentionality? 104 3.7. Are Intentions in the Head? 1ll 3.8. Intentionality and Ambiguity: Concluding Remarks 119 4. The Default De Re Principle 121 4.1. Propositional Attitudes: A Close-Up 122 4.1.1. Propositional Attitudes, Attitude Sentences, and Attitude 122 Reports 4.1.2. The 'Ortcutt Puzzle' and Its Interpretations 124 4.1.3. Factives and Non-Factives 130 4.1.4. The Object of Belief 131 4.1.5. Defaults for Belief Reports or Defaults for Referring Expressions? 133 4.1.6. Varieties of De Re and De Dicto 135 4.1.7. Substitutivity? 137 4.2. Whose Meaning? On Sense and Mode of Presentation 142 4.2.1. In Defence of Speaker-Dependent Senses 142 4.2.2. Types of Modes of Presentation and Default Semantics 147 4.2.3. The Persistence of Sense 158 4.2.4. Sense-Based Semantics? 161 4.3. De Re, De Dicto1, and De Dicto Proper 166

Contents

ix

4.4. Red Giants and White Dwarfs: Context-Dependence of Attitude Ascription 176 4.5. Default De Re 183 4.6. Interim Conclusions 192 4.7. Loose Ends: Context and Presupposition Projection 194 5. The 5.1. 5.2. 5.3. 5.4. 5.5. 5.6.

Lexicon and the Power of Referring An Overview Redefining Referring Expressions Proper Names and Modes of Presentation Degrees of Referring: A Unified Approach Referring by Indexicals Referent Accessibility and the Strength of Referring 5.6.1. The Givenness Hierarchy 5.6.2. Accessibility Theory 5.6.3. Classification and Explanation 5.6.4. Intentions Again 5.7. Conclusions

203 204 205 211 215 216 223 223 227 230 232 234

6. Vehicles of Thought in Attitude Ascription 6.1. Thoughts and Acts of Thought 6.2. Vehicles of Thought 6.3. Vehicles and Attitudes 6.3.1. Beyond Language 6.3.2. Language: The First Vehicle and the Identification of the Referent 6.3.3. Concepts and the First Vehicle 6.3.4. Mental Images: The Second Vehicle 6.3.5. Actions: The Third Vehicle 6.4. Default Interpretations 6.5. The Core of Meaning 6.6. Summary

237 238 244 249 249 251 256 262 267 271 272 274

7. Discourse Representation Theory and Propositional Attitudes 7.1. Semantics and Pragmatics Revisited 7.2. Context 7.3. Belief Expressions and Belief Reports in DRT 7.4. DRT and Intentions 7.5. Concluding Remarks

277 278 280 284 296 298

8. Belief Reports in a Contrastive Perspective 8.1. Belief, Culture, and Translation 8.2. Contrastive Semantics and Pragmatics

301 302 306

x

Contents

8.3. Context and Markedness 8.4. Complementizer That in Contrast 8.4.1. English and Polish: A Surface Overview 8.4.2. Degree of Clausiness 8.4.3. A Lambda-Categorial Language and Sensitivity to Structure 8.4.4. The Application: Defaults and Clausiness 8.5. Final Remarks: What Languages Reveal and Hide

310 306 311 311 320 324 332

9. Denouement: Double Occam's Razor 9.1. The Semantics of Common Sense 9.2. Beyond Propositional Attitudes 9.3. Final Remarks

335 336 337 338

Bibliography

341

Index

357

Preface The theory presented in this book is a voice in the ongoing debate concerning the semantics/pragmatics interface which was set off by the controversy surrounding the cases of the so-called semantic ambiguity back in the early 1970s. The main gist of my approach can be summarized as follows. Conversation relies on default interpretations, the hearer arrives at such a default interpretation without going through the stage of choosing between understandings of a 'semantically ambiguous' sentence. Pragmatic aspects of what is said are already present at the basic semantic level of utterance interpretation. Together with the logical form understood as the output of grammar, they constitute the semantic/propositional representation of the utterance. This is what I call Default Semantics. This semantic representation can be overridden, leading to various degrees of departure from the default. The Default Semantics is a reflection of intentionality that characterizes all mental acts. For the purpose of utterance interpretation, I demonstrate how intentionality can be translated into various types of intentions in communication. This theory sits mid-way between semantic ambiguity and underspecification but falls happily under the 'monoguist' label and 'pragmatic intrusionism' as perceived in dynamic approaches to communication such as Discourse Representation Theory. The approach is spelled out in the example of sentences reporting on propositional attitudes, and in particular reports on beliefs. The main ambiguity I am concerned with is, naturally, the ambiguity exhibited by expressions used by speakers to refer, and the related species of the de re/de dicto ambiguity of belief reports. Hence, the book presents the theory of Default Semantics applied to puzzles associated with referring, signals its applicability to natural language negation and, albeit more tentatively, other sentential connectives, as well as instances where the expansion or completion of the sentence is needed before the hearer can arrive at a plausible candidate for what is said. It is hoped that future detailed studies of particular phenomena classified as instances of semantic ambiguity or underspecification will further strengthen its appeal. The book should be of interest to linguists of postgraduate and advanced undergraduate level, as well as other researchers in linguistics, philosophy of language, social anthropology and cognitive science. It can be read by

xii

Preface

anybody interested in the subject who has some elementary knowledge of linguistic semantics, pragmatics, and first-order logic. The balance between review sections and the exposition of the new theory makes it possible to use the book on a more advanced as well as a more basic level. All symbols and abbreviations used are listed below. Many people contributed at various stages to the crystallization of this work, beginning with my doctoral research conducted between 1988-1992 at the University of Oxford up to the present day. I owe thanks to Kent Bach, Gordon Baker, Gillian Brown, Billy Clark, L. Jonathan Cohen, David Cram, Richard Dammann, Michael Dummett, Vicky Escandell-Vidal, Rom Harre, Roman Kalisz, Ewan Klein, Barbara Lewandowska-Tomaszczyk, Peter Matthews, Richard Matthews, Stephen Schiffer, Pieter Seuren, Barry Smith (SUNY), Timothy Williamson, and Deirdre Wilson. My particular thanks go to Ken Turner, the co-editor of CRiSPI, for reading the first draft and supplying some useful publications. To Chris Pringle from Elsevier Science I am grateful for his support for the CRiSPI series. Finally, I owe thanks to my husband Charles Berthon for his encouragement. Some of the ideas included in this book, including Default Semantics, have crystallized over the period of several years of thinking about these issues. The consecutive stages of this research were reported in several journal articles. I owe thanks to several publishing houses for granting permission to adapt relevant ideas and/or relevant sections of the papers for the purpose of this book. In particular, I would like to thank the publishers listed below. The bibliographical details of the relevant papers are given in brackets. Elsevier Science ('Relevance and Infinity: Implications for Discourse Interpretation'. Journal of Pragmatics 25 (1996), 703-722; The "Default De Re" Principle for the Interpretation of Belief Utterances', Journal of Pragmatics 28 (1997), 315-336; 'Default Semantics, Pragmatics, and Intentions', in K. Turner (ed.), 1999, The Semantics/Pragmatics Interface from Different Points of View. CRiSPI, vol. 1, 199-232); Walter de Gruyter ('Discourse about Beliefs', Theoretical Linguistics 24 (1998), 1-28); Julius Groos ('Reports on Beliefs: Default Interpretations and Default Intentions', Journal of Literary Semantics 27 (1998), 31-42; 'Referring in Discourse: Referential Intention and the "Taking for Granted" Principle', Journal of Literary Semantics 27 (1998), 96-109); Adam Mickiewicz University, Poznan ('De Re/De Dicto: A Semantics of Belief Sentences', Papers and Studies in Contrastive Linguistics 28 (1993), 39-64).

Preface

xiii

The proposed Default Semantics should be seen as a step in the ongoing discussion rather than an attempt to overturn accepted truths. The affinities with the extant proposals of the neo-Gricean theory of implicature, default reasoning, and the neo-Fregean approach to reference should be obvious. To anticipate possible criticism, I should add that Default Semantics is not a complete theory. It is presented here as a new programme, an alternative to semantic ambiguity on the one hand and underspecification on the other. It is a proposal of a theory and an attempt at its application to one area of language. It is hoped that this proposal will meet with some interest and support and will be subsequently tested in its application to various fragments of various natural languages. 'Default intentions' and 'vehicles of thought' must remain partly justified theoretical proposals until humans can look into their neural states and say with certainty what they see there. Cambridge, January 1999

This page intentionally left blank

List of Abbreviations and Symbols +>

'conversationally implicates' truth-functional negation & truth-functional conjunction ->, 3 truth-functional implication V universal quantifier (cf. 'all', 'every') 3 existential quantifier (cf. 'a', 'some') ISg first person singular number 3Sg third person singular number Ace accusative case ACI accusative with infinitive construction (accusativus cum infinitivo) Dem demonstrative pronoun DI Degrees of Intentions principle DRT Discourse Representation Theory DRS Discourse Representation Structure fn footnote FoL first-order logic GCI generalized conversational implicature Gen genitive case iff if and only if ILF interpreted logical form Ind indicative mood Inf infinitive Instr instrumental case LF logical form Loc locative case MoP mode of presentation MOR Modified Occam's Razor NP noun phrase Past past tense Past Pple past participle PI Primary Intention principle POL Parsimony of Levels principle Pres present tense

xvi

Refl Subj SVO TCS VP wff

List of abbreviations and symbols

reflexive pronoun subjunctive mood subject-verb-object word order Theoretical Contrastive Studies verb phrase well-formed formula

Introduction "We shall never get rid of the temptation to perceive the universe as a secret script to which we stubbornly try to find the clue." Kolakowski (1988: 118)

Other people's beliefs can be creatures of darkness but only insofar as they are not expressed. This book is about ambiguities, beliefs, language, communication and cognition. It deals with the fundamental issue of the interpretation of the speaker's utterance expressing a belief and reporting on beliefs of other people in the form of oratio obliqua. The main objective is to dispel the mystery associated with the process of interpretation of utterances expressing beliefs and belief reports. The sentences under consideration are of the form of (1) and (2). (1) is an instance of an expression of belief and (2) is a report on a belief. (1)

The man in the brown hat is a spy.

(2)

Ralph believes that the man in the brown hat is a spy.

An utterance of (1) potentially exhibits an ambiguity between a referential and nonreferential (attributive) reading of the definite description, and the belief report in (2) is potentially ambiguous between reporting on a belief concerning a particular person that he is a spy (the de re reading) and on the belief about the situation as a whole (the de dicto reading). Proper names, although normally not analysed in terms of referential and nonreferential reading, are also used in referring to a known or an unknown individual, and this situation is reflected in belief reports. Like definite descriptions, they can also be used in a mistaken act of referring, i.e. referring to somebody who does not hold that name. So far the problem of referential ambiguities has remained largely unresolved. Linguists tend to adopt one of the available stands ranging from the semantic ambiguity to the generality of sense. I develop here Kent Bach's idea that "an expression's referring function (...) is (...) not part of its semantic contribution to the sentence" (1987a: 58). I demonstrate that

xviii

Introduction

referring is the default function guaranteed by the referential intention (embedded in the communicative intention) and is transported to semantics by the intrusion of pragmatic processes before the propositional representation is formed by the hearer. This standpoint has far-reaching implications for the notion of reference, the sense of the term 'referring expression', and the theory of intensional contexts. The category of referring expressions has been a matter of controversy at least since Frege and Russell brought it to the fore. Definite descriptions, for instance, are sometimes classified together with quantifiers (pome, few, all,...) and sometimes with referring expressions (proper names and pronouns). I follow Bach (1987a) in classifying them as referring expressions, for reasons to be spelled out shortly. On the topic of referring, it is argued here that it is a pragmatic phenomenon and it helps establish the semantic representation of a sentence by means of the intrusion of pragmatic processes. This intrusion is guaranteed by a very important feature of communication, namely its intentionality. Intentions help establish one propositional form and enable an unambiguous semantics. By semantics I mean here a semantic theory applicable to the particular speaker - an idea based on the following proposal from Larson and Segal (1995: 187): "... an important aspect of deciding when a semantic theory applies to a given speaker will lie in deciding whether an axiom may truly be associated with a concept in the speaker's mind (...) and when it may not." This choice between axioms is in fact performed instantaneously by the hearer by means of intention recognition: the default interpretation guaranteed by intentionality is always tried first. In other words, it is presumed. Differences in referring which are exemplified by the readings of (1) and (2) give rise to puzzles that are ample in the philosophical literature. These differences give grounds to the view that belief expressions and reports on beliefs are referentially ambiguous, the first as extensional contexts where the expression used to refer can either refer to an individual or name whoever or whatever holds the name, and the other as intensional contexts where the substitution of coreferential expressions may produce a sentence with a different truth value. The referential ambiguity in (1) is well acknowledged in the case of definite descriptions but also seems to occur with proper names, no matter how directly referential we try to make them for semantic analyses. In addition, both can be used correctly or incorrectly, in the latter case engendering referential mistakes. Now, the semantic/pragmatic perspective of discourse representation as opposed to a semantics of isolated sentences allows one to use information from intentions in communication (Grice 1969;

Introduction

xix

Strawson 1964a; Recanati 1993; Burkhardt, ed. 1990; Bach 1987a, b, 1994a, b), default reasoning (Bach 1984; Lascarides and Oberlander 1993; Levinson 1995) and assumption recovery performed by the hearer (Sperber and Wilson 1986a; Jaszczolt 1996a) in establishing the semantic representation of these problematic constructions. I propose and defend the idea of default interpretations of belief reports and expressions of belief. Potential semantic ambiguities caused by referring terms do not in fact take place and this is due to the interaction of the semantic information (semantic representation of the sentence) and intentions in communication. This interaction produces a default interpretation, which in practice reduces the putative semantic ambiguity to the case of (a) defaults in understanding and (b) overridden defaults producing a justified marked interpretation. Looked at more closely, defaults in the interpretation of belief expressions are ensured by intending to refer which I call the primary referential intention. Naturally, this primary intention interacts with other intentions in communication, to be spelled out in more detail in what follows (see also Jaszczolt 1997a, 1999). Default Semantics is governed by three main principles which I call the Parsimony of Levels (POL), Degrees of Intentions (DI), and the Primary Intention (PI): POL:

Levels of senses are not to be multiplied beyond necessity.

DI:

Intentions come in various sizes, i.e. they allow for degrees.

PI:

The primary role of intention in communication is to secure the referent of the speaker's utterance.

These principles summarize the idea introduced above, namely that the semantic representation of belief utterances is the product of the logical form (LF) and intentions in their default and non-default values. Some linguists call this level of representation propositional, to distinguish it from the underspecified semantics. In my account, these two levels are conflated and underspecification disappears, but so do ambiguities. If, as I hope to show, this account works, its advantages go without saying. The proposal of the Default Semantics for apparent semantic ambiguities is carefully defended throughout this book and is given formal support in the Discourse Representation Theory (DRT) of Kamp and Reyle (1993) which is a suitably 'intrusionist' model in that it allows for the incorporation of the pragmatic information in the semantic analysis (see Levinson 1988; Jaszczolt 1998a). The reasons for utilizing DRT are twofold. First, it allows for a formal presentation of the dynamism of utterance interpretation that comes from the flow of conversation and the availability of new contextual information. Second, the idea of the interaction of intentions in communication with the semantic representation is an improvement on DRT itself. It narrows

xx

Introduction

down the possibilities of interpretation of the speaker's utterance to one. Defaults in interpretation organize the multiple possibilities of discourse representation in DRT in a systematic way. Instead of talking about alternative discourse representation structures (DRSs), we can now talk about the default DRS and departures from the default. In the research on referential ambiguities to date, attention has been focused on intra-theoretic concerns of philosophers (here principally the truthpreserving substitutivity of coreferential expressions, i.e. substitutivity salva veritate) and model-theoretic solutions in formal linguistics which allow for a varying degree of contribution from pragmatics. The latter is still frequently voiced more as a credo than a sound proposal. A conversational approach that would account for the roles of the speaker and the hearer in meaning construction is badly needed. By this I mean an approach that would reveal and utilize aspects of conversation that have been neglected both in formal semantics and in philosophical discourse on the de re/de dicto ambiguity of prepositional attitude reports. Intentions in communication are one such aspect, vehicles of thought are another, and conversational data in a contrastive perspective yet another. This book deals with all three. In other words, belief utterances are analysed as sentences, activities and cognitive processes, to allow for a wide understanding of context. In this way, context acquires an epistemological grounding in corresponding carriers of thought: language, mental images, and actions. It goes without saying that this analysis requires an important assumption of the universality of the semantics and pragmatics of language systems, or what Turner and Jaszczolt (1996: 5) discuss as the Semantic Universals Hypothesis and the Pragmatic Universals Hypothesis: the semantic, as well as pragmatic, systems of the world's natural languages share at least some properties. This is a condition sine qua non for any semantic or pragmatic analysis which aspires to cross-linguistic status. The intrusion of pragmatics, in the form of intentions, into establishing the semantic representation is at the same time a voice in the dispute concerning the semantics/pragmatics interface, and a fortiori the dispute concerning the boundary between the two.

In more detail, the content of the chapters is as follows. The topic of the first chapter is the state-of-the-art in the semantics-pragmatics debate, followed by a critical assessment of ambiguists and monoguists, culminating in a combination of the strengths of the two camps in the proposal of the Default Semantics in Chapter 2. It is suggested that the POL principle restricts the usefulness of the concept of underspecified sense and leads to its replacement with the interaction of intentions with the LF on the level of

Introduction

xxi

semantic representation. Unambiguous semantics acquires a new meaning, dissociated from underspecification. Chapter 2 spells out the idea of semantic defaults and introduces the types of intentions in communication. It contains a discussion of the PI principle and the tenet that intentions allow for degrees, summarized in the DI principle. These degrees account for various readings for potentially semantically ambiguous utterances. The idea of semantic defaults and principled departures from these defaults is dubbed the Default Semantics. It follows with the application of the Default Semantics to definite descriptions and signals its application to other problematic domains of language. The POL principle is applied to definite descriptions, juxtaposing them with bona fide referring expressions and stressing their similarities. The referential intention is said to play an important role in communication and secure the unmarked interpretation of a potentially semantically ambiguous sentence, which is summarized in the PI principle. This intention interacts with other intentions to produce a range of marked interpretations. It is also argued that meaning is established neither by the speaker nor by the hearer but rather in-between, in the process of assumption recovery and assumption construction in what is called here a 'doubly-dynamic' perspective of 'attack and defence'. The proposal of the 'default de re* reading of belief reports receives epistemological support in Chapter 3 in the presentation of the association of intentions in communication with the intentionaliry of mental acts. Intentionality secures default interpretations or 'jumping to conclusions' instead of 'thinking twice' or considering possible paths in which potential semantic ambiguity would, putatively, otherwise lead the interpretation. 'Putatively', because conversation without intentionality is inconceivable. Indeed, relevant mental states without intentionality are hardly conceivable either. Chapter 4 gets to the core of prepositional attitude ascription. It contains a critical assessment of the debates surrounding the failure of truth-preserving substitutivity of expressions in the subject position of the embedded sentence of belief reports and combines it with a proposal of the 'default de re' reading. This reading is supplemented with the enumeration of possible forms that the departure from the default can take. It is suggested that the substitutions made in conversation differ from the truth-preserving substitutions discussed in the philosophical literature. Some notion of sense is shown to be indispensable for the semantics of attitude reports and it is suggested that it takes the form of modes of presentation (MoPs) that can be of various degrees of detail and can make various degrees of contribution to the semantic representation. Chapter 5 puts together the findings concerning referring properties of expressions and redefines the class of referring expressions by including definite descriptions in their default, unmarked reading. It contains some arguments for the degrees of strength of referring. Default Semantics is juxtaposed with the theories of Referent Accessibility and the Givenness Hierarchy. Chapter 6 takes the lesson from

xxii

Introduction

intentionality further into a discussion of the relation between utterances and beliefs. Language turns out to be one of the vehicles of thought, which allows us to jump to a more practical sphere of conversation and answer the question of 'How do I know what you mean?' without the help of notoriously inapplicable sociolinguistic generalizations. Meaning is carried by linguistic expressions, mental images, and actions and these vehicles of thought interact. As a result of this 'going beyond language', the notion of sense is extended to cater for all these vehicles and Husserl's term noema is adopted for it. The universality and culture-dependence of concepts are also taken on board. Chapter 7 presents the theory of Default Semantics in the framework of DRT, emphasizing the strengths of intrusionism in dynamic semantics, as well as the weaknesses of DRT as a model that assigns equal status to all possible understandings of a potentially semantically ambiguous sentence. I demonstrate how DRT can avoid this weakness when combined with the proposal of the Default Semantics. The degrees of referential intention are translated into the levels of embedding of the discourse referent that stands for an individual in the DRSs. Continuing on a practical note, Chapter 8 looks into the properties of that-clauses in a contrastive Polish-English perspective to find independent evidence for the default semantics of sentences reporting beliefs. This independent support is found in the unmarkedness of the Polish construction which corresponds to the de re reading of English negative sentences of the type "I don't believe that B cps." Here I make use of the idea of 'structured compositionality' and different sensitivity to the structure of the that-clause exhibited by the complementizer that in its different understandings. The final chapter sums up the results, points out the common-sense plausibility of the Default Semantics, and offers some guided speculations concerning other domains of apparent semantic ambiguity, with directives for the application of the approach developed here to their analysis.

CHAPTER 1

Semantic Ambiguities and Semantic Underspecification "Razor, where is thy sting? Parsimony, where is thy victory?" Horn (1989: 432).

Contents 1.1. Ambiguous Discourse? 1.2. Highlights from the Ambiguity Debates 1.2.1. Logical Form and Prepositional Representation 1.2.2. Ambiguity Tests 1.2.3. Ambiguity of i f . . . then! 1.2.4. Ambiguity of Negation? 1.2.5. Ambiguity of and! 1.2.6. Ambiguity of Number Terms? 1.2.7. Ambiguity of Definite Noun Phrases? 1.2.8. Ambiguity of Indefinite Noun Phrases? 1.2.9. Ambiguity of Prepositional Attitude Sentences? 1.2.10. Interim Conclusions 1.3. What Is Said 1.4. Towards Delimiting Implieatures 1.4.1. Neo-Griceans and Ineradicable Conventions 1.4.2. Relevance Theory and Frames 1.5. Conclusions

DISCOURSE, BELIEFS, AND INTENTIONS Current Research in the Semantics/Pragmatics Interface, Vol. 2 K.M. Jaszczolt © 1999 Elsevier Science Ltd. All rights reserved 1

2 5 5 9 11 13 18 20 23 24 26 26 27 37 37 42 44

2

Discourse, beliefs, and intentions

Chapter 1

1.1. Ambiguous Discourse? The invocation in the motto sums up the current state of affairs in approaches to ambiguities in utterance interpretation. The boundary between 'what is said' and 'what is communicated' is drawn in a variety of ways. The major standpoints can be classified as follows: (i) Some sentences are semantically ambiguous; (ii) There is no semantic ambiguity and the differences in meaning between the two (or more) understandings can be attributed to implicated information; (iii) There is no semantic ambiguity and the differences in meaning belong to what is said rather than what is implicated. Semantics is underspecified as to these aspects of meaning; (iv) There is no semantic ambiguity, semantics is underspecified, but the differences in meaning belong to the middle level located between what is said and what is implicated. The first standpoint is represented, most prominently, by Russell, the second by Grice, the third by Atlas, Kempson, Sperber and Wilson and their acolytes, and the fourth by Bach and Levinson, albeit interpreting the middle level rather differently. The affinities between (iii) and (iv) are significant. The more so that the status of the 'middle level' is as yet somewhat dubious. Semantic ambiguity or underspecification has to be distinguished from instances of grammatical ambiguity which can be either lexical ('Let's go to the bank') or syntactic ('Visiting relatives can be boring'). Grammatical ambiguity groups instances where the grammar provides two (or more) distinct interpretations. In the case of semantic ambiguity or underspecification, the grammar renders one representation that corresponds to various conditions of truth and falsity, i.e. various states of affairs. The difference can be summarized as follows: whereas lexical and syntactic ambiguities result in sentences which can be unrelated and cannot be traced back to one meaningful representation, the so-called semantic ambiguities are characterized by a general semantic representation from which a truth-conditionally evaluable proposition can be derived by means of pragmatic rules. This avoidance of unnecessary ambiguities is best spelled out in Grice's (1978) principle of parsimony called Modified Occam's Razor (henceforth: MOR): MOR: Senses (linguistic meanings) are not to be multiplied beyond necessity.

Section 1

Semantic ambiguities and semantic underspecification

3

(See Recanati 1989a, b; 1994 for a discussion). The merits of this methodological principle are not to be doubted. However, its functioning in utterance interpretation is unclear. Is it a rule that guards a semanticist from postulating unwanted entities or is it a generalization on a stage in the process by which hearers arrive at a correct interpretation of an utterance? If the first, then it has not been successful, as the motto from Horn indicates. If the latter, what is the use of spelling out what is not there? The main objective of this chapter is a critical assessment of both the underspecification (sense-generality) and ambiguity positions. It is argued here that neither is successful and neither of them is needed. The ambiguity stance has been taken on board by followers of underspecification but it is time to critically assess the latter as well. Once the position of semantic underspecification is challenged, the rest of the book will be devoted to introducing a more successful alternative which is the theory of default interpretations and default semantics. One of the standard examples of alleged semantic ambiguity discussed in the literature involves negation. Sentence (1) has two readings. (1)

The present king of France is not bald.

According to one, the existence of the king is negated. There is no king of France and hence he cannot be bald and the proposition is true. According to the other, more natural, reading, the property of being bald is negated and the sentence is sometimes seen as lacking truth value (Strawson 1950). According to Russell's standard interpretation, the sentence is ambiguous due to the fact that negation can take a wide or narrow scope as represented in (la) and (Ib), where 'x' is an individual variable (cf. e.g. Neale 1990). (la)

--ix (King of France (x) & Vy (King of France (y) -* y = x) & Bald (x))

(Ib)

3x (King of France (x) & Vy (King of France (y) -+ y = x) & -Bald (x))

But (la) is not a good representation of the first reading. The wide-scope negation can correspond to both readings: 'It is not true that...' can mean an objection due to the non-existence of the person as well as due to the fact that he has hair on his head. In fact, since the two types of negation are not logically independent but rather the second entails the first, semantic ambiguity should not be postulated (cf. Kempson 1979). Instead, the hearer relies on pragmatic processes in establishing which understanding was intended by the speaker. In a more theoretical jargon, the semantic form is either general, underspecified, or corresponds to the logical form with

4

Discourse, beliefs, and intentions

Chapter 1

the negation taking a wide scope (la). Pragmatic processes work on it and establish the unique prepositional representation. This prepositional representation is an 'enriched' variant of the semantic representation and it is this prepositional representation that can be assessed as to its truth and falsity. In current pragmatic research, it has been widely acknowledged that the semantic representation can be underspecified and can be further enriched by pragmatic processes according to Atlas's pastiche of Kant: "Pragmatic inference without sense-generality is blind, but sensegenerality without pragmatic inference is empty." Atlas (1989: 124). (See also Atlas 1979 and Turner 1991,1997 for discussion.) Instead of selecting multiple readings of a sentence, the hearer relies on the context in constructing one interpretation by filling in what is missing in the sentence meaning revealed by its grammar and lexicon. This is how pragmatic inference prevents ambiguities from arising. In fact, on this account, it does not make sense to say what it would mean for semantic ambiguities to arise: they are exorcized globally by means of a theory of discourse interpretation which assigns a powerful role to pragmatic processes. Sentence (1) does not have a single set of truth conditions that could be assigned to it prior to pragmatic inference. An interpretation is assigned first and then truth conditions are clear. So, the fact that a sentence can be assigned two logical forms as in (la) and (Ib) does not yet mean that it has these logical forms, neither does it mean that it is ambiguous (cf. Bach 1987a: 207). Now, we can go further and argue, by the same non sequitur, that the possibility of assigning more than one logical form does not yet mean that the semantic representation is underspecified as to the meaning and truth conditions. Underspecified semantics and underspecified sense seem an easy solution to cases such as (1). When a sentence has two readings as in (la) and (Ib), and no satisfactory explanation of this situation is provided by spelling out the logical form, i.e. by the scope of negation, there is no semantic ambiguity. It is assumed that semantic ambiguity should rely on the existence of two independent logical forms (see Kempson 1977, 1979; Kempson and Cormack 1981). Even if one accepts this theoretical assumption, one is forced to take a big leap from there to claim that the semantic representation is general, underspecified, underdetermined, vague, neutral - these are only some of the adjectives used in the relevant literature (see Zwicky and Sadock 1975: 2 and Green 1996: 1). In effect, one is committed then to the claim that there is no unique sentence meaning, and all there is, is utterance meaning arrived at through pragmatic processing. This seems to make the semantic layer somewhat weak and insignificant, as well as contradict the common-sense intuitions concerning sentence meaning. While the rejection

Section 2

Semantic ambiguities and semantic underspecification

5

of the semantic ambiguity of negation seems justified, it also seems to call for a more extreme move, that is a rejection of the level of underdetermined semantics. It is easy to see the need for it when one entertains the possibility that the ambiguity/underdetermination dilemma is a wrongly posed problem. Returning to the success in implementing Grice's MOR, Horn's concern about parsimony seems to be justified. Ambiguity is exorcized from semantics, only to appear in the form of multiple possibilities of pragmatic enrichment of the semantic representation. If one avoids calling the relation of (1) to its meanings by its name, one is forced to group (1) together with such instances of underspecification as (2) or (3), which is equally unparsimonious in that it places even more burden on pragmatic processes. (2)

I haven't eaten.

(3)

Steel isn't strong enough.

(from Bach 1994a: 268). At the same time, one pours out ambiguities together with the useful distinction between having two or three salient readings on the one hand, as in (1), and having an infinite number of context-dependent enrichments as in (2) and (3). Parsimony, where is thy victory?

1.2. Highlights from the Ambiguity Debates 1.2.1. Logical Form and Propositional Representation In order to stress the drawbacks of postulating ambiguities, I would like to begin with the opening passage of Empson, Seven Types of Ambiguity: "In a sufficiently extended sense, any prose statement could be called ambiguous. In the first place it can be analysed. Thus, The brown cat sat on the red mat' may be split up into a series: 'This is a statement about a cat. The cat the statement is about is brown', and so forth. Each such simple statement may be translated into a complicated statement which employs other terms; thus you are now faced with the task of explaining what a 'cat' is; and each such complexity may again be analysed into a simple series; thus each of the things that go to make up a'cat' will stand in some spatial relation to the 'mat'. 'Explanation', by choice of terms, may be carried in any direction the explainer wishes; thus to translate and analyse the notion of 'sat' might involve a course of anatomy; the notion of 'on' a theory of gravitation." Empson (1930: 1).

6

Discourse, beliefs, and intentions

Chapter 1

The debate concerning the relative limits of semantics and pragmatics exemplified by the heterogeneous set of examples including quantifiers, negation, conjunction, disjunction, and implication, has been brought to the fore and maintained by arguments of Wilson 1975; Kempson 1975, 1977, 1979, 1986; Zwicky and Sadock 1975; Sadock 1984; Atlas 1977, 1979, 1989; Kempson and Cormack 1981; Sperber and Wilson 1986a; Carston 1988,1994, 1996; Recanati 1989b, 1993,1994; van Deemter and Peters, eds, 1996, to name only a small representative sample. The commitment to sense-generality varies within this group and seems to be lower-bounded by an attempt to explain away unwanted ambiguities (Grice 1978; Horn 1972,1985,1988,1989; Levinson 1988; Atlas and Levinson 1981). First, a terminological clarification is due. The terms logical form, semantic representation and proposition have been used with various shades of meaning and scope. Logical form has undergone an interesting transformation since the introduction of intensional logics to linguistics in the 1970s, primarily by Montague and Hintikka (see Lycan 1984: 7-15). Sentence meaning was more and more often analysed in terms of possible worlds. Logical forms written in terms of intensional logic resemble the representations of surface grammar. Their raison dietre is to provide an unambiguous representation of the sentence. This representation makes it possible to specify what the world would have to be like for the sentence to be true and what it would have to be like for it to be false. In other words, logical forms give access to the theory of truth for the language. This theory of truth is founded on Tarski's so-called 'convention T': it must contain instances of 'jc is true if and only if/?', where 'x' is replaced by a name that describes the logical form of a sentence V of the object language, and 'p' is replaced by's' (cf. Lycan 1984: 15; adapted from Harman). In other words, to follow Davidson's formulation, to know the meaning of a sentence is to know the conditions under which the sentence would be true. A theory of truth constitutes a successful theory of meaning for Davidson (cf. e.g. 1984a), all we have to do is assign truth conditions to every sentence. There are two main problems with this account and both have to do with the meaning of 'meaning'. First, if 'means that' equals 'is true if and only if (henceforth: iff), then not every declarative sentence will license such a substitution.1 The logical form may have to be made more precise before truth conditions can be assigned. This was the case with examples (l)-(3) above. Second, the psychological plausibility of the move of equating the theory of meaning with the theory of truth is somewhat strained. In Putnam's (1975b, see also Section 3.7 below) example of H2O and its Twin-Earth equivalent which has the chemical composition XYZ, the sentence (4) corresponds to the same content in the head of the Earthling and the Twin-Earthling.

1

Problems with imperatives and interrogatives will be left aside here.

Section 2

(4)

Semantic ambiguities and semantic underspecification

1

Water is wet.

So, meaning is not in the head and semantics is not part of psychology. We require a theory of truth which will tie sentences with facts. Solipsistic semantics will not do (cf. Lycan 1984: 240). An independent motivation for this requirement comes from the fact that natural languages are normally public, are shared in a linguistic community and one needs a common tool to talk about the content of our propositional attitudes. Hence, Davidson's realism about truth seems to win over methodological solipsism. However, the difficulty with the psychological significance of this move has to be emphasized. The more so that if to know the meaning is to know the truth conditions of a sentence, then, as Davidson concludes, speakers would have to have tacit knowledge of these truth conditions. And yet there are many undecidable sentences, to mention only (5). (5)

A city will never be built here.

(from Lycan 1984: 251). So, if meaning cannot always be explained with truth conditions, it presumably cannot be equated with truth conditions either. The first difficulty is attended to by the followers of semantic underspecification in distinguishing semantic representation/logical form from the enriched propositional representation. In fact, it was also attended to by Grice (1975: 25) who recognized a limited amount of pragmatic processing that is necessary before truth conditions are assigned. This processing consisted of reference assignment and disambiguation of grammatical ambiguities. So, propositional representation differs from the semantic representation/logical form in that it is enriched by pragmatic processing. The second difficulty, i.e. the psychological plausibility of the level of logical/semantic form, is not successfully attended to. There is no certainty that hearers go through the stage of an underspecified semantic representation which they subsequently fill in by pragmatic processing. The conflating of these two levels seems then a justified hypothesis worthy of closer investigation. Hence, as a preliminary terminological distinction, I shall try the following. Semantic representation and logical form are not distinguished for the purpose of utterance interpretation as separate levels, but neither is propositional representation. All of the terms are needed, though. Whereas in utterance interpretation we talk about semantic representation, or, in DRT terms, discourse representation, in analysing sentences we talk about logical forms as the output of grammar. Hence, there can be more than one logical form corresponding to the sentence but only one logical form qua semantic representation corresponding to an utterance. This theoretical setup prevents one from saying that a sentence can be semantically ambiguous. Semantic ambiguity means a commitment to the effect that a

8

Discourse, beliefs, and intentions

Chapter 1

sentence has more than one semantic representation and this commitment is rejected in agreement with the underspecified-semantics theorists.2 And yet, no commitment to this underspecificity is made here. Logical ambiguity, on the other hand, is used with respect to sentences which allow for more than one representation in the language of first-order logic. This term does not involve any epistemological commitment: the fact that a sentence can be ascribed a logical form does not yet mean that it has this logical form. Kempson (1986: 79) says that"... a sentence is ambiguous if with respect to a single set of circumstances it can be both true and false". This is not yet a semantic ambiguity, although we can freely call it an interpretative ambiguity. 'Interpretative ambiguity' is intended as a loose, non-theoretic term (as far as it is possible) meaning a difficulty with assigning a unique interpretation to an utterance, at least when taken out of context. Semantic form and logical form have to be distinguished for yet another reason. On this account, semantic form, when enriched by pragmatically derived information, remains a semantic form. Calling it by a different name (prepositional form) would signal a psychological reality of these two levels and force one to treat them as consecutive stages in utterance interpretation. This is too important a commitment to be accepted without support, and in fact the onus of proof seems to be on the side of the dualists. Hence, instead, we shall entertain the possibility that semantic representation can be enriched and remain semantic representation, in contrast to logical form where talking of any enrichment would be nonsensical. LF is tied to the grammar of the sentence whereas semantic representation is tied to utterance interpretation. Preserving this distinction is a remedy to radical pragmatics which, in Sadock's words, confuses functional (pragmatic) descriptions of language with its structure, and attempts, quite mistakenly, to choose between the two:

"... the radical pragmatist would be like an anatomist who, realizing that birds fly, loses all interest in the structure of their wings." Sadock (1984: 142).

Now, what does it mean to say that semantic representation remains semantic when enriched? In a coarse generalization, it means that there are some general rules that make one of the interpretations unmarked, default, arrived at prior to (and instead of) any other interpretations. Independent support for the theory of semantic defaults constitutes the rest of what follows. 2

See Jaszczolt 1999.

Section 2

Semantic ambiguities and semantic underspecification

9

7.2.2. Ambiguity Tests Although underspecified semantics ended up conflating Russellian ambiguities of the King of France type with bonafide underdetermination, there is no shortage of criteria for telling ambiguity and underspecification apart. But first, some more terminological clarification. Underdetermined sense and underdetermined semantic representation were explained above with reference to example (1). When a sentence has two readings and the explanation of these two readings cannot be provided by two independent logical forms, then there is no ambiguity but rather some other difficulty which is of an interpretative nature and which we tentatively called interpretative ambiguity in 1.2.1. Zwicky and Sadock in their seminal paper on ambiguity tests (1975) define ambiguity in precisely this way, namely as having several underlying semantic/logical representations. This is opposed to generality in the case of which the sentence has one single representation corresponding to different states of affairs. The terms 'generality', 'vagueness', 'indeterminacy', 'nondetermination', 'indefiniteness of reference', 'neutrality', 'unmarkedness', 'lack of specification' (ibid.: 2) are treated as names for essentially the same phenomenon. Over two decades later, terminology still gives rise to confusions3 but some order seems to be emerging: sense-generality seems to signal the epistemological commitment to the intrusion of pragmatic inference into the truth-conditionally relevant representation (cf. Atlas 1989), underspecification is used with reference to the information carried by logical forms (cf. van Deemter and Peters, eds 1996), indeterminacy and underdetermination tend to be used when utterance interpretation as performed by the hearer is analysed (cf. Bach 1994a). But this is only a rough intuitive guide and the differences between the terms do not seem to carry any theoretical load. Zwicky and Sadock (1975: 2-3) contrast (6) and (7) with (8). (6)

They saw her duck.

(7)

He cooked her goose.

(8)

My sister is the Ruritanian secretary of state.

The first two are instances of ambiguity, whereas (8) involves a lack of specification as to the properties of the sister: Is she younger or older than the speaker? When did she take up the post? And so on, perhaps ad infinitum. All of these examples involve differences in understanding and hence this difference is only a necessary but not a sufficient condition As a striking example of this see Zhang 1998.

10

Discourse, beliefs, and intentions

Chapter 1

for ambiguity. Zwicky and Sadock proceed to discussing various types of ambiguity tests. The Semantic Difference test requires that the readings of the sentence have very little in common for the sentence to count as ambiguous. The test of Contradiction is more precise: 'dog' is ambiguous because 'dog' as a species stands in privative opposition4 to 'dog' as a male representative of the species. Inconstancy Under Substitution of semantically related lexical items also signals ambiguity. They add to this syntactic tests such as an argument from having two syntactic structures, disambiguation by transformation (e.g. 'Her duck was seen by them' as passivization applied to example (6)), and, perhaps most importantly, the Identity Test. According to the latter, conjunction reduction should be possible only when the conjoined constituents have matching understandings, as in (10) which is a result of the application of this test to (6) and (9). (9) (10)

They saw her swallow. They saw her duck and (her) swallow.

(from Zwicky and Sadock 1975: 18). Crossed understandings do not arise for ambiguous sentences, while they are freely permitted for underspecification.5 None of the tests is reliable, though. Identity tests prove to be too strict for some types of constructions, including privative oppositions, neither do they work for some adverbials, metaphors, irony, jokes, or exaggeration. Crossed understandings are sometimes excluded although there is no ambiguity involved and this fact is attributed to the interference of pragmatics.6

4

Privative opposites are terms joined by the relation where one understanding includes a specification which the other lacks. 5 This test has close affinities to G. Lakoff 's (1970: 357) test using the following type of reduction: '... and so did George','... and the same goes for George','... but George wouldn't do so'. 6 This point was elaborated further e.g. by Lascarides et al. 1996 who claim that word punning effect in conjunction reduction (zeugma) is not a satisfactory test for lexical ambiguity. Some examples of juxtapositions of two senses work better than others, which suggests that there is no ambiguity involved; the punning effect may be caused by syntactic or pragmatic factors as well as lexical-semantic. The authors allow for the 'pragmatic ambiguity' instead, perhaps better called underdetermination of coherence or an uncertainty of the rhetorical connection. However, as I demonstrate throughout Chapter 2, pragmatic ambiguity is an unjustified postulate and can be avoided by discovering rules according to which default interpretations work. The analysis of rhetorical structure of discourse falls short of providing an explanation why certain interpretative rules are followed rather than others. Intentionality of mental states and a fortiori of linguistic expressions provides this explanation.

Section 2

Semantic ambiguities and semantic underspecification

11

1.2.3. Ambiguity of i f . . . then? The early history of the ambiguity debate also features solutions to conditional perfection where a conditional ('if.. .then') is frequently taken to mean a biconditional ('if and only if) as in (11) and (12). (11)

If you mow the lawn, I'll give you five dollars.

(12)

If and only if you mow the lawn will I give you five dollars.

(van der Auwera 1997: 169, after Geis and Zwicky 1971). Sentence (12) has an equivalent in a more common idiom of (13). (13)

I'll give you five dollars just in case/only if you mow the lawn.

This enrichment does not signal an ambiguity of implication in English. Geis and Zwicky claim that (11) invites an inference to (14) and this inference can be analysed as a straightforward Gricean implicature, triggered by Grice's (1975) Co-operative Principle.7 (14) 7

If you don't mow the lawn, I won't give you five dollars.8

The Co-operative Principle: "Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged." Grice (1975: 26). The Maxims of Quantity: 1. Make your contribution as informative as is required (for the current purposes of the exchange). 2. Do not make your contribution more informative than is required. The Maxims of Quality: Try to make your contribution one that is true: 1. Do not say what you believe to be false. 2. Do not say that for which you lack adequate evidence. The Maxim of Relation: Be relevant. The Maxim of Manner: Be perspicuous: 1. Avoid obscurity of expression. 2. Avoid ambiguity. 3. Be brief (avoid unnecessary prolixity). 4. Be orderly. Grice (1975: 26-27). 8 Van der Auwera discusses conditionals that refer to states of affairs rather than epistemic or speech act conditionals (respectively) of the following sentences (1997: 170): (i) If she's divorced, she's been married, (ii) John has left, in case you haven't heard.

12

Discourse, beliefs, and intentions

Chapter 1

Now, these inferences are pragmatic strengthenings which do not warrant semantic ambiguity. Van der Auwera (1997: 172) uses the idea of scalar implicature (see Section 1.4 and Levinson 1983: 132-147) to account for this strengthening. His scales, however, are somewhat idiosyncratic. They contain whole sentences rather than a list of relevant expressions arranged from the strongest to the weakest, because, as he says, pragmatic scales cannot rely on units smaller than a sentence. Sentences (15) and (11) constitute such a scale: (15)

Whatever may be the case, I will give you five dollars.

(11)

If you mow the lawn, I'll give you five dollars.

The fact that (15) could have been used but was not, constrains the interpretation of (11). The reconstruction of the interpretation of a sentence with a conditional looks as follows: if p, q, implicates that not: if p, q and if r, q; if p, q and if s, q; etc.9, that is only if p, q. In neo-Gricean pragmatics, this verbal parsimony is explained by Horn's (1984) Principle of Relation and Atlas and Levinson's (1981) Priniciple of Informativeness which limit the conversational contribution to what is necessary, i.e., to put it very crudely, to saying no more than one must. But resorting to informativeness is not very interesting because the sentence which is higher on the scale (here, (15)) relies on the 'only' implicature spelled out as (13). 'Only' implicatures are derived from scales and from the principle of Quantity, saying that the speaker says as much as he/she can, given the counter-balancing effect of the Principle of Relation. For example, the existence of the quantifier 'all' strengthens the meaning of 'some' to 'some and only some, not all'. In terms of the above reconstruction, we can say that 'if is strengthened to if and only if p, q, not if r, q; if 5, q; and so on. Be that as it may, implication in English is not ambiguous, it is pragmatically conditioned. The speaker said as much as he/she could by indicating that/? (here: mowing the lawn) will ensure the hearer five dollars. According to van der Auwera, no other activity will ensure this income because if it did, it would be suggested by the speaker. According to Horn and Levinson no other activity will ensure this income because this particular activity was selected as salient and because this is how 'if' tends to work. The intricacies of this difference will not be relevant until Section 1.4. Cf. van der Auwera (1997: 172-173) for acknowledgements to Horn, Ducrot and Cornulier.

Section 2

Semantic ambiguities and semantic underspecification

13

1.2.4. Ambiguity of Negation? The identity test, and in particular Lakoff's version of it, is sometimes overly stretched to fit the current dogma. Atlas (1977,1989) argues for the so-called Atlas-Kempson Thesis, that is that sentences of the form 'The A is not B' are not ambiguous but semantically general, non-specific or vague.10 It is so because the narrow-scope predicate-negation entails the wide-scope sentence-negation. Now, Lakoff's test is used in sentence (16): (16)

The King of France is not wise, and the same (thing) goes for the Queen of England.

(from Atlas 1977: 326; 1989: 76). Atlas allows here for the crossed understanding and claims that (16) is not anomalous. Atlas's intuitions on this point are by no means universally shared. If they were, and if in addition the test was decisive, he would have a powerful argument against semantic presupposition. For the presupposition that the King of France exists to be present, negation has to be ambiguous and the test would clearly show that it is not. Facing the sea of uncertainties, Atlas maintains that the propositions corresponding to the sentence of the form A is not B' are not its meanings; the sentence has one meaning and two uses. Now, whether the logical form of the sentence corresponds to the wide scope negation remains controversial. Atlas (1989: 69) concludes that sentences of the type 'The F is G' and 'The F is not G' do not have traditional Russellian logical forms and their semantic representations are not logical forms. If we were to go one step further and put forward a hypothesis of defaults in interpretation, the discrepancy between the logical form and the semantic representation would not have to be preserved and parsimony would win. The referential reading of definite descriptions would be produced automatically and so would the narrow-scope predicate-negation. But all we can do at this stage is signal the hypothesis and point out its methodological strength by appealing to MOR and POL, also maintaining the definitions of semantic representation and logical form put forward in Section 1.2.1. As was remarked earlier, the Russellian logical forms of sentences of the form in (1), presented in (la) and (Ib) above, reflect the difference in the scope of negation but cannot be treated as representations of the two readings: the wide-scope negation is compatible with both readings. Moreover, these logical forms are not utilized as steps in utterance interpretation. The speaker uses the sentence with the referential or the attributive understanding and Atlas (1977) attributes these terms to Quine, Zwicky and Sadock, and G. Lakoff, respectively.

14

Discourse, beliefs, and intentions

Chapter 1

these are not represented in the logical forms (la) and (Ib). The referential reading assimilates the description to the status of a proper name and, for our purposes, can be represented as (Ic) where 'k' is an individual constant standing for the person who is the king of France (cf. Jaszczolt 1999). (Ic)

- Bald (k)

Russell's account does not provide for this reading. As it in addition places a strong requirement of uniqueness on the described entity, it is generally regarded as untenable. The representation in (Ic) is theoretically important. It is also frequent in the practice of utterance interpretation. This referential reading seems to have a different status from the wide-scope reading. The suggestion that this interpretation is the default one conforms to the intuition that the sentence is problematic: either outright false, or neither true nor false due to not having a referent. In a sense, the scope of negation comes later. Unlike the interpretation in (Ic), the attributive reading with the narrow scope of negation has to be justified by contextual information, and so does the attributive reading with the wide scope of negation. The intuition seems to support the conjecture that the readings of (1) do not have equal semantic standing. Hence, looking for supporting arguments for default readings is justified, and so is the search for default semantics which, by MOR, if true, would be the best solution. To sum up, the semantic ambiguity of negative sentences is a standpoint represented by several philosophers and linguists such as Russell, Karttunen and Peters, and proponents of three-valued logics. It has its respectable roots in Aristotle's work. But the dominant position nowadays is that of unitary semantics for negation, which is a truth-functional or at least sense-general operator (cf. Horn 1989: 362). The rejection of ambiguity, as well as semantic presupposition and truth-value gaps, can be attributed to Atlas (1977, 1979, 1989), Kempson (1975, 1986), Gazdar (1979), Carston (e.g. 1996), and many others. Semantic presupposition has been successfully accommodated to pragmatic presupposition or entailment and implicature (see Horn 1996). Now, Horn (1989: 308-309) argues that humans have a tendency to assign a lower-clause reading to a higher-clause negation. For example, (17) normally requires a reading of (17a) rather than (17b). This is a psychological trait, generalized in syntax as a Neg-raising phenomenon. (17)

X does not believe that p.

(17a)

X believes that not-/?.

(17b)

It is not the case that x does believe that p.

Section 2

Semantic ambiguities and semantic underspecification

15

In other words, contraries become assimilated to contradictories: p v q becomes reinterpreted in speaker's mind as p v-ip, i.e. obeys the law of excluded middle. In other words again, if something is not good, it is bad; there is a tendency towards simple contrast in our psychology. From the position of the hearer, then, contradictories are read as contraries. As Horn remarks, the embedded placement of negation may not be available at all, as indicated by the pair of French sentences in (18a) and (18b). (18a)

Je ne veux pas que vous sortiez. I don't want you to leave.

(18b)

??Je veux que vous ne sortiez pas. I want you not to leave.

(from Horn 1989: 311). Since the lower-negation meaning entails the higher-negation one, we have an instance of privative opposition here, of Zwicky and Sadock's 'dog-canine' vs. 'dog-female' type. And in the case of privative oppositions, ambiguity is difficult to maintain. The strengthening from the contradictory to the contrary performed by the hearer is said to be pragmatic, i.e. an instance of an implicature. It occurs frequently with 'weak', non-factive propositional attitude verbs, such as 'believe' or 'want'.11 This implicature does not have to be arrived at in utterance processing but rather is the default understanding (or, in the common jargon, the implicature is 'short-circuited'12 and becomes a convention).13 The conclusion drawn above that neither the sense-generality nor the ambiguity accounts are satisfactory is not new. Horn signals clearly that both are partially correct. "Both views contain much insight and some truth, yet both are incomplete. While two distinct uses of sentential negation must indeed be admitted, the marked, nondescriptive variety is not a truth-functional or semantic operator on propositions, but rather an instance of the phenomenon of METALINGUISTIC NEGATION - a device for objecting to a previous utterance on any grounds whatever, including the conventional or conversational implicata it potentially induces, its morphology, its style or register, or its phonetic realization." Horn (1989: 363). 11

As opposed to 'claim', 'regret', 'know'. Cf. Horn (1989: 325-6; 345). The term was first used by Morgan. See Horn (1988: 138). It can also be seen as a result of application of Horn's R principle (Principle of Relation) mentioned briefly in Section 1.2.3 and discussed in more detail in Section 1.4 below. 12

13

16

Discourse, beliefs, and intentions

Chapter 1

Although there are two types of it, negation is not semantically ambiguous. Instead, there is a "built-in duality of use" (Horn 1985: 132), or pragmatic ambiguity. Context helps decide whether negation is descriptive or metalinguistic. Metalinguistic negation, a notion revived from Ducrot, does not fare well with the positions of Kempson, Wilson-Carston, Cormack, or Burton-Roberts, called by Horn the London School of Parsimony (see Horn 1989: 432). The latter group see negation as truth-functional: it is the one-place propositional connective of Fregean semantics. The differences of interpretation are said to contribute to the propositional form by means of pragmatic processes. Horn opts for what he calls the Mill-Grice view. By producing metalinguistic negation, the speaker "... conveys an unwillingness to assert a sentence that would induce a misleading implicatum, even though this sentence would be true under these circumstances." Horn (1989: 384). Metalinguistic negation is then submitted to a battery of tests, e.g. from the behaviour of morphologically incorporated negation, polarity items, and the behaviour of 'but' (see ibid.: 392-413).14 Its main advantage over the sense-generality position is supposed to be eliminating the multiplicity of propositions that can be engendered by various readings of negation: the multiplicity which is, in fact, a (semantic or pragmatic) ambiguity in disguise. Ambiguity disappears for negation, only to reappear on the level of propositions. However, whether there are strong advantages of postulating another negation operator for English is not certain. After all, ambiguity reappears on the level of sentential operators and calling it pragmatic rather than semantic will not yet get rid of it. Finally, the difference between the position of metalinguistic negation and that of truth-functional negation operating on enriched propositions (Carston 1996) is not as important as it is claimed to be. Carston says that negation is truth-functional but operates on 'echoic use' of language. For example, the behaviour of polarity items in (19a) and (19b) signals that in (19b) material is 'put in quotes' and then negated: the negative polarity item 'ever', otherwise standard in such sentences, is not present. (19a) is an instance of descriptive negation, i.e. descriptive use of the material in the scope of the negation operator. 14

Cf. (i) and (ii) of Horn (1989: 408). Only the 'A is not X but Y'construction is admissible for metalinguistic negation. (i) He's not tall, but very tall. (ii) *He's not tall, but he's very tall. But, as McCawley (1993: 190) observes, (i) is standardly used for contrastive negation and hence the duality of the operator is dubious, as his example in (iii) demonstrates (p. 190). (iii) John drank not coffee but tea.

Section 2

Semantic ambiguities and semantic underspecification

(19)

Mary is sometimes late.

(19a)

She isn't ever late; she's always punctual.

(19b)

She isn't sometimes late; she's always punctual.

17

(from Carston 1996: 324, my emphasis). We are still left with descriptive or echoic negation, redefined as the 'duality of use'. This time the focus of explanation falls on the negated material, i.e. the use of a proposition. The major weakness of this account is obvious. Negation is truth-functional on one hand, but operates on propositions which can be enriched by just about any type of information: intonation, register, grammar, non-linguistic communication, and so on, perhaps ad infinitum. And, "... when we bear in mind what a truth function must be a function of, we recognize the implausibility in the view that negation is invariably truth-functional." Horn (1989: 434). All in all, both solutions eradicate ambiguity from semantics and both admit a duality of use: one assigns it to the negation operator, and the other to the prepositional content of the negated material. As McCawley (1993: 204) rightly observes,'metalinguistic' seems to be & function of negation rather than its form and it is only natural to employ the negation operator when one wants to convey objection to the proposition on any grounds. We can conclude that semantic ambiguity is either still there in a new disguise, or dispensed with at the cost of opening the Pandora's box of heavy context-dependence. In neither case is sense-generality necessary. A different attempt to eradicate ambiguity has been very recently proposed by Geurts (1998: 293, fn 17). He suggests an improvement to Carston's and Horn's proposals and says that negation of the linguistic form of an expression is not just echoic use or metalinguistic negation but rather negation aimed both at 'linguistic objects' and 'objects in the world'. It seems that levels of senses are treated with parsimony here. As a result, negation is not ambiguous. However, it is as yet necessary to wait for further developments of this idea in order to assess its validity. The idea may prove to draw on the common-sensical claim in cognitive linguistics that negation is a cognitive universal.15 Negation is a universal concept which, although further divisible or gradable, does not allow for ambiguities. 15

Cf: "... negativity may be partly conceptualized (...) in terms of spatial/topological relations and image schemas such as BLOCKAGE or RESISTANCE, ABSENCE rather than PRESENCE, LESS rather than MORE, DISJOINING rather than CONJOINING, SPLITTING rather than COMBINING, etc." Lewandowska-Tomaszczyk (1996: 34).

18

Discourse, beliefs, and intentions

Chapter 1

7.2.5. Ambiguity of and ? In addition to implication and negation, sentential conjunction also seems to invite a conclusion of ambiguity. Sentences (20a) and (20b) clearly differ in truth conditions. (20a)

They got married and had a baby.

(20b)

They had a baby and got married.

The test of embedding p&q and q&p under the scope of logical operators such as implication and negation demonstrates it even more clearly, as in example (21): (21)

If the old king died of a heart attack and a republic was declared Sam will be happy, but if a republic was declared and the old king died of a heart attack Sam will be unhappy.

(from Carston 1988:172; cf. Cohen 1971: 54). In addition to the straightforward logical conjunction, the sentential connective and can mean a temporal conjunction ('and then') and consequential or causal conjunction ('and therefore', 'and as a result'). The logical conjunction is exemplified in the standard understanding of sentence (22), and the causal one in (23): (22)

I ate cheese and drank wine.

(23)

I turned the key and the engine started.

As in implication and negation, ambiguity is exorcized here rather consistently across the board, albeit by means of different tests and solutions. The particular meaning of the connective and can be arrived at through pragmatic enrichment. The logical conjunction reading and the temporal/causal, enriched, meaning are not functionally independent, the latter is the development of the prepositional form of the first, and hence, by this criterion (Carston 1988, 1998a), there is no ambiguity.16 Section 1.3 will expose some flaws in this criterion, but this weakness does not change the orthodoxy concerning the unary, non-ambiguous semantic status of and. Since the test of functional independence either does not always work or 16

Nota bene^&nd therefore' is sometimes regarded as parasitic on 'and then'. But the consequential and need not presume the temporal and. States can be fully overlapping and stand in the causal relation to each other as in (i). (i) Susan is underage and can't drink, (from Wilson and Sperber 1998: 6).

Section 2

Semantic ambiguities and semantic underspecification

19

has to be regarded as rather vague and intuition-based (cf. Carston 1998a), Recanati (1989b) proposes an overtly intuitive criterion.17 Recanati agrees with Grice that postulating semantic ambiguity is methodologically unsatisfactory and says instead we should allow for pragmatic processes to contribute to the semantic representation of the utterance. In other words, he subscribes to MOR. But he says that we can only tell whether pragmatic information is necessary as enrichment if we already know what is said. And this is circular. Instead, the enrichments are determined intuitively: "In deciding whether a pragmatically determined aspect of utterance meaning is part of what is said, that is, in making a decision concerning what is said, we should always try to preserve our pre-theoretic intuitions on the matter." Recanati (1989b: 106).18 He calls this suggestion the Availability Principle. Its usefulness for drawing the boundary between semantics and pragmatics need not interest us until Section 1.3. In addition, Recanati revives the so-called Scope Principle suggested by Cohen in 1971 (see example (21)) and accepts the following: "Scope Principle: A pragmatically determined aspect of meaning is part of what is said (and, therefore, not a conversational implicature) if - and, perhaps, only if - it falls within the scope of logical operators such as negation and conditionals." Recanati (1989b: 114). Sentential conjunction is not so straightforwardly truth-functional, though. A great deal of adjustment of the prepositional representation was needed in order to accommodate the thesis of unitary semantics. Carston (1988, 1994) resorts to the argument from the enrichment of the original logical form (functional dependence of the two prepositional forms) and to the constraint 17

Recanati rejects Carston's Independence Principle because it seems not to work for some examples. For instance, in the example below, talking about a party, B replies to A and the reply carries an implicature which entails what is said: 'Somebody was there'. So, the Independence Principle does not work (adapted from Recanati 1989b: 111). A: If anybody was there, Jim was there. B: Somebody was there - I saw John going there. Implicature: Jim was there. 18 So, what is said is consciously accessible - and in delimiting it we appeal to common sense, as some experiments seem to prove, cf. Gibbs and Moise 1997. This conclusion is undermined by another set of experiments, cf. Nicolle and Clark 1999. Also, we must allow for speakers' intuitions to be fuzzy at times.

20

Discourse, beliefs, and intentions

Chapter 1

on this enrichment coming from relevance theory. Sperber and Wilson's (1986a) two counter-balancing principles in communication that keep the processing effort and the cognitive effect in tandem, ensure that the effort of processing the utterance is not unnecessarily great but rather is constrained by the informative value of the communicated assumptions: any expenditure must be justified. In such processing, we are said to utilize knowledge of stereotypical situations, frequently used scripts (Carston 1994: 697). I shall say more about the role of this principle in Section 1.4.2 (see also e.g. Wilson and Sperber 1986a).

1.2.6. Ambiguity of Number Terms? Numerals are also potential candidates for semantic ambiguity Sentence (24) can correspond to the state of affairs in which Max has three or more children, or only exactly three, as the respective developments in (24a) and (24b) demonstrate: (24)

Max has three children.

(24a)

Max has three children, and in fact he has four altogether.

(24b)

Max does not have three children, he has four.

It is commonly held that e.g. 'three' in (24) is logically bound to mean 'at least three' and pragmatically bound to mean 'at most three' or 'exactly three' (cf. e.g. Horn 1985; Levinson 1988; Kempson and Cormack 1981). On the other hand, the sense-generality account allows numerals a weak, underdetermined sense and accommodates the natural, upper-bounded 'at most' understanding of (25). (25) She can have 2000 calories without putting on weight. It also accommodates the necessity of the punctual 'exactly «' meaning of 'three' in (26) which would be nonsensical if 'three' meant 'at least three'. (26)

More than three people came.

(both examples from Carston 1998b: 204). However, sense-generality is not the most successful option. Punctual semantics for numerals ('exactly «') seems to work at least equally well, and in addition it seems to be more in the spirit of MOR by avoiding the postulate of an additional level of generality. First, if n had the semantics of 'at least «', then 'at most' and 'at least' would not have symmetrical

Section 2

Semantic ambiguities and semantic underspecification

21

functions: 'at most' would mean the numbers below and including the stated number, and 'at least' would suspend the implicature derived from Grice's first maxim of Quantity (or its neo-Gricean derivatives, see Section 1.4.1 below). Sentence (24) also fares happily with the punctual semantics, and so do sentences with obvious collective readings such as (27) where downward entailments do not arise: (27)

Three boys carried a sofa up the stairs.19

Distributive readings may not carry downward entailments either, as in example (28). And yet, sentence (28) has upward entailments: failing four or five times will also exclude the addressee. (28)

If you fail three times, you're excluded.

(both examples from Koenig 1993:143). Punctual semantics is best suggested by sentences expressing measurements such as (29) and (30) which do not seem to have downward or upward entailments, although judgements here may vary: it is possible to think up situations where the entailments will hold. (29)

John weighs 200 Ibs.

(30)

This book costs $20.

(both examples from ibid.: 144). Koenig argues that punctual semantics is also a satisfactory point of departure for interval interpretations. If we interpret 'three' in terms of a set of cardinality 3, downward entailments will follow anyway: the predicate holds for each element of the set (cf. ibid.: 146-147). Greater informativeness is tied to the property of being a proper subset of a set, so Grice's submaxim "Make your contribution as informative as is required" (cf. 1975: 26) is satisfied for the set reading.20 "Submaxim of Quantity: If the truth set of a sentence A is a proper subset of the truth set of a sentence B, A is more informative than B." Koenig (1993: 147). So, uttering (24) implicates that Max does not have four children. Finally, examples involving the relative scope of numerals also seem to point to punctual semantics. Sentence (31) is said to have four readings (Kempson 19

(27) does not entail that two or one boy carried the sofa upstairs: they carried it together. The relevant neo-Gricean principle, Horn's Q Principle is also satisfied. It essentially says 'Say as much as you can (given that you shouldn't say more than you must)'. See Section 1.4.1. 20

22

Discourse, beliefs, and intentions

Chapter 1

and Cormack 1981: 267). The two most obvious ones are listed as (31a) and (31b). The collective meaning in (31a) appears to be more natural than the other. In (31a), 'between them' sounds redundant, whereas 'each' in (31b) is informative. But this judgement, like many other intuitions on ambiguity, has a very limited use. In some contexts (31) will perfectly well be taken to mean (31b), for example as an answer to (32): (31)

Two examiners marked six scripts.

(31a)

Two examiners marked (a set of) six scripts between them.

(31b)

Two examiners marked (some or other) six scripts each.

(32)

How many examiners managed to mark more than five scripts in a day?

The set reading of (31) consists of a group of exactly two people marking a group of exactly six papers. By Koenig's Submaxim of Quantity, this is the most plausible reading. Also, intuitively, (31b) only makes sense as an answer to (32) if 'six' means 'exactly six'.21 Permutations of scope and collective and distributive readings also render other, more context-dependent, possibilities (see ibid.: 267). Other natural language quantifiers are also potential candidates for semantic ambiguity. 'Some' can mean 'some but not most/all' or 'some and perhaps most/all', that is it can be bilaterally bounded (cf. 'exactly «') or lower-bounded, respectively. The analysis of 'some' does not quite follow that of the number terms, though. Scale-reversal to 'at most' (as in sentence (25)) does not seem to occur (Carston 1998b: 211) and the 'not all' meaning may well be a mere implicature. However, intuitions do not seem to be reliable here: we can envisage a situation is which 'some' means 'some or less'as in (33): (33)

Some small bars of chocolate are allowed every now and then without making you put on weight.

Examples of the type (25) and (33) should not be relied on too heavily: there is a sense of negation or denial in this type of construction ('You mustn't eat more than... if you don't want to get fat.') and hence the implicature may be derived from the construction as a whole or from the verb 'allow' instead. 21

Kempson and Cormack adhere to the 'at least' interpretation and an underspecified semantics. Cf.: "We wish to say (...) that [31] is true if for at least two examiners there is some relationship between them and at least six scripts that were marked but without specifying that there is a necessary relationship between any particular members of the set of examiners and any particular members of the set of scripts." Kempson and Cormack (1981: 270).

Section 2

Semantic ambiguities and semantic underspecification

23

In fact, it has also been claimed that different types of expressions allow for different semantics: some default to the 'exactly' meaning and some to 'at least'. Geurts (1998) ascribes this alleged state of affairs to the 'semantic transfer'of meaning which may result in lexicalization. For some expressions, "... the 'exactly' sense is primary, while in others it is the 'at least' sense which is basic. Into the first category I would place cardinal expressions and perhaps determiners like a few, while adjectives like warm, bad, and intelligent would go into the second category." Geurts (1998: 296). So, 'five books' means 'exactly five', while 'an intelligent woman' is not incompatible with her being a genius. Expressions under discussion are standardly called scalar in that an assertion of a weaker expression, e.g. 'warm' implicates that the stronger does not hold, e.g. 'hot' (see Section 1.4 below for a discussion of scales). Now, postulating different default senses for different expressions may seem a costly move. Geurts convincingly argues, however, that the lexicon allows for such varieties of senses and we must not expect uniformity with respect to the 'at least' and 'exactly' meaning. There is nothing methodologically unparsimonious with simply admitting that this is what the language is like. But it seems doubtful that predicates really conform to Geurts's classification. Unitary negation and lexeme-specific precision of meaning seem difficult to reconcile on the basis of arguments such as that denying that it is warm does not mean that it is hot, whereas denying that one has five sisters may mean that one has more than five. More about it in Section 2.6 below.

7.2.7 Ambiguity of Definite Noun Phrases? Throughout this chapter I have been discussing and in effect contesting the idea that the underspecified semantic representation always functions as a cognitively viable step in the analysis of interpretatively ambiguous expressions. The MOR indicates that semantics is uniform and is supplemented with accounting for any differences in meaning by means of pragmatic principles.22 Definite descriptions, that is definite noun phrases in sentences of the type (34), allow for two interpretations: the referential one which makes them akin to proper names (about a particular person, e.g. 22

Gricean, neo-Gricean, or relevance-theoretic, to be discussed in more detail in Section 1.4.

24

Discourse, beliefs, and intentions

Chapter 1

Mr Jones), and the attributive one which makes them about whoever or whatever satisfies the description (about whoever Smith's murderer may be. See Donnellan 1966; Neale 1990). (34)

The murderer of Smith is insane.

The ambiguity of reading is best seen as an interpretative ambiguity. It has been regarded as pragmatic (see e.g. Kripke 1977), although the exact placement of the pragmatic component differs among current approaches (cf. Schiffer 1995a). According to the sense-generality account, definite descriptions have a unique, underspecified semantic representation which is further pragmatically enriched into a prepositional form before being assigned a truth value. But postulating indeterminacy here is equally unsatisfactory as maintaining semantic ambiguity. Before resorting to indeterminacy, it has to be investigated whether both readings have equal semantic status. If one is unmarked and more basic than the other and if this unmarkedness manifests itself on the level of semantics, sense-generality is a redundant level of interpretation. This markedness is investigated in Section 2.5.3.

1.2.8. Ambiguity of Indefinite Noun Phrases? Indefinite noun phrases are mainly used as indefinite descriptions. The latter are sometimes claimed to have even more readings than definite descriptions. For instance, sentence (35) is frequently said to have a specific reading which is postulated as a reading intermediary between the referential and the attributive. It describes the case where the speaker does not intend the hearer to individuate the person (cf. Ludlow and Neale 1991; Larson and Segal 1995). (35)

A friend of mine from Oxford paid me a visit.

The use of an indefinite signals that the identity of the person spoken of is not relevant for the purpose of the conversation. However, indefinites do not seem to exhibit semantic ambiguity. The use of an indefinite serves the purpose of saying either that [1] the identity of the referent is unknown to the speaker or [2] it is known to the speaker but irrelevant for this particular discourse either because the hearer would not know the referent anyway, or because whoever it is, it is only the indefinite description that matters. Therefore, rather than postulating manifold ambiguities, MOR and POL suggest two interpretations: specific and nonspecific, which is, after all, a well acknowledged and established distinction (see Kamp and Reyle 1993). So, indefinite descriptions can be used specifically or

Section 2

Semantic ambiguities and semantic underspecification

25

nonspecifically. The specific use signals that the speaker is talking about a particular, known individual or object and that this individual or object is either taken to be new to the hearer or its identity is irrelevant for the conversation. The nonspecific use signals that the speaker is talking about an unknown individual or object. The difference between the specific and nonspecific use is truth-conditionally decisive in the examples where it triggers the ambiguity of scope as in (36) (from Kamp and Reyle 1993: 279), where on one reading there is one particular problem that preoccupies all politicians, and on the other reading every politician has some problem or other that preoccupies him or her. (36)

A problem about the environment preoccupies every serious politician.

However, the ambiguity of scope seems to be engendered here by the quantified expression 'every politician'. Therefore the argument about the truth-conditional impact of the reading of an indefinite NP is not decisive. Hence, Ludlow and Neale's (1991) distinction into referential, attributive, and specific use of indefinites appears rather contentious. In order to postulate it, the authors had to interweave a change of perspective from that of the recognition of the speaker's intentions by the hearer to that of the speaker's background knowledge, which is methodologically spurious. Attributive and referential readings are presented as straightforward cases of the recognition of the speaker's intentions by the hearer. In the specific use, more than the intention has to be recognized: the hearer knows that the speaker knows the person talked about and that the speaker does not intend to communicate this information. Ludlow and Neale distinguish here the speaker ground, the proposition meant, and the proposition expressed. A different account will be tried in Section 2.5.4, according to which the change of perspective is unnecessary. It is an account of communication in terms of the referential, communicative and informative intentions. After all, the aim is to analyse the process of utterance interpretation and the establishment of meaning by the speaker and not the differences between the speaker's background knowledge and speaker's meaning. The perspective is that of the hearer recovering the speaker's intentions and thus of the language use rather than that of the totality of cognitive processes involved in utterance interpretation. In other words, there are only two readings that matter: specific and nonspecific seen as intended as specific and nonspecific. The totality of the speaker's thoughts and attitudes is irrelevant. The confusion is still rife in the literature but it is important not to run the two perspectives together.

26

Discourse, beliefs, and intentions

Chapter 1

All in all, the grounds for ambiguity are even weaker in the case of indefinites than in that of definites. But equally weak are the reasons for underspecification.

7.2.9. Ambiguity of Propositional Attitude Sentences? Sentence (2) of the Introduction, repeated here as (37), is commonly regarded as ambiguous between the de re and de dicto readings.23 A similar question should be asked here: are propositional attitude sentences semantically ambiguous? Are they underdetermined? The common, albeit not fully reliable, intuition seems to be that sentences of the type of (38) and (40) are standardly taken by the hearer to be about a particular, known person: the description in (38) is referential and the belief report in (40) is de re. (38) and (40) would normally trigger a question as in (39): (37)

Ralph believes that the man in the brown hat is a spy.

(38)

The best architect designed this church.

(39)

Really? Who was it?

(40)

Sally believes that the best architect designed this church.

(cf. Jaszczolt 1998d and Klein 1981). This seed of doubt concerning ambiguity and at the same time underspecification is all that is required to trigger an investigation into our third possibility: default, intention-based semantics, to be returned to in Chapters 2 and 3.

1.2.10. Interim Conclusions The conclusion that emerges from this section is that both ambiguity accounts and underspecification accounts have their weaknesses. There is a third position available, that is the one according to which there are default interpretations of seemingly semantically ambiguous sentences. Presenting the case for these defaults will constitute most of what follows. See Introduction and Jaszczolt 1997a, 1998d.

Section 3

Semantic ambiguities and semantic underspecification

27

1.3. What Is Said This section is essentially a review of the current standpoints concerning what is said. The status of what is said is still subject to debate. Grice (1978) suggested that in the case of utterances that exhibit an ambiguity of interpretation which cannot be attributed to the lexicon or grammar, what is said should be regarded as the same in the two readings, whereas the conversational implicatures are different. It has been argued since then that what is said is better seen as semantically underdetermined. So, for instance, the difference between the referential and the attributive reading of definite descriptions is placed on the level of what is said because what is said is underdetermined (Recanati 1989b). The degree to which semantics is reduced and pragmatics enlarged remains in dispute. I have introduced briefly the idea of traditional semantic ambiguity, as well as unitary semantics supplemented by implicatures, and sense-general semantics where pragmatic aspects of meaning contribute to what is said. A distinction is made here between the semantics of natural language and the semantics of the conceptual representation system, with pragmatic factors bridging the gap. Pragmatic factors are understood as contextual information plus the pragmatic principle of co-operation, including an account of speaker's intentions. The process of supplementing the semantic representation is called completion (or saturation) and expansion (or strengthening; Bach 1994b; Recanati 1989b respectively. See also Carston 1988). I shall examine the ideas of expansion and completion, as well as arguments in favour of and against placing them under what is said. It will be concluded in this section that expansion and completion do not warrant sense-generality. As a result, a fourth solution is required and here I shall propose non-ambiguous, unitary semantics achieved through the interaction of the speaker's intention with the logical form of the expression, as put forward in Chapter 2. We have established that the boundary between semantics and pragmatics is difficult to draw.24 To use Lyons's (1987:157) exposition, some of the options can be put as follows:

(1) (2) (3) (4)

SEMANTICS concerns meaning concerns competence concerns conventional aspects of meaning is a matter of rules

Cf. also the discussion in Cole (1981: xiv).

PRAGMATICS concerns use concerns performance concerns non-conventional aspects of meaning is a matter of tendencies, principles, maxims or strategies

28

(5) (6) (7) (8)

Discourse, beliefs, and intentions

Chapter 1

deals with truth-conditional deals with non-truth-conditional meaning meaning; PRAGMATICS = MEANING - TRUTH CONDITIONS deals with literal meaning deals with non-literal meaning concerns the meaning of concerns the interpretation of sentences utterances deals with contextdeals with context-dependent independent meaning meaning

Most of the pragmaticists I have discussed assume (5) and distinguish between truth-conditional semantics and a narrower linguistic semantics which is the output of grammar: word meaning plus the structure in which the words are involved. Unless stated otherwise, semantics will mean here the wider, truth-conditional semantics. Grice distinguished three types of implicature which differ in the degree of departure from the semantic content of the utterance. First, particularized conversational implicature is the furthest from the semantic content as it depends on the situation. Second, generalized conversational implicature is closer to the semantic content, being context-independent. Finally, conventional implicature takes care of the lexical meaning which does not contribute to truth conditions and hence it properly belongs to linguistic semantics. I shall now discuss this distinction between what is said and what is implicated in more detail, beginning with an attempt to define what is said. Kent Bach (1994a) observes that people often speak loosely, non-literally, and it is often more appropriate and efficient to do so. To repeat a familiar example: the mother utters (41) to reassure a child who cut his finger: (41)

You're not going to die, Peter.

(from Bach 1994a: 267). It is unlikely that the child will conclude that he is going to live forever. Why didn't the mother add "... from this cut"? Because it was obvious that this is what she meant. But the point remains that what she said differs from what she meant; she was not speaking literally. She used the sentence non-literally, although she did not use any figures of speech; no part of the sentence was used non-literally. This type of leaving words out is called sentence-nonliterality or 'loose talk'. Similarly, sentence (42) means that the speaker hasn't eaten recently, in the past few hours, rather than ever in his or her life: (42)

I haven't eaten.

(from Bach 1994a: 268). It is quite a common feature of talk that we leave words out and rely on the hearer to infer their meaning.

Section 3

Semantic ambiguities and semantic underspecification

29

In the case of the two examples just mentioned, we say that the sentences have literal meaning, express a proposition with clear truth conditions: that the boy is immortal, and that the speaker has never eaten, respectively. This is called the minimal proposition, expressed by the sentence but not identical with what the speaker meant. The speaker meant an expansion of this minimal proposition, such as these in (41a) and (42a) respectively. (41a)

You are not going to die from this cut, Peter.

(42a)

I haven't eaten breakfast today.

Bach also distinguishes another phenomenon, quite similar to expansion, exemplified in (43). (43)

Steel isn't strong enough.

(from 1994a: 268). This sentence does not express a complete proposition, there is no way of specifying its truth conditions. It is semantically underdeterminate. This is an instance of semantic generality, non-specificity, better known as semantic underdetermination. Naturally, we assume that indexical references are fixed, and ambiguity and vagueness are resolved in all these examples. The meaning of the sentence provides here only a so-called propositional radical, which requires completion to become a full proposition. For (43), this completion involves stating: strong enough for what? Underdetermination differs in this respect from non-literality, where the proposition expressed is complete, but minimal. Here, what the speaker says is the propositional radical, and what the speaker means is the completed proposition. A slogan from Bach (1994a: 269) renders it most succinctly: "... with sentence non-literality a minimal proposition is fleshed out; with semantic underdetermination a propositional radical is filled in." Grice, in distinguishing what is said and what is implicated, did not talk about expansion and completion. But the recognition of the material required for expansion and completion relies on the same pragmatic process as the recognition of the implicature (cf. Bach 1994a: 269-270). So, before the hearer calculates the implicature, he or she may have to flesh out or fill in what is said. This is done by means of the same pragmatic, inferential processes. All that Grice observed (1975: 25) is that we need some pragmatic processes on the level of what is said to assign reference to indexicals and disambiguate expressions. The rest, including expansion and completion, he would classify as implicature. This does not satisfy post-Griceans:

30

Discourse, beliefs, and intentions

Chapter 1

"... there is no line to be drawn between what is said and what is implicated. Instead, there is considerable middle ground between them." Bach (1994a: 270).

We have three options as to the classification of expansions and completions (also known as strengthening and saturation): (i) Gricean, as implicature (Grice 1975); (ii) Relevance-theoretic, as explicit content of utterances (Sperber and Wilson 1986a; Carston 1988); (iii) Bach's, as middle ground (Bach 1994a). For Grice, what is said is closely related to the conventional meaning of the sentence, and hence he has to treat these elaborations as implicatures. For relevance-theorists, these elaborations are not implicatures; they are explicit (cf. Carston 1988), they belong to what is said. Here we can include Sperber and Wilson, Carston, and Recanati. This seems intuitively plausible. Grice's notion of what is said is rather narrow: as what is said is always close to the word meaning and sentence meaning, metaphor and irony are classified by him as "making as if to say" (Grice 1975: 30). So, we have (a) what is said, (b) making as if to say, and (c) implicatures - including even conventional implicatures which do not belong to what is said but to what is meant. And, for Grice, what is said entails what is meant. Hence, he suggests 'making as if to say' instead of saying one thing and meaning another, in the case of metaphor.25 In fact, what Grice calls 'what is said' is better named 'what is explicitly stated', extending the notion of saying to allow for saying that (i) the speaker means what he says plus something else, (ii) says one thing and means something else, or (iii) says something without meaning anything. Now, Bach says that expansion and completion on the part of the hearer result in the hearer constructing a proposition which is richer than the explicit content of what was explicitly stated by the speaker. He calls this enriched proposition impliciture, comparable to the relevance-theoretic explicature, but defined by Bach not as part of what is said: "Implicitures go beyond what is said, but unlike implicatures, which are additional propositions external to what is said, implicitures are built out of what is said." Bach (1994a: 273). 25

Nota bene, for Grice, metaphor is an implicature, which is puzzling, because it is not an 'addition' to what is said (cf. Bach 1994a: 272).

Section 3

Semantic ambiguities and semantic underspecification

31

They constitute the middle ground: they are not implicatures, but neither do they belong to what is said. There is another problem with Grice's notion of what is said. He claims that the essential meaning of certain lexical items constitutes a conventional implicature rather than part of what is said - e.g. in 'but', 'therefore', and perhaps also 'manage'. And it is so because these words do not contribute to truth conditions of the sentence. Bach claims that, in fact, what is said does include the connective 'but': comparing 'but' with truth-functional conjunction is unnecessary. Also, "For example, in 'Bill managed to finish his homework', the truth-conditional content includes both the finishing and the entailed difficulty." Bach (1994a: 289, fn 14). In the case of connectives, we can quite successfully say that they convey some illocutionary act (cf. ibid.: 277). So, the main criticism of Grice's notion of saying is that it is too strong: it does not distinguish between saying something and meaning something. If we weaken the notion of saying, we can include non-literal utterances as acts of saying something, rather than 'making as if to say'. But this is not to deny that saying is closely related to sentence meaning: it should normally be arrived at compositionally, with added disambiguation and reference assignment to indexicals, if needed. All the rest of what we called filling in and fleshing out belongs to the 'middle level' - between what is said and what is implicated. It is this middle level that Bach calls an impliciture. Let us now have a closer look at this middle level of meaning. The minimal proposition in (44) is obviously true. What is meant is stronger, provides more information. In (45), the minimal proposition is obviously false. What is meant is logically weaker. (44)

I have had breakfast.

(45)

Everybody went to Paris.

(from Bach 1994a: 278). This process of fleshing out a minimal proposition (or: expansion, lexical strengthening) amounts to logical strengthening in the first, and logical weakening in the second case, or some other process as in (46) that adds presumed 'before': (46)

I have eaten chocolate [+ before].

32

Discourse, beliefs, and intentions

Chapter 1

(from ibid.: 279). The middle level is not unanimously recognized. To repeat, Recanati (1989b) and Carston (1988), and, generally, pragmaticists working within relevance theory, subsume it under 'what is said'. Recanati calls it 'pragmatic constituents of what is said', and Sperber and Wilson, and mainly Carston, call the enriched proposition an explicature. Bach argues that we need to preserve the middle level in order to be able to distinguish between saying and stating - and between sentence meaning and speaker's meaning. There are also other problems with subsuming the middle level under what is said. Let us imagine that by uttering (42) the speaker means (42b). (42)

I haven't eaten.

(42b)

I haven't ever eaten in my life.

So, the minimal proposition is (42c): (42c)

There is no time prior to the time of utterance at which the speaker ate.

Here we are made to imagine that these words are uttered by a robot in a science-fiction story. Then, what is said would have to differ from one case to another, or, if what is said is the expanded proposition, then what is meant is less than what is said. And this is counterintuitive. So, "... I am not disputing the idea that underlies [the] contention that there are pragmatic aspects to what is said as well as to what is implicated (...), but in my view these aspects are properly regarded as pertaining to what is implicit in what is said." Bach (1994a: 280)26 The category of filling in, i.e. semantic underdetermination, is rather hazy. It is defined as sentences expressing propositional radicals as in (47): (47)

I am too tired

[+ for what?]

Some examples used by Bach (1994a: 286) do not seem to be underdeterminate: (48)

The lamp is old

[+ relative to what?]

(49)

John wants a taxi

[+ to do what with?]

References to Recanati and internal quotation are omitted.

Section 3

Semantic ambiguities and semantic underspecification

33

Sentences (48) and (49) seem to be less plausible examples of the category than 'I am too tired', which is very obviously incomplete.27 So, Bach concludes that Grice's distinction between what is said and what is implicated requires amending. It can be done by adding the middle level of expansion or completion. But his classification of types of expansion and completion is at best ad hoc and controversial (see ibid.: 284-287). As a result, allocating these pragmatic processes to a putative 'middle level' seems to be insufficiently supported. Let us now have a look at the approaches which include the expansion and completion in what is said instead of postulating the middle level. According to Grice, the propositional form results from the conventional meaning of the sentence, as well as disambiguation and reference assignment to referring expressions. But as we know from Bach, propositions may require expanding or completing, so these processes have to be added to those recognized by Grice. Carston (1988) handles the problems with the proposition explicitly expressed (explicature) as follows.28 The discussed additions are not merely implicatures; without the completion, it would not be possible to tell what would make the sentence true or false. Hence, they belong to the explicature (or impliciture). As was mentioned in Section 1.2.5, Carston suggests that explicatures are functionally independent from implicatures, and also that the process of enrichment of the logical form should not be extended unnecessarily so as not to include what is truly an implicature (cf. Carston 1988: 156). This independence means that the truth conditions of the explicature and the implicatures are independent of each other. Sentence (51) is an answer to (50) and utilizes an implicature in (52). It is called an implicature because its truth conditions differ substantially from these of (51). (50)

Have you read Susan's book?

(51)

I don't read autobiographies.

(52)

The speaker [of 51] hasn't read Susan's book.

In general, implicatures are said to have distinct propositional forms and function independently of the explicature. The question remains, how big a contribution pragmatic processes make to the propositional form (explicature). As was demonstrated in Section 1.2.5, even the meaning of and can be regarded as part of an explicature. If the 27

Semantic underdetermination has to be distinguished from ellipsis. Semantically underdeterminate sentences do not have syntactic gaps. 28 Explicature is also known as the propositional form of the utterance, which is a narrower term of Sperber and Wilson's.

34

Discourse, beliefs, and intentions

Chapter 1

temporal meaning was an implicature, then the implicature would entail the more general (logical '&') explicature and there would be no need to store both in memory. Therefore, instead, it is better to assume the criterion of functional independence and say that 'and then' belongs to the explicature. Now, Carston says that causal and also exhibits sense-generality and allows for pragmatic contribution to the logical form, although it would be difficult to envisage any slots in the logical form which require filling in.29 In general, there need not be such slots for adding information to the logical form. The prepositional form may be syntactically complete, and fully truth-evaluable. But the proposition expressed is very likely to be more precise, so the form requires (in Bach's terms) further expansion as in (53) and (54): (53)

The park is some distance from where I live.

(54)

It will take us some time to get there.

(from Carston 1988:164, my emphasis). So, the explicature intuitively requires this specification of distance or time in these examples. Otherwise the speaker would not be co-operative, i.e. relevant or informative. Now, calling all this embellished content 'proposition expressed' or 'what is said' just because the implicature should not entail the explicature may seem somewhat unnatural. On the other hand, as Sperber and Wilson (1986a: 182) say, explicitly communicated assumptions are these which are developments of the utterance's logical form - its semantic representation. And this is intuitively plausible but not formally constrained: we can go on adding details. Hence, they employ the principle of relevance to constrain this elaboration of the logical form: the effort spent in processing the utterance and the gain in information must go hand in hand.30 This postulate of constraints imposed by relevance does not seem sufficiently informative in that it is difficult to see how the postulate of the upper bound which is tied to the principle of relevance can be used in practical calculations of pragmatic effects. And, neither Bach's middle level nor the above suggestion seem to suffice for postulating the level of underdetermined semantics. Whether we call the pragmatic intrusion a middle level or pragmatic aspects of the explicature, the semantics remains a separate, underspecified, level of analysis. If it is to be regarded as a mere theoretical construct, it is uninteresting and methodologically redundant. If it aspires to a psychologically plausible stage in utterance processing, then it is both methodologically and psychologically redundant. Psychologically, because there is no independent support for its existence. 29 30

For temporal and, such slots are easier to envisage: we can quantify over times. More discussion of relevance theory will follow in Sections 1.4.2 and 2.9.

Section 3

Semantic ambiguities and semantic underspecification

35

Carston sees linguistic semantics as a sub-part of truth-conditional semantics. The linguistic sense (also known as logical form, semantic representation) often does not give a full prepositional form which is truth-evaluable. Expanded or completed logical form does, but expansion and completion are pragmatic processes. They are pragmatic and at the same time subsumed under truth-conditional semantics. She evokes another standard test to support her point, namely falling within the scope of logical operators such as negation, disjunction and the conditional (see example (21) in 1.2.5 above). This does not prove the correctness of the explicature view either. This test was used by Cohen (1971) to propose rich semantic content of e.g. 'and' with cancellable features of 'then' or 'as a result'. So, the test is not conclusive. As was signalled in Section 1.2.5, the principle of functional independence is not a precise criterion; in fact, it does not always work. Recanati (1989b) approaches the issue of implicature in the following way. He agrees with Grice that when the sentence seems semantically ambiguous, then rather than assuming two literal meanings, we prefer a pragmatic account - this is dictated by MOR. Then, these pragmatic aspects may have to contribute to what is said:

"... it is no longer possible to contrast 'what is said' with those aspects of the interpretation of utterances that are pragmatically rather than semantically determined; for what is said turns out to be, in a large measure, pragmatically determined. Besides the conversational implicatures, which are external to (and combine with) what is said, there are other nonconventional, pragmatic aspects of utterance meaning, which are constitutive of what is said." Recanati (1989b: 98).

There are three levels of meaning: sentence meaning, what is said, and what is communicated. So far Recanati agrees with Carston. Now, as to the criteria for distinguishing conversational implicatures from pragmatic constituents of what is said, he opts for an intuitive basis of recognition of what is said (cf. Section 1.2.5 above and his Availability Principle31). All in all, Recanati disagrees with Bach's claim that we should take the minimal proposition as what is said and say that the speaker speaks non-literally. 31

The Availability Principle accounts for what was standardly called generalized conversational implicature, subsuming it under what is said as in (i) and (ii): (i) Tom has [exactly] three children, (ii) Tom studied at Oxford or Cambridge [but not both].

36

Discourse, beliefs, and intentions

Chapter 1

Levinson (1995: 90) also talks about three levels of meaning but the processes of utterance interpretation are classified in yet another way: (a) sentence meaning, (b) utterance-type meaning, and (c) utterance-token meaning. The middle level consists of, among others, generalized conversational implicatures: "This third layer is a level of systematic pragmatic inference based not on direct computations about speaker-intentions, but rather on general expectations about how language is normally used. These expectations give rise to presumptions, default inferences, about Levinson (1995: 93). both content and force..." This level also includes conventions of language use. When the default meaning is conveyed, the hearer stops at the level of utterance-type, i.e. the default, preferred interpretation, without going to the effort of considering the speaker's intentions. Levinson talks here about default interpretations of utterance-types, governed by his heuristics of the type: "Marked descriptions warn 'marked situation'" (ibid.: 97), as in sentence (55). In (55), the degree of reliability is lower than in (56): (55)

The mail is not unreliable.

(56)

The mail is reliable.

(from ibid.: 105). In view of this addition of the level of default meaning, the differences between the views of Carston, Recanati, and Bach are considerably reduced: conventions may have to be allocated a more significant place. To sum up, the controversies in the semantics/pragmatics boundary dispute concern solutions to the following problems: (i) how many levels of meaning there are; (ii) how they are to be defined; and (iii) what criteria to use to distinguish implicature from what is said. As yet, we have found no sufficient argument in support of the thesis of the lavish pragmatic input into what is said. A fortiori, the questions (i) to (iii) beg for a different answer to the ones discussed above. And, with a different kind of answer, we may try a different kind of theory which does have independent support, such as Default Semantics suggested in Chapter 2.

Section 4

Semantic ambiguities and semantic underspecification

37

1.4. Towards Delimiting Implicatures 1.4.1. Neo-Griceans and Ineradicable Conventions In the preceding sections, occasional vague reference has been made to Grice's principle of co-operation, Levinson's principle of Informativeness, and Horn's principles of Quantity and Relation. These and other post-Gricean approaches require a more detailed discussion in order to delimit implicatures with respect to what is said. Only then can we attempt the semanticization of the contentious layer that falls between bona fide implicatures and bona fide semantic content. Discussing the role of context in the study of meaning, Stalnaker (1998:18) observes that the traditional Gricean view is still very plausible, namely that one should invest in the study of conversation as a rational activity instead of overburdening semantic theory. But overburdening semantics does not necessarily mean positing semantic ambiguities. Allowing context to contribute to the semantic representation in an unguarded manner is equally overburdening as unjustified ambiguities. For Grice, the boundary between what is said and what is implicated was quite clear. What is said contains truth-conditional aspects of meaning, what is implicated is non-truth-conditional. Implicatures can be conventional, attached to particular words such as 'but' and 'manage', or, more importantly, non-conventional with a sub-category of conversational implicature. These can arise either independently of the context or be engendered by the contextual information.32 Conversational implicata are calculable through the Co-operative Principle and the maxims of conversation.33 Now, even if the general idea retains its appeal, the strict boundary between the prepositional content and implicatures proved to be untenable. The only pragmatic processes allowed by Grice to contribute to the semantic representation were reference assignment and grammatical disambiguation (see Section 1.2.1 above). As we know now from the attempt to delimit what is said, the pragmatic intrusionism is much more omnipresent if we want to retain the idea of semantic representation as a truth-conditional representation. The subsequent attempts to improve on the set of Grice's maxims also had to account for this difficulty. Horn (1984) revives Zipf's Principle of Least Effort from the 1940s to generalize over the tendencies in human communication and stresses the dialectic opposition between 'speaker's economy', using as few words for

32 33

See the diagram in Horn (1988: 121). See fn 7 in Section 1.2.3 above.

38

Discourse, beliefs, and intentions

Chapter 1

concepts as necessary, and 'auditor's economy', using as many expressions as messages to make the communication clear. This revival results in the Q Principle and the R Principle, briefly referred to in Section 1.2.3: The Q Principle: Make your contribution sufficient. Say as much as you can (given R) The R Principle: Make your contribution necessary. Say no more than you must (given Q). (adapted from Horn 1984:13). The Q Principle is a lower-bounding principle: if the speaker says that some people came to the party, an upper-bounding implicature to the effect that not all (invited) people came is induced. It is a principle of maximization of the informational content and it gives rise to scalar implicatures. The R Principle is upper-bounding: saying that I broke a finger implicates that the finger was mine. It is a principle of minimization of form. According to this principle, the use of a marked expression such as a complex, more effortful one when a simple one is available, renders a marked interpretation, i.e. gives rise to implicatures. Now, whether the scales that arise out of Horn's Q principle such as < . . . two, three, four,.. .>, < and, or> are truly context-independent is a moot point. Matsumoto (1995) stresses that the choice of a weaker expression over a stronger one can sometimes be attributed to the observance of maxims other than Grice's Quality and Quantity-1, and then the implicature does not arise. The critique is perhaps not a very powerful one since Grice's sub-maxims seem to be well distributed between Horn's two principles34, but the allocation of an arising implicature to Grice's sub-maxims does not seem to be testable anyway once we assume Horn's over-arching principles. The pragmatic theory would have to be overly fine-grained and overly transparent from the methodological point of view to keep delivering explanations below the level of Q and R. Levinson (1987) follows the same pattern of amendments to Grice, adding a distinction of the minimization of form and minimization of content. Semantically general expressions are said to be preferred to specific expressions, and shorter expressions to longer ones. The Q-implicature is taken from Horn (1972) and Gazdar (1979), and I-implicatures are the result of the Principle of Informativeness of Atlas and Levinson (1981): "The 'best' interpretation of an utterance is the most informative one consistent with what is non-controversial" (Levinson 1987: 66). His Q-Principle says the following:

34

The Q Principle replaces Grice's Quantity 1 and Manner - 'Avoid ambiguity, avoid obscurity', and the R principle replaces Grice's Relation, Quantity 2 and Manner - 'Be brief.

Section 4

Semantic ambiguities and semantic underspecification

39

"Speaker's Maxim: 'Make your contribution as informative as is required for the current purposes of the exchange.' Specifically: don't provide a statement that is informationally weaker than your knowledge of the world allows, unless providing a stronger statement would contravene the I-principle." Levinson (1987: 67). For example, sentences (57) and (58) have conversational implicatures marked by '+ >': (57)

I often take sugar in my coffee.

+ > not always

(58)

I believe that John is away.

+ > not know

(from ibid.: 64-65). The hearer infers on the basis of this principle that the speaker made the strongest statement consistent with his/her knowledge. The I-Principle works as below: "Speaker's Maxim: The maxim of Minimization. 'Say as little as necessary' i.e. produce the minimal linguistic clues sufficient to achieve your communicational ends, bearing Q in mind." Levinson (1987: 68). The hearer will normally enrich the content of the utterance up to the most specific interpretation that is judged to be intended. For example, (59) and (60) become enriched as signalled by '+>'. (59)

John turned the key and the engine started.

+ > and then

(60)

Harry and Sue bought a piano.

+ > together

(from ibid.: 65). Following his distinction into minimization of content and minimization of form, Levinson adds the M-Principle which says that the speaker does not use a prolix, obscure or marked expression without reason: if he/she does so, then implicatures are inferred: "Try letting the I-principle win in the first instance, i.e. go for minimal forms; if that doesn't work escalate step by step towards a Q-principle solution." Levinson (1987: 119).

40

Discourse, beliefs, and intentions

Chapter 1

Implicatures are ordered on a scale of adding them to the prepositional content: Q precedes M which precedes I. The order of adding information to the prepositional content extends beyond implicatures. First, background assumptions are added, then semantic entailments, followed by conversational implicatures: Q-clausal (cf. example (58)), Q-scalar (cf. example (57)), M-implicatures, I-implicatures, and finally presuppositions, if we distinguish them as a separate category (cf. Gazdar 1979: 135. Adapted to Levinson's taxonomy of principles by Huang 1994: 15). This is also an ordering of cancellation procedure: if added material is inconsistent with what had already been added, it becomes cancelled. Generalized conversational implicatures obtain strong support in Levinson's middle level of meaning, the level of conventions of language use (1995: 93). This middle level seems to defend itself from objections of other orientations: "... that intermediate level is constantly under attack by reductionists seeking to assimilate it either to the level of sentence-meaning or to the level of speaker-meaning; thus, for example, in the case of the inferences we are calling GCIs, many theorists (Kamp, Peters, Barwise and others) have suggested that they should be in effect semanticised, while Sperber and Wilson and some so-called local-pragmatics theorists have presumed that on the contrary they should be assimilated to matters of nonce-inference at the level of speaker-intention. But generalised implicatures are not going to reduce so easily in either direction, for they sit midway, systematically influencing grammar and semantics on the one hand and speaker-meaning on the other. I shall therefore presume that we do indeed need such a three-tiered theory of communication." Levinson (1995: 95). The semanticization of the middle level of meaning is best represented in DRT, although at present DRT lacks a satisfactory account of the mechanism that triggers this semanticization. And it is this semanticization that seems to be the most adequate solution. It avoids the unnecessary proliferation of levels of analysis and hence adheres to MOR and POL. It is economical and simple and seems to work well for prepositional attitude sentences once it is merged with an intention-based account, as the analysis in Chapter 7 demonstrates. Now, how does the delimitation of what is said fare with the neo-Gricean amendments to the maxims? Horn opts for generalized conversational implicatures, some of which become 'short-circuited', then further conventionalized, as from use convention to meaning convention, metaphor to idiom, or rhetoric to grammar (cf. Horn 1988: 139). Levinson stresses

Section 4

Semantic ambiguities and semantic underspecification

41

the level of conventions in a similar way, taking it out of the semantics-pragmatics dispute. However, the internal coherence of this middle level cannot be taken for granted. There is evidence to the effect that scalar implicature is more context-dependent, or at least structure-dependent than Horn admits (Matsumoto 1995, and in particular van Kuppevelt 1996). Van Kuppevelt (p. 393) points out some serious problems with scales, the need to reduce, reverse or suspend (not activate) a scale, as well as with scale coherence. He suggests that the generation of scalar inference may depend on the construal of the topic as indicated by topic-forming questions. For example, in (62), the scalar implicature of 'no more than four' is suspended and this fact is indicated by the topic-forming question in (61). (61)

Who bought four books?

(62)

Harry[comment] bought four books. In fact he bought seven.

The implicature goes through in (64), though. (63)

How many books did he buy?

(64)

*He bought four[comment] books. In fact he bought seven.

(from van Kuppevelt 1996: 411). In (64), the quantifying term 'four' has comment status and gives rise to an upper-bound scalar inference. This test does not work for all constructions but it certainly indicates a form of context-dependence of scalar inference. Hirschberg (1985) is much more generous towards context-dependence and simply states that there are no defaults in generalized implicatures. Instead, they are all context-dependent. She suggests a list of orderings that generalize over instances, such as set/proper subset; whole/part; type/subtype; entity/attribute but admits that they do not constitute a theory and can perhaps be created ad hoc and ad infinitum (1985: 114). Also, listing of scales is impossible because speakers can form unpredictable scales in which items are hardly related. Instead, Hirschberg suggests 'partial ordering' that is salient in a particular context (see ibid.\ 65). An instance of an implicature arising out of the entity/attribute ordering is (65). An instance of an ad hoc, non-linear scale is presented in (66). (65)

A: Are you a doctor? B: I have a PhD.

(from Hirschberg 1985: 112). (66)

A: Did you get Paul Newman's autograph? B: I got Joanne Woodward's.

42

Discourse, beliefs, and intentions

Chapter 1

(from ibid.: 50). There are a number of other more or less destructive counter-arguments to context-free implicatures (Green 1995; Carston 1988, 1998b). They all point to the weakness of the current analyses of the 'middle level' of meaning. They do not seem to endanger the existence of the middle level as such but, if its assimilation to semantics proves satisfactory (as it is beginning to be for the fragments of natural languages analysed in DRT and File Change Semantics (Heim 1988)), the defence of the middle level will become redundant.

1.4.2. Relevance Theory and Frames Sperber and Wilson (1986a) analyse human communication as a process of construction and recovery of assumptions. An assumption is said to be relevant in a context just in case it has cognitive (contextual) effects in this context (cf. ibid.: 122). The more cognitive effects an assumption has, and the less processing effort it requires, the more it is relevant. In this way Gricean maxims are replaced by one principle, the principle of relevance. Every utterance which is an act of ostensive communication carries with itself a presumption of its optimal relevance (cf. ibid.: 158).35 This radical reductionism is certainly methodologically appealing. However, in its present state, it subsumes some claims which still require further scrutiny. For example, it subscribes to sense-generality which possibly leaves too great a role to contextual diversity, as the attempt to constrain it to defaults presented in the following chapters will demonstrate. Next, it provides no precise mechanism for measuring efforts and effects which would yield successfully to computer modelling (cf. Levinson 1989). Finally, depicting utterance interpretation as a process of the recovery of speaker's assumptions may be overly constraining, as demonstrated in my reanalysis of this process in terms of meaning-construction which happens in-between the interlocutors in Jaszczolt 1996b and Section 2.9 below. Now, the inference from 'and' to 'and then' and 'and therefore' can be accounted for in two ways. First, we can postulate stereotypical situations in a neo-Gricean manner. The situations can be either listed and the speaker's knowledge of them is assigned to acquired conventions which have very little in common with semantics or pragmatics (cf. Levinson 1988, 1995), or, they can be generalized over by over-arching scales. The first, if present, pertains at least to informativeness, if not all generalized implicatures. The other pertains, if at all, to scalar Q-implicata. 'If present', 'at least' and 'if at all' 35 Ostensive-inferential communication means that the speaker points out to the addressee the intention to convey information, from which further implicatures can be drawn (cf. Sperber and Wilson 1986a: 54).

Section 4

Semantic ambiguities and semantic underspecification

43

signal here the plethora of counterexamples and contrary intuitions. Alternatively, we can analyse the frames or scripts to which the hearer refers in terms of processing effort and cognitive effect: 'and' is used instead of 'and then' because it takes less effort to process and produces the same effect (cf. Carston 1994). But this is so because the hearer refers to such scripts. The question arises whether scripts alone will not do as a successful explanation of this parsimony. Wilson and Sperber (1998) claim that (67) normally has to be interpreted as a statement that the speaker has had breakfast within the past few hours as otherwise it would not have any cognitive effects. On the other hand, (68) imposes no such interval constraint and is happily interpreted as a statement that the speaker has visited Tibet some time in the past: (67)

I have had breakfast.

(68)

I have been to Tibet.

(from ibid.: 11). In order to find the suitable interval for the interpretation of such sentences, the hearer has to have some general, encyclopaedic knowledge of what people normally do and say. In other words, he/she has to have access to schemas of previously encountered events, states, and sequences of these. Encyclopaedic knowledge about such frequently encountered situations triggers some interpretations, or, in Wilson and Sperber's terminology, makes some assumptions more accessible (cf. ibid.: 15). Occasionally, scripts pertaining to world knowledge seem to be supplemented by linguistic devices that do not contribute to the propositional content of an expression but rather provide instructions as to the way in which the utterance should be interpreted. Blakemore (1987, 1989) lists here 'so', 'moreover', 'you see', 'after all', 'but', and many other discourse connectives. Hence, it is not only the case that the sentence requires further contextual embellishment before the proposition is recovered, but some parts of this sentence may have to be classified as part of what we would call here contextual information: they direct the process of interpretation, constrain the derivation of implicatures. This division into truth-conditionally relevant and irrelevant items is not as straightforward as it would seem from the analysis of connectives: there can be truth-conditionally relevant elements that correspond to processes rather than concepts, such as personal pronouns, and non-truth-conditional elements that are conceptual such as illocutionary adverbials, e.g. 'frankly', 'confidentially', 'unfortunately' (see Wilson and Sperber 1993). Hence, scripts appear to operate on a rather diversified material. Semantic representation can be enriched by truth-conditionally relevant elements that are not conceptual but signal the need for pragmatic processes to operate (as in the case of reference resolution for pronouns),

44

Discourse, beliefs, and intentions

Chapter 1

and the procedural non-truth-conditional elements further direct the recovery (or production) of implicatures. As relevance theorists say, they clarify the relevance of the utterance. Scripts seem to fit in rather freely at more than one place of this many-stage process.36 They can be activated as soon as relevance to the particular script becomes salient to the hearer, which, naturally, defies any more radical constraints and ordering. It appears from these two analyses that schemas or frames serve a very important role in utterance interpretation. They are relied on both by neo-Griceans and relevance theorists and it seems that for both they are not reducible to the pragmatic theory. Wilson and Sperber (1998: 17) claim that the temporal and causal interpretations of sentences conjoined by 'and' are engendered by the interaction between sentence meaning, general cognitive factors, and the criterion of consistency with the principle of relevance (see also Carston 1998b: 197). It seems that the cognitive factors have to operate prior to the assessment of the consistency with the principle of relevance and the output of their help is assessed by subjecting it to the principle itself. This temporal precedence seems to make them an independent stage in utterance processing: a stage which is perhaps hastily defined by Levinson as pragmaticized by relevance theorists.37 All in all, whether we accept the neo-Gricean informativeness principles and special status of generalized implicatures or follow optimal relevance as a route to pragmatic inference in general, there are some presumptions about what the world is like which are a part sine qua non of utterance interpretation. And, what the world is like, includes what speakers are predisposed to do, perhaps by the way intentions normally work. In what follows, I shall try to explore the middle level of meaning without making an assumption that it is an irreducible level, neither that it belongs to semantics or pragmatics. The irreducibility,'semanticity'or 'pragmaticity' will be the result of the investigation into how intentions work in communication.

1.5. Conclusions Semantic ambiguity seems to have been successfully exorcized by means of MOR. The unitary semantics complemented with conversational implicatures was not a powerful enough replacement because it did not account for the diversity of pragmatic processes that help semantic representation become 36 See also Richer 1997 who suggests that discourse connectives work as so-called parenthetical performatives: 'but' is interpreted as the truth-functional connective '&' plus a performative utterance 'I suggest that these two propositions contrast', or 'I ask you to notice that this contrasts'. He objects to the classification of discourse connectives as uniformly procedural (cf. ibid.: 68). 37 See quotation from Levinson (1995: 95) in Section 1.4.1.

Section 5

Semantic ambiguities and semantic underspecification

45

truth-evaluable. In order to accommodate this intrusion, the clear-cut boundary between semantics and pragmatics had to be abandoned. It was replaced by underdetermined semantics according to which sentences can be sense-general and pragmatics contributes widely to what is said. Other variants include the middle level of pragmatic aspects of what is said or the middle level of conventions of language use which is neither semantic nor pragmatic. I have signalled in this chapter that there is a possibility of another approach which is worth scrutinizing. This approach does not postulate a level of sense-generality but rather semanticizes the pragmatic contributions to the semantic representation in the form of default interpretations or departures from the default. I have not said much about this fourth view as yet. The purpose of the first chapter was to prepare the ground by pointing out some weaknesses of the current dogmas such as the methodological weakness inherent in postulating an unnecessary level of representation. The difficulties of accounting for semantic underspecification, as well as the expansion of the minimal proposition and completion of the propositional radical without giving way to a flood of pragmatic processes on the one hand, and/or postulating a non-linguistic level of conventions on the other, raise doubts as to the adequacy of underspecification for the analysis of utterance interpretation. The more so that conventions or schemas seem to give rise to default interpretations and there is only one step from there to saying that these conventions and schemas can sometimes operate on the level of semantics. Underspecification can thus be suspected of being parasitic on tidy and predictable world knowledge. If this view is sufficiently supported, underspecification should be exorcized.

This page intentionally left blank

CHAPTER 2

Semantic Defaults "The theory of utterance-type meaning should be a theory of default interpretation". Levinson (1995: 109-110).

Contents 2.1. 2.2. 2.3. 2.4. 2.5.

2.6.

2.7. 2.8. 2.9. 2.10.

Setting the Scene Intentions in Communication The Primary Intention Principle and Definite Descriptions The Principle of the Parsimony of Levels Default Semantics? 2.5.1. Default Reasoning 2.5.2. Defeasible Knowledge 2.5.3. Defaults for Definite Descriptions: Interaction of Intentions 2.5.4. Defaults for Indefinite Descriptions Other Applications 2.6.1. On Semantic Representation and Negation Again 2.6.2. And, or, and if 2.6.3. Number Terms Defaults in Dynamic Semantics Interim Conclusions The Doubly-Dynamic Approach: Assumption Recovery or Creation? Conclusions

DISCOURSE, BELIEFS, AND INTENTIONS Current Research in the Semantics/Pragmatics Interface, Vol. 2 K.M. Jaszczolt © 1999 Elsevier Science Ltd. All rights reserved 47

48 49 51 56 58 58 60 62 65 66 66 67 68 70 75 76 85

48

Discourse, beliefs, and intentions

Chapter 2

2.1. Setting the Scene Semanticization of the middle level of meaning can easily result in concealing ambiguities under a new label. DRT, by providing analyses of problematic expressions such as number terms, assigns them various discourse representation structures (DRSs) and makes the choice context-dependent (see Kamp and Reyle 1993). This need not be so. In stressing the need for a better understanding of utterance-type meaning, Levinson (1995: 93) appeals to people's general expectations about how language is commonly used in given situations. Pragmatic inference performed by the hearer is based on these expectations rather than on the calculation of speaker's intentions. That is a reasonable starting point for an analysis of this underdeveloped level of meaning and one can anticipate that, once it is better known, it will cease to be seen as an isolated curiosity, i.e. as a mysterious level between sentence-meaning and speaker-meaning. As was noted earlier, both pragmatization and semanticization have been tried. The first one led to letting pragmatics loose on a weak semantic form and the other led to hidden ambiguities, albeit attended to under context-shifts of dynamic semantics. In this chapter, I try another approach. It consists in a non-ambiguous semantics achieved through the interaction of the speaker's intentions with the logical form of the expression. The problem of ambiguity will not arise because intentions 'intrude' into the semantic representation. The view is thus compatible with the pragmatic intrusionism of dynamic-semantic approaches such as DRT (Kamp and Reyle 1993), where pragmatic factors 'intrude' into a unitary semantic domain. There is one, discourse-level representation, to which syntax, semantics and pragmatics contribute. However, whereas DRT treats all possible resulting representations equally, the account proposed here orders them on the scale of salience and predictability from the default one to the furthest departure from the default. I suggest that the idea of default semantics, in addition to faring well with MOR, enjoys an indelible psychological plausibility. Intentions in communication direct utterance interpretation to the correct semantic representation. Hence, I suggest an unambiguous and unitary semantics, obtained essentially along the lines of Kamp's (1984) dynamic semantics. However, I also attempt to resolve the problem of the various possible interpretations of such 'interpretatively ambiguous' sentences which are assigned to them by Kamp's DRT. A proposed addition to Occam's Razor will be employed to perform this task. This 'addition' consists of the principle of economy with respect to the postulated levels of interpretation, so far mentioned very briefly as POL and to be elaborated on in Section 2.4 below. First, it is necessary to decide what types of intentions are involved in the communication process and what exactly their role is.

Section 2

Semantic defaults

49

2.2. Intentions in Communication As was discussed at the beginning of Section 2.1, there are cases of utterance interpretation which do not rely on a computation of speaker's intentions. What this means is that this computation can be 'short-circuited', preempted by conventions. However, the difference between the two ways of arriving at speaker meaning need not be theoretically significant. The calculation of intentions retains its explanatory value even when it ceases to happen in practice. The more so that this reliance on intentions often decreases gradually, following the increase of our world knowledge and world experience. Hence, as a theoretical tool, intentions remain indicative of utterance interpretation whether they are relevant for a particular situation or not. Conventions enter the picture only at the stage of explaining how information was communicated rather than what was communicated. And it is the first that we are interested in while weighing sense-generality and semantic defaults against each other. After all, it is widely acknowledged that sentence meaning ultimately has to be analysed in terms of intentions and other psychological states (Grice 1989; Sperber and Wilson 1986a; Recanati 1993). A brief recent history of intentions looks as follows. Grice (1957,1969) says that the hearer understands the speaker's utterance by recognizing the speaker's intention. As Bach and Harnish (1979: xi) put it, "... a communicative intention has the peculiar feature that its fulfillment consists in its recognition".1 Sperber and Wilson (1986a) add here the distinction between the communicative and the informative intention, the second embedded in the first. The communicative intention is defined as making it "mutually manifest to audience and communicator that the communicator has this informative intention." (Sperber and Wilson 1986a: 61). Bach and Harnish (1979: 7) do not distinguish a separate informative intention. They talk about the illocutionary-communicative intention guaranteed by the 'communicative presumption': that whenever the speaker says something to the hearer, he/she is doing so with some (illocutionary) intention. Bach (1987a) adds here the referential intention. He develops Grice's (1969) original account of intentions in communication and calls the basic intention speaker (communicative) intention (Bach 1984, 1987a): "Which thought an utterance of a sentence expresses depends not only on the meaning(s) of the sentence but also on the speaker's communicative intention. If the sentence is ambiguous, the speaker's intention determines which of its meanings is operative 1

See also discussion in Bach 1987b.

50

Discourse, beliefs, and intentions

Chapter 2

(...). If the sentence is not being used literally, that is because of the speaker's intention. In this case he must intend the hearer to recognize a certain connection between what the sentence means and what he means in order for the hearer to identify the latter (...). And if the sentence contains an indexical, its meaning underdetermines what the speaker means (even if the sentence is unambiguous and is being used literally), and how the gap is to be filled depends on the speaker's referential intention!" Bach (1987a: 65-66; my emphasis). The referential intention constitutes a part of the communicative intention. The fulfilment of the communicative intention consists in its recognition by the hearer (cf. Bach 1987b). Also, ".. .what is said, to the extent that it is not fixed by linguistic meaning, is determined by speaker intention, which itself can include the intention to refer to what one is demonstrating." Bach (1992: 140). Intending means essentially acting out of one's beliefs and desires. Grice observes that for a speaker to mean something by uttering a sentence he/she has to intend the hearer to [1] produce a particular response (cognitive or physical, as Strawson 1964a emphasizes), [2] think that the speaker intends that the hearer produces a response and [3] produce the response on the basis of thinking that this is what the speaker intends (see Grice 1969: 92). Except for some sophisticated situations where there is a need to step down to the wth level sub-intentions, the schema seems to work, although in the case of conventionalized procedures, such as bidding in a game of bridge, conventions are said to do the job and secure the uptake of the intention automatically (see Strawson 1964a). Whether one agrees with the role of conventions so assigned or not, it is conspicuous that one essential element is missing in these accounts and that is the explanation of what exactly it means to secure a response, according to what procedure and by means of what information. An account of speaker meaning that espouses intentions is wasted if stopped before letting the idea of intentions perform its task. And the task, as I see it, can extend to resolving the problem of interpretative ambiguity. In order to do so, we have to say more about intention as securing the referent in conversation.

Section 3

Semantic defaults

51

In other words, the fact that a speaker means something by uttering a sentence is equivalent to saying that he/she utters it with "... the intention of inducing a belief by means of the recognition of this intention" (Grice 1957: 219).2 First of all, the speaker intends the object or person spoken about to be easily individuated by the hearer. This is the main preference in communication and the main point of reference for evaluating the role of referring and quantified expressions in utterance interpretation. In what follows, the referential intention will be tried out as a criterion for deciding between various interpretations of an utterance involving an expression that is used to refer.3 It will be demonstrated that the referential intention helps to order these interpretations according to their salience. To sum up: across the board, there have been three major types of intentions distinguished in the literature: communicative, informative, and referential. The simplest case of referring is when the syntactically and semantically simplest expression (i.e. simplest among its functional equivalents) is used with the intention to communicate that the referent is to be recovered and to inform about this very referent. The present account makes use of such a combination of three types of intentions: reference-securing, communicative, and informative.

2.3. The Primary Intention Principle and Definite Descriptions The role of the referential intention in communication is best observed in the interpretation of definite descriptions. In the referential use of definite descriptions, the intention to refer is present: the speaker intends to make a certain individual salient rather than merely talk about a state of affairs. As can be observed in conversational practice, the hearer often, if not always, presumes that this intention is present when a definite noun phrase was used by the speaker, unless the context makes it clear that this was not the intended meaning. To give this intuition some theoretical support, we can posit with some initial plausibility that perhaps this presumption of the referential intention is derived from the intentionality of the underlying mental act. Intentionality is a property of mental states that makes them about, or of objects and states of affairs (cf. also Searle 1983, 1990a). Only some mental states are intentional: beliefs, hopes and fears are, as opposed to nervousness or elation that are not. The claim goes back at least to the nineteenth-century idea of intentionality, put forward by Brentano (1874) and developed by Husserl (1900-1901, 1913), although the connection between contemporary 2

For examples of meaning that cannot be straightforwardly explained by an intention to produce a (cognitive or linguistic) response, see Avramides (1997: 76-78). 3 The term 'referring expression' is deliberately avoided at this stage.

52

Discourse, beliefs, and intentions

Chapter 2

studies of intentions in communication and the phenomenological tradition is not always acknowledged and not always appreciated, perhaps because HusserPs philosophy had undergone many turns. Intentionality says that acts of consciousness are directed towards an object. These objects trigger a semantic interpretation.4 Meaning is seen here as social, established in the process of social, meaning-giving acts. Knowing the object depends on fulfilling intentions either by thought, by perception, or by imagination. As language is one of the possible vehicles of thought (Jaszczolt 1992, 1996a), intentionality also governs its use. Some beliefs (thoughts) are externalized by means of language. This is how intentionality becomes a property of linguistic acts (by being, so to speak,'inherited' from mental acts). Hence we can say that some intentional states are linguistic. And just as intentionality means aboutness, so there is an intention to speak about an object. On uttering a sentence, the speaker intends a particular object or person to be salient to the hearer and this intention constitutes the internal, undetachable property of the constructed representation. Intentionality of securing a referent ensures that semantics is atomistic, about the real world, and that acts of meaning are acts of experience of the real world, about the objects of the world, and primarily aimed at the objects of the real world (in agreement with Husserl 1913). Intentionality so perceived incorporates the workings of intentions in communication, since language is one of the vehicles of thought (paceW. Lyons 1995). The speaker's intentions to communicate something and to inform the hearer about something also rely on this primordial intention to secure the referent. So, the referential intention can be tested for its default properties in utterance interpretation. This thesis was put forward in the Introduction as the Primary Intention principle (PI): The primary role of intention in communication is to secure the referent of the speaker's utterance. The 'referent' is understood as a result of intentionality and hence can be either an entity or an eventuality (state, event, or process, after Kamp and Reyle 1993: 509).5 Following Kent Bach, I distinguish a special kind of intention that performs the reference-securing task. Naturally, one has to allow for the instances where this intention is absent either by force of the type of the expression used or by force of particular circumstances. Not every sentence concerns an individual. But we shall entertain the thesis, to be tested in various examples below, that whenever a possibility of a referential interpretation is viable, it is secured by the referential intention which is assumed by the hearer to be present. This is the general idea behind PI. It has to be pointed out that the claim made in PI is considerably stronger than the current speech 4

What Husserl earlier called objectifying, meaning-giving acts. See Jaszczolt 1996a and Chapter 3 below. 5 The term 'eventuality' is due to Emmon Bach (see Bach 1981: 69).

Section 3

Semantic defaults

53

act/intentionality tradition allows for. Bach (1987a: 52), when discussing prepositional attitude contexts, admits that when the speaker expresses an attitude towards an individual, the referential intention must be present. However, this claim does not yet aid the indiscernibility problem of referential and attributive readings of definite descriptions. Other works in the intentionality tradition follow suit. The general idea is this: as long as communication is said to incorporate only the need to inform or perform any other illocution (using illocutionary intention) on the part of the speaker, assumed by the hearer to be present, then the fact that the hearer normally comes up with only one interpretation remains unexplained. After all, whether one subscribes to the underspecified semantics view or not, one has only very vague explanations available of the fact that although there is more than one interpretation of an utterance involving the interpretative ambiguity of negation, the hearer does not hold all of them in the mind, pondering which one to choose.6 The referential intention, when present, seems to explain this fact of instantaneous interpretation. The device of referential intentions, suggested for the cognitive level of interpretation, opens the possibility of maintaining it for the semantic level as well. Some theorists would not see a problem with having a technical device in their semantics that does not serve a cognitive explanation, i.e. does not have a psychologically plausible equivalent.7 Hence, they see no danger in multiplying levels of explanation and adding underdetermined representations. But this is methodologically ill-founded. On the other hand, the price to pay for our rigid semantic parsimony may be high as we may have to revindicate some long abandoned ideas of speech act theory such as the default correlation between an utterance and a speech act. However, the referential intention makes only a small demand in this contentious revival: there is no direct correlation between sentence types and speech acts. Instead, there is a correlation between utterances and default situations they are used to describe. Moreover, defaults are rather weak: they are easily overridden when contextual information makes them so. Or, rather, from the processing point of view, they do not arise then. After all, PI allows for nonreferential interpretations which perform the same speech act as the referential reading of the same utterance, as in example (38) of Chapter 1, repeated here as (1):

I mean here the incremental nature of utterance interpetation and post-Gricean accounts of conversational inference. 7 Victoria Escandell-Vidal (personal communication) says: "... I would probably still prefer a two-layered model without claiming any psychological reality (...), at least until a way out of the problems of the notion of speech act is found." and: "... what you gain in psychological verisimilitude gets somewhat lost in theoretical elegance; and vice versa, what you gain in theoretical elegance you lose in psychological reality."

54

(1)

Discourse, beliefs, and intentions

Chapter 2

The best architect designed this church.

The problem under discussion concerns only the class of representatives. In addition, the issue of indirect speech acts remains open to interpretations; it can also be left untouched. All in all, if intentions contribute to the semantic interpretation, they render the reading that depends on the presence or absence of the referential intention and this fluctuation has an indirectness flavour itself.8 In sentence (1), when the speaker is taken to be talking about a particular, known individual, e.g. Christopher Wren, the referent is easily secured. But when the speaker talks about whoever happened to have designed the church and who is otherwise unknown to him/her, then the intentions involved seem to be aimed at a person but fail to provide an identifiable referent. And, finally, one has to establish the object of the utterance when the speaker is obviously referentially mistaken. This would be the case if in (1) the speaker was taken by the hearer to be talking about John Smith thinking mistakenly that Smith designed the church. Since the social, recognizable act (speech act) secures reference, Smith should be the referent. However, one may also say that since intentionality is social, it reflects the use of language in a linguistic community and thus that intentions proceed towards the bearer of the name or description because this is what they normally do for a hearer who is not referentially mistaken. But let us suppose that the hearer is not sure whether the speaker meant Smith or Wren, i.e. the true architect of the church. This confusion may be engendered by the prior assertion of the speaker to the effect that Wren is known to the speaker as an eminent architect. To systematize these differences in interpretation one can again resort to the PI and see it as a matter of degree, as in the principle of the Degrees of Intentions (DI) introduced earlier. The principle says that intentions allow for degrees. It accounts for the ordering of the three readings of definite descriptions in the following way. In the case of (i) a referential use of the description in (1), intentions are the strongest and the referent is secured outright, as if by default. In the case of (ii) a referential mistake, intentions proceed towards two different objects, one along the path of 'social intentionality' and one along the 'individual path'. By 'social intentionality' Bach and Harnish (1979: 93) say on this point that there is no standard strategy for recovering the illocutionary intention. My addition to this minimalism is to claim that although there is no general recipe for interpretation, there is one for the standard interpretation in sentences that involve the multiplicity of senses, secured by the interaction of intentions, and, where applicable, by the referential intention. Bach (1987a: 66) recognizes the referential intention in the case of indexicals: their meaning depends on the referential intention of the speaker. From here there is only one step to saying that if things get complicated in interpreting informationally fuller referring terms (descriptions, proper names), the hearer also resorts to the speaker's intentions.

Section 3

Semantic defaults

55

I mean here the route which would be representative of an average native speaker who refers correctly, in agreement with the common wisdom. An 'individual path' represents the intentionality as it really occurs in the particular case of a particular speaker who can be referentially mistaken. The gap between them is reduced by utilizing contextual clues and one intended referent is established. In the case of (iii) an attributive use of (1), the referential intention sets off as normal, but does not reach any particular referent. We can say that here we deal with the weakest type of intention. The intention is still there: in the attributive reading, the speaker still intends a referent to exist, whoever it might be. The particulars are irrelevant until Chapter 3.9 All in all, the basic idea is this. The semantics allows for the intervention of pragmatics to various degrees. It is essentially a dynamic semantics, exemplified in the 'intrusionistic' models of Kamp (1984); Kamp & Reyle (1993); and Heim (1988) but differs from them in supplementing POL with an ordering of resulting interpretations. In other words, the degree to which the referential intention is present has its direct bearing on the semantic representation of the utterance of the sentence, which can be summed up as follows. The referential intention in its strong form secures the referent by its intrusion into semantic representation. The weaker the intention, the more information has to be read off the rest of the semantic representation. The ultimate case is the attributive use of (1) where the semantic representation alone secures the referent, whoever he or she might be. Now, saying 'the utterance of the sentence' may seem like another guise of an ambiguity: after all, sentences, not utterances, are said to have logical forms and semantic representations. But utterances have them by simple transitivity: they are utterances of sentences that have semantic representations. And, as has been known since Grice and certainly since the sense-generality debates, the utterance is equally important for the semantics as the sentence because primary pragmatic processes in utterance interpretation interact with the logical form.10 All that I have done here is conflate the levels of underspecified 9

I developed this proposal in detail in Jaszczolt 1997a. Turner (1997:167) summarizes Carston's (1988) account of utterance interpretation as follows: Linguistic Meaning —> Primary Pragmatic Processes —> Truth-conditional Semantics —> Secondary Pragmatic Processes The diagram depicts clearly the process of utterance interpretation, albeit without giving due attention to the differences in status of these components. 10

56

Discourse, beliefs, and intentions

Chapter 2

semantics and primary pragmatic processes. Let us say, at least some primary pragmatic processes for the time being, that is the ones that rely on the workings of the referential intention.

2.4. The Principle of the Parsimony of Levels This section further develops the idea of pragmatic intrusionism. As was discussed in Chapter 1, context aids in establishing the propositional form of an expression to a much larger extent than Grice originally suggested.11 In addition to lexical and syntactic disambiguation, there is a plethora of pragmatic processes that are activated before the propositional representation is complete. Here we also have to include expansion of minimal propositions and completion of propositional radicals. They ensure that MOR is satisfied as semantic ambiguities are exorcized. My problem with this picture concerns the justification for semantic representation as a different level from propositional representation. In other words, must semantic representation be seen as underdetermined and separate from the pragmatic processes that enrich it? This generosity can be dispelled with the help of intentions. Some sentences require further embellishment on the level of semantics but this does not yet guarantee the leap from this incompleteness to the epistemological commitment that semantics is underdetermined. From the methodological point of view, the question arises as to what advantage is gained from maintaining an underdetermined semantic representation which is prior in the order of cognitive processing and the order of explanation to a representation with a truth value. If intentions are let in, there is a reading of interpretatively ambiguous sentences in which the distinction between the semantic form and the truth-evaluable propositional form is redundant. If there are such defaults, the reasons for maintaining the above distinction will have to be reassessed. It is difficult to envisage what these reasons would be. It is possible to give an equally adequate account of sentence meaning in terms of a sole representation. This parsimony of levels of analysis can be seen as an extension to, or application of, MOR and has been briefly referred to in the Introduction as the Parsimony of Levels principle (POL). To repeat, the principle says that levels of sense are not to be multiplied beyond necessity.12 POL boils down to a restriction on postulating the two levels of sentence interpretation: semantic representation (here equated with logical form), and propositional representation as separate contributions to meaning (sense) when they are not discernible with cognitive plausibility. If what is said was interpreted as a form that is 11 12

See Grice (1975: 24; 1978: 46); Recanati 1989a, 1994. The principle, together with the DI and PI principles, originated in Jaszczolt 1999.

Section 4

Semantic defaults

57

transitional between the semantic form and a truth-conditionally identifiable content, then POL would be violated and, in the most radical case of underspecification requiring both completion and expansion, we would have to distinguish between (1) semantic form, (2) 'prepositional' content of what is said, and (3) the 'true' prepositional form which contains all the necessary information for the assignment of truth value. 'Necessary' rather than 'all' because otherwise the prepositional form would contain too much information. Here we require a separation between truth-conditional and non-truth-conditional meaning by shifting some implicatures below the line of prepositional content. In other words, we require only that part of pragmatic information which contributes to truth conditions. This much is widely acknowledged.13 After all, if any meaning that is not derived by linguistic decoding was implicated (as Grice claims, see Carston 1988), then in many cases we would not be able to arrive at a single proposition since decoding need not necessarily always precede inferential processes. Or even, as Escandell-Vidal (1996 and personal communication) claims, decoding is always preceded by inferential processes which result in building a schema or frame, ready to admit decoded information. To sum up, what is communicated by the utterance of (1) may differ depending on the context of interpretation. In order to account for this difference, one could follow either of two procedures: follow the radical pragmatics path and approve of the generality of sense, or construct an account of intentions that secure reference. In order to avoid being between Scylla and Charybdis, a convincing decisive argument is required. Both traditions are acclaimed to enjoy a certain psychological plausibility, one derived from Chomsky's account of Universal Grammar and the other from the idea of phenomenology including the intentionality of acts of consciousness. Both pertain to innate abilities of the mind, with the different scope attributed to linguistic phenomena.14 So, if we begin the account of utterance interpretation with the specification of intentions, we obtain, in most cases, a truth-evaluable level of prepositional form as the first-level sense because the referent is secured. In other words, POL is satisfied. The level of an underdetermined semantic representation is then excluded by definition: securing reference is the most natural and expected process in utterance interpretation performed at an early stage, after some initial pre-decoding

13

Cf.: "A number of philosophers have contended that Grice completely overlooked the fact that inferential processes of essentially the same sorts as those involved in implicature enter into determining what is said." Bach (1994a: 269-270). In other words, what is implicit in what is said differs from implicature sensu stricto. 14 For an account of the vehicles of thought see Chapter 6 and also Dummett 1991a; Jaszczolt 1996a.

58

Discourse, beliefs, and intentions

Chapter 2

inferences, and necessarily simultaneously with decoding.15 'Necessarily', because securing the referent is automatic on hearing the expression that is used to refer, just as the quantifier analysis is automatic on hearing an indefinite description. And this tendency to secure a default, unambiguous interpretation has to be captured by any cognitively plausible theory of communication. Kamp and Reyle (1993: 288-304) call it a general tendency of a type of expression that has to be captured by construction rules for discourse representations, i.e. rules for building their intrusionistic semantic representations of utterances. The conjecture to be made at this point is that the underspecified semantic representation may lack psychological plausibility: normally, intentions secure the object of the discourse and the unique reading of an utterance. The idea behind POL is that full prepositional representation can be equated with the semantic representation. This proposal was tested here on definite descriptions. But this formulation of the idea is only the tip of an iceberg and thus requires detailed justification. I have indicated that the core of this justification lies in intentions in communication as discussed by Grice (1969) on the one hand, and the tradition in the speech act theory on the other (cf. e.g. Searle 1983,1992; Bach and Harnish 1979). The resolution is given in the following section, followed by a presentation of its application to some chosen examples.

2.5. Default Semantics? 2.5.7. Default Reasoning The idea of semantics based on default interpretations appears to be not so controversial when derived from the account of default intentions and default reasoning. It is founded directly on Grice (1978) and Bach (1984). Over a decade ago, Bach argued that utterance interpretation is governed by the principle of default reasoning and 'jumping to conclusions' because we 'know when to think twice'. The view was based on the postulate of degrees of belief and intention but has never been developed into a fully-fledged account of utterance interpretation. The present account provides the missing theory16 It is demonstrated that degrees of intentions account for the existence of more and less salient interpretations of potentially referentially ambiguous utterances. 15 As Escandell-Vidal (personal communication) observes, pure decoding may have no place in this account. 16 On the proposed development of Bach's account see Jaszczolt 1998e.

Section 5

Semantic defaults

59

Bach observes that in order to work out his sketch of default interpretations into a full theory of default reasoning one needs an account of beliefs and intentions because beliefs and intentions influence reasoning in an important way. Here he suggests the "degree of strength of belief and intention" (1984: 49), without elaborating on this suggestion. Now, putting POL, PI and DI together provides such a development. They account for a salient interpretation which constitutes the 'first unchallenged alternative'.17 Bach proposed that in the case of referring to an entity, there is a referential intention present which constitutes a part of the communicative intention (Bach 1987a), and that reasoning is principally governed by a 'taking for granted' principle, it proceeds to the first unchallenged alternative (Bach 1984), i.e. to the first sense of the utterance that springs to the hearer's mind and is not immediately refuted in another thought. Default reasoning is automatic, it relies on generalizations and stereotypes of everyday life. People normally take it for granted that the interpretation that springs to mind is correct; no further reasoning process is involved. They do so in accordance with the rule that makes them calculate costs and effects of making judgements and decisions. In other words, in answering the question of priorities of interpretation of utterances with referring expressions, one is governed by the rule that there should be no expenditure of processing effort in utterance interpretation if it does not come with an additional gain in cognitive, information-providing or information-enriching effects. This balance is guaranteed by the principle of co-operation in conversation, be it Gricean (1975), neo-Gricean (Levinson 1987), or relevance-theoretic (Sperber and Wilson 1986a). The compatibility of Bach's account of conversation with the principle of co-operation is evident from the following statement: "... I take for granted that the first thought that comes to mind is the right one - unless some reason against that or some alternative Bach (1984: 45). comes to mind."

17 The intention-based account of meaning illustrates one general point remarked upon by Thomason: if intending to assert/? does not differ from intending to assert q, then the meaning of;? and q is the same. Intensionality of meaning and intensionality of intention go hand in hand (see Thomason 1990: 349). As Thomason observes, Grice's maxims are too loose to account for intentions in conversation. He suggests a theory of shared plans and goals. Thomason hopes for a theory of default reasoning that would be on a par with conversational implicatures. It is my belief that such a theory has to be based on intentions in communication, default reference, MOR, POL, and above all parsimony of processing, controlling and controlled by gained information.

60

Discourse, beliefs, and intentions

Chapter 2

Jumping to conclusions is a cost-avoiding and effective default. It is effective because, in Bach's terms, the hearer knows when to jump to conclusions and when to think twice. There are shortcuts in reasoning, the hearer takes things for granted and proceeds to the default interpretation, unless he/she has evidence that the 'marked' interpretation may be the case. As Bach (1984: 46) puts it, most things we do because we do not decide not to do them. Equipped with PI and DI, we can also say that defaults correspond to the strongest intentions, undipped by any conditions governing their fulfilment. Ceteris paribus, the interpretation of definite descriptions proceeds as referential because people normally talk about other individuals: we jump to conclusions, there is no need to think twice.

2.5.2. Defeasible Knowledge The idea of taking semantic forms together with other knowledge sources in order to draw inferences has been formalized with a high degree of success in defeasible logic. In defeasible logic, conclusions may follow from a set of premises but not from a superset. Applied to causal and temporal interpretation of events described in text, defeasible logic gives as an output the fixed interpretation. The relation between the descriptive order of events in text and the temporal order can be described in terms of two sets of defeasible rules: world knowledge pertaining to causal relations between events and states, and linguistic rules pertaining to Gricean maxims. For example, in (2) where the events are falling and pushing, world knowledge tells us that normally pushing causes falling (cf. Lascarides and Oberlander 1993: 13). (2)

Max fell. John pushed him.

(from ibid.: 2). This interpretation of (2) shows the backward movement of time in narration. It is an instance of a causal rule. It is a 'ceteris paribus law' and may not hold in some circumstances. Defeasible logic also uses, for example, a Narration Rule which says that the sentence order p.q normally means that the eventuality p immediately precedes eventuality q. Causation in (3) may equally plausibly go either way and the Narration Rule is responsible for not allowing an ambiguity to arise by specifying the order of precedence of eventualities:

Section 5

(3)

Semantic defaults

61

The bimetallic strip changed shape. The temperature fell.

(from ibid.: 4). Another rule says that states overlap.18 So, if the sentence order is that of an event followed by a state, the direction of causation can be inferred from this rule and the state causes the event as in (4): (4)

Max opened the door. The room was pitch dark.

(from ibid.: 2, see also p. 21 for discussion). Lascarides and Oberlander conclude that there is no ambiguity on the level of sentence/utterance interpretation because world knowledge (called the 'knowledge base') and linguistic knowledge do not allow it to arise. The only sense in which it is plausible to talk about ambiguity is that (2) can have different interpretations relative to differences in the knowledge base. But even on this level there are default interpretations. In the absence of information to the contrary, (2) is interpreted as a reversal of the causal order. This default can easily be overridden and (2) can be interpreted as conforming to the Narration Rule. This possibility is guaranteed by the theory with its assumption of non-monotonic inference: the inferred temporal order of eventualities may change as the context becomes enriched. Lascarides and Oberlander entertain three classical possibilities as far as the semantic representation of text (and by extension discourse) is concerned: (i) the semantic structure (and logical form) is neutral as to the relation between the eventualities and pragmatics secures the ordering of eventualities; (ii) the ambiguity position, where there are as many logical forms as possible connections between eventualities and the choice between them is left to pragmatics and the reader; and (iii) the logical form of the text is constructed dynamically as a pair of sets. The first set is a set of logical forms of sentences, and the other is a set of relations between the eventualities depicted in the first set. In dynamic semantics, discourse interpretation is incremental. For instance, we obtain p, and then q, from the first set and supplement them with the rule coming from defeasible logic.19 Defeasible logic ceases to be a pragmatic phenomenon and instead actively constructs the semantic representation (see also Lascarides and Copestake 1998 on defaults coming from the lexicon). This idea of dynamic logical forms is an instance of the semanticization of primary pragmatic processes also advocated by DRT and by the present account. Lascarides and Oberlander end on a non-committal note; all three placements of the boundary between semantics and pragmatics are possible. See also Dowty 1986 on a more detailed analysis of temporal relationship and aspectual class. For the precise logical forms in dynamic semantics see Lascarides and Oberlander (1993: 33).

62

Discourse, beliefs, and intentions

Chapter 2

However, it is signalled that the preferences for (iii) are rather stronger than for the other two options, at least on the part of one of the authors (see ibid.: 33 ).20 Indeed, the dynamic perspective best fares with default interpretations. Intentions and other contextual information fit in as interpretation proceeds and intrude into the process of constructing the semantic representation. This standpoint bears obvious affinities to the position defended in this book.

2.5.3. Defaults for Definite Descriptions: Interaction of Intentions It was pointed out above that the referential use of definite descriptions makes them akin to proper names in their referring properties. Let us now have a look at these two types of expressions. I claimed that the degrees of referential intention indicate the 'degree of referentiality' of the interpretation, or the degree of remoteness from the referential interpretation of the utterance under scrutiny. These would include utterances of sentences with proper names21 as well as definite descriptions in the subject NP position in extensional contexts as in examples (5)-(6), and in sentences embedded under the prepositional attitude verb, as in (7). (5)

Ortcutt is a spy.

(6)

The man with a martini is flirting with your sister.

(7)

Ralph believes that ... + (5) or (6).22

It was proposed that the referential reading corresponds to the strongest referential intention. Intuitively, this reading constitutes the presumed, salient interpretation. The reading that involves sorting out the speaker's referential mistake corresponds to a 'weaker' referential intention. It is the reading in which, so to speak, something goes wrong in the act of referring. The nonreferential reading corresponds to the lack of referential intention. This reading is only activated when the salient, default one is contextually prevented from being the case. In the latter case the speaker does not talk 20

In subsequent papers, Lascarides researched rhetorical structure of discourse in terms of a version of discourse representation theory. See e.g., Asher and Lascarides 1995. The paper also includes some general heuristics for the disambiguation of lexical expressions, albeit amounting to spelling out what is common-sensically well known rather than to novel rules. They pertain to avoiding incoherence and the reinforcement of rhetorical connections. 21 Proper names are normally regarded as directly referential terms. However, they can also be accompanied by a stronger or weaker referential intention. 22 Pronouns and demonstratives are left out of the picture because they pose slightly different problems of referring and, in any case, the referential intention is there assumed to be present (see Bach 1992 on intentions and demonstrations).

Section 5

Semantic defaults

63

about any particular entity but rather about the event or state as a whole. Then, if one can speak about referring in such cases, it is referring to the state or event. Referring to an eventuality corresponds to a 'weaker' referential intention than referring to the constitutive parts of the eventuality. It can be graded as corresponding to the weakest form of intentionality because in referring to an individual, the speaker, so to speak, refers both to the eventuality audio the individual as a constitutive part of this eventuality. One could even say that referring to an eventuality is in fact equivalent to the lack of a referential intention. According to Bach, the communicative intention accounts for transmitting propositions between interlocutors (see also Jaszczolt 1996b, 1997a). The presence of the referential intention makes the communicative intention 'stronger': referring adds to the overall strength of intending to communicate and thus to the conversational value of the recognition of the intention (as well as to the conversational value of the communicated attitude). By the conversational value I mean here informativeness, or a ceteris paribus informativeness, presuming that both referential and nonreferential interpretation of the speaker's utterance are relevant to the hearer. It remains to be seen how this account of referential intentions fares with the semantics/pragmatics interface. By entertaining the presumption that the referential intention is present in its strongest form, the hearer supplements the linguistic meaning with the referent as if this reference belonged to the linguistic meaning, if nothing in the context indicates otherwise. There is a gradation of attachment (or, perhaps, incorporation) of the referential intention to the linguistic meaning and the possible interpretations are not equally salient: the strongest referential intention corresponds to the default reading. In the default referential case, the linguistic meaning leads directly to the propositional form as far as reference is concerned if no further expansion or completion is required. The further the interpretation is from the default, the smaller role can be ascribed to the referential intention. In this way we have the discourse representation as a sum of the syntactic, semantic and pragmatic information in the form of accessible defaults and less accessible departures from defaults. Intuitively, one can concede that on most occasions we talk about real people and real objects and refer to them explicitly. So, there is a tacit presumption that the referential interpretation is the default. Now, why the strongest referential intention corresponds to the default reading can only be answered by negative arguments at this stage, that is envisaging the difficulties with postulating the opposite or no defaults. The positive psychological argument from intentionality and the workings of our mental states, signalled only briefly above, has to wait until Chapter 3.

64

Discourse, beliefs, and intentions

Chapter 2

Coming back to Donnellan's distinction, it has been remarked about it that "... in some cases, we use the description to get to the referent and then throw away the description as it were" (Levinson 1988: 47). The hearer does so in the case of a referential interpretation of a mistaken description. In a sense, he/she does so in all cases of referentially used definite descriptions. He/she gets to the object as if by default: normally, people talk about known, existing entities. Thanks to the presence of such preferences in conversation, we can say that this process belongs to dynamic, context-sensitive semantics. Levinson (1988: 62) gives here a convincing example to illustrate the thesis that enriched semantics is preferable to post-semantically, pragmatically resolved reference: (8)

The ham sandwich is getting restless.

In (8), the definite noun phrase refers, obviously, to a client who ordered a ham sandwich. Levinson claims (after Sag, for references see ibid.: 62) that denotations should be allowed to change within semantics, for instance along the lines suggested by Kaplan's (1989a) distinction between character, which is a function from contextual indices to contents, and content.23 As is well known, instead of talking about reference, Donnellan suggests talking about reference attribution by the speaker. He adds that the referent should be "historically or (...) causally connected to the speech act" (Donnellan 1970: 356). But the theoretical explanation of this link is not provided. A theory of the recovery of the intended referent founded on a theory of intentionality can fulfil this role. After all, who (if anybody) the speaker referred to is a matter of the speaker's intentions, or rather their recognition by the hearer (Donnellan 1966: 297). And so is the referential or attributive use of a definite description. So, the ambiguity of sentences such as (6) is neither properly semantic nor pragmatic.24 It is a semantic ambiguity only when we accept a 'static' semantics that lists the truthconditional impact a description can have in all sorts of contexts, including intensional ones such as (7). The seeming interpretative ambiguity is successfully accounted for by the dynamic semantic representation that allows for default interpretations of the utterance. The utterance is normally unambiguous and its interpretation proceeds towards the default reference, unless some contextual impediments stop it from doing so. And I have suggested that in the case of definite descriptions intentions are strongest when the speaker communicates a proposition about an individual known 23

Cf. also Jaszczolt 1997a, fn 4. I discuss this issue in more detail in Chapter 4. In one of my earlier papers (Jaszczolt 1993) I claimed that in Polish there are lexical and syntactic means of distinguishing between talking about a known and an unknown individual in belief reports (de re and de dicto). However, the use of this distinction is also pragmaticallydriven to some extent. 24

Section 5

Semantic defaults

65

to the hearer. The hearer automatically processes the utterance as if it concerned a particular, known individual. Nota bene, both the communicative and the informative intention are in their strongest form: the particular person is spoken of and it is intended that this information is recovered by the hearer. If, on the other hand, the speaker communicated a proposition about whoever fits the description, the informative intention would make the process of reference assignment stop at a certain stage: either at the very beginning, if the informative intention was strongly supported by the communicative one, or later on, if the communicative intention was weaker. This is due to the interaction of the primary, communicative and informative intentions. The strongest intentions correspond to the smallest effort in processing the message on the part of the hearer. This is the default situation: the message is optimally informative, minimally demanding in assumption recovery, and thus the situation is optimally advantageous for securing relevance and informativeness. Intentionality, directedness at a referent, is secured without going through an intermediate, underspecified stage in utterance interpretation: the person or object is intended and this message 'automatically' gets through to the hearer.

2.5.4. Defaults for Indefinite Descriptions In utterances with indefinite descriptions such as sentence (35) from Chapter 1, repeated here as (9), when accounted for by means of primary, communicative and informative intentions, it seems natural to admit that the nonspecific interpretation is rendered as a default: (9)

A friend of mine from Oxford paid me a visit.

Indefinites are not referring expressions, the primary intention is missing and this fact allows the hearer to read more information off the semantic representation than would be the case when the primary intention was in operation, i.e. in the case of definite descriptions. The semantics is not ambiguous: indefinites do not exhibit referential properties. They can, however, be specific. Even then their aim is not to secure a referent but rather by their means the speaker communicates to the hearer an event, at the same time making the hearer alert to the fact that no stronger statement was necessary or possible. The specific and nonspecific readings can be explained as follows. The thought of the speaker exhibits intentionality because all acts of consciousness do. This intentionality can be stronger or weaker. However, the utterance does not exhibit the primary intention of securing the referent. The primary intention does not interact with the informative and the communicative

66

Discourse, beliefs, and intentions

Chapter 2

intentions. This situation can be engendered either by the weakness of the primary intention of the mental act (the nonspecific reading of the indefinite) or by the conflict between the primary intention of the mental act and the informative intention (the specific reading; the referent may not be of interest in the conversation). In any case, the utterance is permeated with the primary intention of the speaker's thought and the specific or nonspecific interpretation is established between the speaker and the hearer. To repeat, the nonspecific interpretation occurs in the case where, according to the hearer, the intentionality of the mental state of the speaker is the weakest, i.e. it fails to reach the object. The specific reading is rendered when the speaker's thought is recognized as strongly intentional but the utterance fails to reflect this primary intention due to the interaction of the communicative and informative intentions that signal the irrelevance of any precise act of referring for the particular purpose at hand. In other words, the underlying thought is either general and the primary intentionality is unrealized, i.e. does not reach the object, or the thought concerns an intentional object but this fact is not communicated at its face value. Let us take stock. Ludlow and Neale (1991: 184) assert that the uses of indefinite descriptions do not prove the existence of a separate, 'semantically' referential interpretation. That is true. And so is the claim that 'donkey'-sentences of the form as in (10) are not ambiguous between the singular and plural interpretation of the description (cf. ibid.: 198). (10)

Every man who buys a donkey vaccinates it.

But proceeding to assigning one of the possible readings has to be more fine-grained, level-neutral, and allow for the dynamism of semantic representation and, a fortiori, for default interpretations.25 Sense need not be unspecified, it just needs to be constructed in situ as default sense or departures from the default, predicted by the recovered degrees of intentions. Only in this way will one avoid multiplying the levels of senses beyond necessity.

2.6. Other Applications 2.6.1. On Semantic Representation and Negation Again It remains to demonstrate that the defaults are indeed semantic. In Chapter 1, I suggested that sentences involving negation such as (1) can have a semantic 25

There are serious implications here for the distinction between being a Russellian and being a referentialist. See Neale 1990. But the opposition may after all boil down to being founded on a terminological mismatch. For arguments in favour of such a reconciliation see Recanati 1993.

Section 6

Semantic defaults

67

representation as in (Ic) repeated here as (11): (11)

- Bald (k)

where 'k' is an individual constant. This seems to be a truly referential reading of the definite description that accommodates it to proper names. The referential reading is also the presumed default interpretation as it corresponds to the strongest primary intention. Hence, the intrusion of the referential intention into the semantic representation seems to signal that there is no separate level of underspecificity: there is no obvious route from the representation with the wide scope of the quantifier to (11) other than by instantiation, and allowing for instantiation would mean letting in semantic ambiguity. The seeming circularity of this reasoning is prevented by the intentionality which is an inherent feature of acts of consciousness.

2.5.2. And, or, and if DI as formulated here also allows for intentions other than the referential to be gradable. This feature enables an explanation of the interpretative ambiguity of some other sentential connectives such as conjunction, disjunction and implication. In sentence (20b) from Chapter 1, repeated here as (12), we can conjecture that the strengthening of the meaning of the connective & corresponds to the strongest communicative and informative intentions. (12)

They had a baby and got married.

As a default, and will mean 'and then'or 'and as a result'. Similarly, if will render 'iff', and or will stand for the exclusive disjunction. In such an interpretation, the contextual effect is strongest and the intention to inform of the situation is also strongest. If we took the semantically poorer sense to be the default, ambiguity would flourish and the corresponding intentions would be present in their crippled, inhibited form. The third option, that of no defaults, is eliminated by force of MOR and POL, and because an interpretation tends to be easily secured by the hearer, as confirmed by common-sense, everyday observation. But this is only a temporary suggestion. After all, in (13), one would be in need of postulating a rule for non-sequentially (cancelling 'and then') and non-consequentiality (cancelling 'and as a result') of and, perhaps coming from the fact that it is immediately diaphanous to the hearer that no sense-connection is meant between the conjuncts: (13)

He is a novelist and he has three children.

68

Discourse, beliefs, and intentions

Chapter 2

2.6.3. Number Terms Numerals are equally problematic but I believe they can conform to default semantics. Sentence (24) from Chapter 1, repeated here as (14), can have a default interpretation arrived at through POL and DI: (14)

Max has three children.

In this interpretation, 'three' means 'exactly three' as this is the case of the strongest informative intention - and, of course, the highest informative content. By definition, quantifier expressions of this sort do not exhibit a primary intention. However, by reasoning analogous to that for the three connectives above, we can have an instance of the strongest communicative and informative intention. It is so because, by DI, both relevant types of intentions allow for degrees. Meaning 'at least three' is also possible. Let us imagine a context where a parent who has three children qualifies for social benefits. On this reading, the object of the predication is only partially revealed and intentionality is weaker. The situation can be explained as follows. Although PI does not apply to numerals, the referential intention frequently accompanies a predication. Uttering (14) is quite likely to induce (15) or (16) in the hearer who is presumed to be acquainted with Max and his family: (15)

What are their names?

(16)

What do they do?

If the postulate of the strength of intentions incorporated into semantics is on the right track, then perhaps they can be seen as standing in a fixed inverse relation to processing effort. And intentions are best seen as being directly proportional in their strength to the informative content of the expression, as strengthening of and best demonstrates. I have not said anything about the resolution of the relative scope of the quantifiers if more than one is involved, and with the collective and distributive reading as posed by (17). (Again, the ambiguity of reading is there only in theory because even here defaults seem to shine through.) (17)

Six men carried two pianos upstairs.

This delimitation of the number of readings is only a preliminary step. Yet, the fact that the addition of 'each' would be meaningful seems to suggest, in agreement with the Co-operative Principle, that its omission is significant and points towards the collective reading. Kamp and Reyle (1993: 321) point

Section 6

Semantic defaults

69

out that any satisfactory paraphrase of a collective reading would have to refer to the fact that the effort of the individuals is combined - either physically or in its outcome. They quote examples where the collective reading is obligatory or at least is strongly preferred and argue that the distributive reading is best regarded as obtained through an optional further step after the collective reading has been obtained (see ibid.: 320-330). The intention-based account succumbs to this strength of the collective interpretation and makes it a default by attributing a primary referential intention to a set, thus incorporating collective intentions which are not reducible to individual intentions. One can plausibly argue that collective intentions can be shown to interact with the other three dominant intentions in communication and strengthen them by strengthening the primary, referential intention. As a result, the default interpretation is produced and it is the collective one. The route to the default interpretation can be directly derived from the interaction of intentions presented in the case of definite descriptions, with the addition of the collective intention. At the moment, this is as far as the account leads, extending Searle's (1990b) view on collective intentions. These are only tentative proposals. The difficulty lies in the fact that intentionality is to no avail. It is simply not decisive. I shall have no more to say about this type of 'ambiguity' and will mainly concentrate on the workings of the referential intention for referring expressions in extensional and prepositional attitude contexts instead. As a final positive note, however, it has to be pointed out that there is significant progress in accepting the theory of default semantics. Geurts (1998), for example, tries to reconcile sense-generality and Horn's pragmatic ambiguities of negation by saying that while some expressions have punctual semantics, others have the 'at least' meaning. As was discussed in Section 1.2.6, he ascribes these differences to the lexicalization of scalar implicatures, for example in numerals, and the lack of this lexicalization in, for example, adjectives such as 'warm' which has semantic meaning of 'at least warm'. Now, Geurts (ibid.: 298-299) defends this diversity by saying that it does not go against Occam's razor because the lexicon just is like that and polysemy has nothing in common with methodological parsimony. In any case, even polysemy is excluded because lexicalizations lead to default meanings. But we have gone further than that. Instead of resorting to 'this is what the world is like', we have elaborated here a constructive explanation why the world, and the lexicon, are like that: they conform to and reflect the interaction of intentions in communication, and at the same time they reflect the workings of the intentionality of mental states attended to in Chapter 3.

70

Discourse, beliefs, and intentions

Chapter 2

2.7. Defaults in Dynamic Semantics Defeasible logic introduced in Section 2.5.2 was demonstrated to fare well with logical forms constructed according to the incremental rules of dynamic semantics. Sentences have default semantic representations and these are constructed as discourse proceeds, with the help of ever increasing contextual information. In this section it is suggested that the same rule of default interpretations can be applied to accounts within Discourse Representation Theory. An analysis of referring constructions in DRTwill have to wait until Chapter 7. First, I discuss the notion of a prepositional form that constitutes part and parcel of an 'intrusionistic' dynamic semantic representation. I shall again take on board the theses of semantic ambiguity and underdetermination and demonstrate that, to a large extent, the polemic is driven by a terminological vagueness and confusion. I shall also summarize the advantages of building a semantics around the idea of default interpretation. The difficulty with defining semantic representation is engendered by the fact that there is no uniform connection between the output of grammar and the situation referred to by the linguistic string. The interpretation of the linguistic string does not undergo any strict semantic rules. But since neither does the situation itself, what is required is a restatement of the relation in more dynamic terms. The reality changes by the moves and actions of the participants and the interpretation of a linguistic string changes with more information becoming available, along the lines described by Heim's (1988) File Change Semantics where the task of the hearer is, so to speak, to construct a file and regularly update it so that it contains all the necessary information revealed by the speaker. So, for indefinite expressions the speaker starts a new file card (introduces a new discourse referent, cf. Karttunen 1976), and for a definite, an existing file card is updated, according to the conditions specified in the semantics. Files, interpretive devices that mediate between the language and the world, perform the role of a monitor that records all the changes of the situation, including both linguistic and non-linguistic information such as that from perception and background knowledge. They do not rely on logical forms (in Chomsky's sense) but rather overlap with them since logical form contains a lot of grammatical information which is irrelevant for constructing meaning. Lappin (1991) makes an even stronger claim that the LF is redundant as a level separate from the S-structure to satisfy MOR and, as we can infer, POL as well. One of the main arguments for this standpoint comes from the ambiguity of the scope of quantified noun phrases that do not seem to require a unified LF in order for one or the other semantic representation to be assigned to them (cf. ibid.: 310-313. See also

Section 7

Semantic defaults

71

Kempson 1991/92 on how syntax and semantics are intertwined). So, even within the domain of syntactic analysis, logical form may be overly informative. Changes in the situation may affect the truth value of an utterance. They cannot be recorded after the propositional form has been constructed because they affect the truth-conditional meaning of the utterance. Heim's solution evades this difficulty by introducing a level on which both utterances and other sources of information write down anything that is required. Essentially the same relation between the two components is proposed by Kamp's DRT. Levinson calls this perspective 'pragmatic intrusionism': "... there is a common slate, a level of propositional representation, upon which both semantics and pragmatics can write..." (Levinson 1988: 22). Pragmatic intrusionism is a more cognitively plausible model of utterance interpretation with its semantic and pragmatic aspects assigned an equally privileged status. Discourse representations are mental representations and are, so to speak, two-aspectual: they rely on model-theoretic semantics but also concern the meaning grasped by the hearer on hearing an utterance. The theory thus has to be composed of three stages: a generative syntax, rules deriving representations from syntactic constructions, and mappings from representations to a model, supplied with a definition of truth. Logical ambiguities vanish on the level of deriving a representation, thanks to the various possibilities of applying DRS construction rules that account for the diversity of meanings (cf. Kamp and Reyle 1993). A discourse representation shows what the world would have to be like for the sentence to be true. Its construction rules account for the introduction of new referents and for adding properties (or, generally, DRS-conditions) to the existing referents, which is essentially based on the linguistic representation. When more information becomes available as the discourse proceeds, the representations are filled in with more properties and referents (see ibid.: 59). This pragmatic intrusionism into semantic representations can be partial or total. Levinson (1988: 22) uses the following metaphor. The semantic and pragmatic contributions to the 'common slate' can be distinguished by the colour of the ink: "Semantics and pragmatics remain modular 'pens'as it were: they are separate devices making distinctively different contributions to a common level of representation." This is partial intrusionism. Total intrusionism would not preserve the two colours of the ink but rather mix them into one resulting colour. 'Two colours' means 'two modules', whereas there is no reliable evidence that pragmatics is a module, while there is significant evidence to the contrary (cf. e.g. Wilson and Sperber 1986b). So modules may not be a suitable jargon for discussing

72

Discourse, beliefs, and intentions

Chapter 2

intrusionism. Levinson (1988: 78, fn 29), while defending modularity of pragmatics, claims that rejecting it does not alter the story: semantics and pragmatics remain different 'pens'. But 'different pens' and resulting colours of the ink are not the same thing. After all, once the semantic representation has been embellished with results of pragmatic processing, the colours may merge and there may not be any need to postulate separate components in utterance interpretation. The existence of default interpretations guaranteed by intentions in communication seems to point towards a merger rather than intrusions: intentions accompany an act of reference rather than follow it. This is so both for the order of explanation and the psychological order. Levinson (ibid.: 60) rejects such a merger26 because semantics is monotonic and pragmatics non-monotonic. But it is not clear why the difference in monotonicity should be an obstacle. If discourse processing is incremental, then surely monotonic and non-monotonic contributions can cohabit the 'common slate'. All that is sacrificed seems to be modularity, which is a controversial property anyway. A DRS is true if there are individuals in the universe of discourse that correspond to the discourse referents, and the conditions of that DRS predicate something that can be truthfully said about these individuals. However, the interpretative ambiguity remains here as a full-blown ambiguity of processing options: in the case of indefinite NPs, for instance, the hearer can process the utterance to render the specific or the nonspecific reading, depending on the DRS in which a discourse referent for 'an jc' is introduced. The appropriate reading can thus be achieved by means of choosing the appropriate order of rule application or by postulating into which DRS (in the case of embedded DRSs) the referent is to be introduced (see Kamp and Reyle 1993: 288-293; Jaszczolt 1998d). The theory does not predict under what circumstances the hearer will follow one or the other option, but neither does it postulate an ambiguous unit or an underspecified representation. Whether there are defaults or merely contextual preferences, the process of utterance interpretation relies on the whole discourse situation that is available and avoids multiplying stages through which this interpretation must proceed. Ambiguity is present because the sets of information contributing to each of the alternative DRSs are different in various respects, both semantic and pragmatic. But it is an ambiguity of the sentence rather than the utterance. Given a sentence, the interpretation of its various utterances can go either of the prescribed ways. Given an utterance, it proceeds along one particular way because of the default interpretation which is engendered by the joint Called by him 'pragmantics' and attributed to generative semanticists. It should be clear that a principled intrusion of intentions into semantic representation is far from the type of merger advocated by generative semanticists in the 1970s.

Section 7

Semantic defaults

73

commitment, i.e. by sharing of the discourse referents between the interlocutors (Kamp 1990) and, if applicable, because of overriding the defaults. According to Kamp, there are also devices of formal and external anchors for directly referential terms in the semantic interpretation: in order for a representation (DRS) to have truth conditions of a singular proposition, the DRS has to have a clear referent, i.e. be connected with an entity by an external anchor. Kamp (1984) believes that the structures that people form during the cognitive process of utterance interpretation resemble the representations of DRT. In cognitive processing, as in DRT, building an interpretation can proceed in various ways. For instance, anaphoric pronouns introduce referential ambiguities. Background assumptions about the world help select referents in such problematic cases. Principles of inference in communication, forming intentions out of beliefs and desires, resemble principles of inference in formal logic and are applied to items with syntactic structure, i.e. logical formulas (Kamp 1990). Now, these items do not yet constitute a semantic representation. Discourse understanding is incremental, proceeds bit by bit and in the process it accommodates changes of the context. In our account, some of these bits are inherently equipped with triggers of intentions, and intentions prevent the semantics from being ambiguous. Instead, they secure the reading either as the default, strongest one or a weaker one when intentions are weaker. For instance, who is meant by using a proper name which has many potential bearers depends on the context - including the speaker's intentions (Kamp and Reyle 1993: 62). Also, in (18), the indefinite description can be represented differently in the DRS depending on the context. The discourse referent for the indefinite description a Porsche has to be introduced in the main, transparent, superordinate DRS which corresponds to the widest scope of the existential quantifier that handles 'a Porsche': (18)

John doesn't like a Porsche. He owns it.

(from ibid.: 106). But there are no strict conditions for when this should be done. Perhaps the issue would be less complicated if intentions were fully applied in a sense of using the fact of reading information off the world or the inherent lack of this process. Then the hearer can be said to arrive at default readings and consecutively weaker readings that correspond to weaker intentions. Only if this schema is of no help, true communicative, interpretative ambiguity ensues. The problem also arises in sentences of the type (36) from Chapter 1 repeated here as (19), as well as (20) and (21): (19)

A problem about the environment preoccupies every serious politician.

74

Discourse, beliefs, and intentions

Chapter 2

(20)

Every boy in Mary's class fancies a girl who Mary doesn't know.

(21)

Bill doesn't know a book that I have read during the past four weeks.

(from ibid.: 279, 288, 303 respectively). Different readings of these sentences owe their existence to the different scopes taken by the indefinite noun phrase. Kamp and Reyle suggest here two possibilities of accounting for the phenomenon: relaxing the order of application of rules for constructing DRSs or externally imposing a status of a quantifier or a referring expression on the indefinite, according to the hearer's judgement. Since such judgement is decisive in the theory, we can infer that it has a role to play in DRS construction and as such it prevents the occurrence of an ambiguity. It does so because the (cognitive) process of DRS construction intentionally aims at one DRS, the one that represents the reading recognized as intended by the speaker. This reading is congruent with the intentions of the speaker but also is collaboratively achieved by the speaker and the hearer in the process of communication. So although "... the processing of indefinite noun phrases must allow for a considerable spectrum of alternatives" (Kamp and Reyle 1993: 292), these are alternatives to the default reading which is obtained by the ordinary, default application of construction rules. By default, indefinites are treated as quantifiers. There is no reference-securing intention involved which would signal direct referentiality as is the case in the default reading of definites. But, as was proposed above, this intention may be present in other non-default readings, although in competition with the informative intention for a surface appearance. It is sometimes signalled by the presence of words such as'certain','particular','given', and relative clauses. The authors say that in such cases indefinites are processed as definites and obtain their default, referential interpretation. I have suggested that they are still processed as indefinites but the primary intention of the mental act affects the informative intention of the utterance. Kamp and Reyle admit that they do not know how to distinguish between different scope readings, i.e. between 'long' and 'short' indefinites and they have no good processing rule to account for this distinction. All they do is to postulate various processing options available to the hearer. However, the hearer 'chooses' the processing option automatically in the context of the conversation, not as one of the possible two. Since there are various reasons for scope ambiguities, one cannot fit the examples into one standard mould of logical variations of scope. There is no unique rule they would conform to. Any such rule would have to be psychologically plausible and the translations from natural language into first-order logic are not sensitive to this plausibility (cf. ibid.: 304). The ambiguity is thus relegated to the level of utterance and resolved by utterance processing that relies on dynamic semantics allowing for pragmatic intrusionism. In this way the ambiguity-

Section 8

Semantic defaults

75

nonambiguity dilemma is exorcized for these cases. It can be successfully replaced with dynamic semantics, combined with contextualism and intentionality. I shall return to the contextualism -— anti-contextualism debate in Section 7.5 below.

2.8. Interim Conclusions To sum up: I have proposed an account of utterance interpretation which develops the ideas of referential intention and default reasoning. It conforms to the Gricean principle of not multiplying senses beyond necessity. It does not postulate referential ambiguities but rather utilizes intentions in a way which allows for the discrimination between more and less salient interpretations of potentially referentially ambiguous expressions. Semantics remains uniform, unambiguous, but fully truth-conditional thanks to securing a unique reading of a sentence by means of intentions incorporated in the dynamic framework. Rather than underdetermination and underspecification resolved later on by pragmatics, what we get is default interpretations. Only by accepting this standpoint can we explain the attributive use and referential mistakes of definite descriptions, as well as the multiple logical ambiguity of numerals. The alternative readings are simply the weaker cases of intentions: primary, referent-securing intention in the case of definite descriptions, and communicative and informative intention in the case of indefinites, numerals, and other cases where the Primary Intention principle is not applicable. They 'mix in' various degrees of information read off the semantic form and intentions in order to render a truth-evaluable representation. The semantics is thus flexible and dynamic: the semantic representation can be strengthened or weakened by intentions and the process is not temporal. In the default case there is no underdetermination as a step in the ladder. Kamp (1984) and Kamp and Reyle (1993) allow for the dynamism to be achieved by contextual specification; Recanati (1994: 166) opts for methodological contextualism, that is a weak version of contextualism according to which the linguistic meaning of an expression is not the same as the contribution the expression makes to the proposition expressed. And as it has been demonstrated here with help from intentions, the dynamism can be taken 'higher up' in the proposition-implicatures schema, to affect the semantic representation. As an additional bonus, we can now observe that the ambiguity-nonambiguity dilemma disappears and turns out to be merely a terminological mismatch. In our analysis, ambiguity is precisely such a flexible (and hence uninteresting) construct. For example, referring in discourse yields to a unitary semantics, blurs the distinction between linguistic meaning and primary pragmatic processes and adds salient, default interpretations and

76

Discourse, beliefs, and intentions

Chapter 2

departures from the default that correspond to the strength of the referential intention with which the utterance was conveyed to the hearer and understood by the hearer to be conveyed.

2.9. The Doubly-Dynamic Approach: Assumption Recovery or Creation? If the communicative intention is envisaged in the light of the theory of relevance, then what is communicated is entirely hearer-dependent. However, it seems at least equally plausible to conceive of conversation as a process of assumption creation performed in-between the speaker and the hearer, assigning to the hearer a more responsible role than just recovering the already existing assumptions. In fact, Bach and Harnish (1979) treat communication as a co-operative process, based on Grice's idea of speaker's intentions addressed at the hearer. I am not sure whether, if the above is true, one has to dispose of the default role of the communicative intention in constructing meaning or make it intersect with the informative one, since the fulfilment of the communicative intention is its recognition (Bach 1987b). It is certain, however, that the hearer is also invited to construct meaning rather than merely recover it (Levinas 1961; Jaszczolt 1996b), and thus that the communicative intention may matter to a lesser degree in communication: making it manifest that 'this is what I want to say' can perhaps be overridden by a principle of how exactly the speaker and the hearer contribute to the meaning that is being established between them in the process of conversation.27 It is important to get this perspective on utterance interpretation right. The strength of intentions may be judged differently according to the adopted perspective and, above all, the degree to which the hearer is free to construct assumptions is an important claim of any pragmatic theory. Sperber and Wilson (1986a) suggest that the role of the hearer is rather passive in that the hearer's aim is to recover the speaker's assumptions. This suggestion is worth a closer look as it may not stand up to scrutiny. I suggest in this section that processes other than intention recognition affect the derivation and creation of assumptions. The process of discourse interpretation is 'doubly-dynamic', i.e. interpretation is created in-between the interlocutors and the hearer may be given freedom to create assumptions rather than recover them. I propose to bring together two theories coming from quite different traditions: relevance theory on the one hand, and the theory of conversation included in Levinas's Totality and Infinity (1961) on the other. While the first is a theory of human cognition and communication, the other A similar standpoint is signalled by Bird 1994.

Section 9

Semantic defaults

77

comes from the French phenomenological tradition and addresses the issues of the structure of being and the lack of mastery of a person over him/herself. It is the latter issue that for Levinas finds its explanation in conversational interaction and which, rather inadvertently, finds its application to the reanalysis of the speaker's and hearer's role in meaning-construction. According to relevance theory, relevant information is derived from the utterances, memory, observation, and the process of inference. However, the theory leaves the process by which such an interpretation is reached largely unexplained. The authors pose two unanswered questions, namely (i) how assumption schemas are filled out, i.e. what is the mechanism of providing an utterance with the interpretation, and (ii) what determines the order of accessibility of hypotheses in the process of utterance interpretation, i.e. what factors in the discourse situation affect the fact that certain interpretations come to the hearer's mind earlier than others and seem to be more natural than others (Sperber and Wilson 1986a: 170). The answer should be expected along the lines of the relation between communicative behaviour and psychological states such as intentions. Sperber and Wilson (1986a), Grice (1989), Recanati (1993), and many others stress that meaning should be analysed in terms of psychological properties, along the path leading from sentence meaning, through utterance meaning and communicative behaviour, to psychological states (cf. Recanati 1993: 20). As yet no successful account of psychological states has been given. It is my belief that states such as fear, anger, annoyance, etc., as well as many non-emotional states cannot be excluded from the study of intentions. They affect assumption recovery, assumption creation and assumption 'pruning', and hence pertain to the problems in (i) and (ii). In communication, we do not aim at the identity of thoughts, similarity is sufficient. As relevance-theorists say, the interlocutors enlarge their mutual cognitive environments instead of duplicating thoughts. Mutual knowledge is not attempted, it does not even exist (see Sperber and Wilson 1990). Instead, communication relies on mutual manifestness of what happens around us and what can be non-demonstratively inferred from knowledge. But how exactly speakers access assumptions remains an open question. In order to answer it one must analyse examples where the speaker's intention to convey certain information is particularly difficult to pinpoint. In (22), Mary's answer can be deliberately ambiguous: (22)

Peter: Do you want some coffee? Mary: Coffee would keep me awake ... [flat intonation]

(adapted from Sperber and Wilson 1986a: 34). Mary may not be sure whether she wants coffee or not, she may rely on Peter's further reaction. She may know that no additional assumptions are available for Peter concerning

78

Discourse, beliefs, and intentions

Chapter 2

her decision on coffee-drinking, or that there are conflicting assumptions available such as that she normally drinks coffee when she wants to stay awake to revise before an exam and that she realizes she should have a good night's sleep before the exam. So, she may have conflicting motives. Now, it seems plausible to say that there are no assumptions to be recovered concerning this occasion of coffee-drinking. Assumptions are left for the hearer to create. The general rule in such difficult cases of interpretation would be to judge the strength of the assumptions. It is a relatively easy task for referring expressions (including definite descriptions) because the referential intention secures the referent of the referring term in accordance with the PI principle. However, judging the strength of assumptions is more difficult in the cases where the primary intention is not applicable. Moreover, it seems that before we attempt the task of tracing the process of assumption recovery, an important methodological assumption is needed. Namely, we have to assume that the objects of beliefs are propositions and that these propositions are recognizable cross-culturally, i.e. we have to accept cultural rationalism. Sperber (1985a, b, 1996, 1997) says that culture is formed through the transmission of ideas which propagate easily, such as religious beliefs or scientific hypotheses. He calls these ideas 'contagious' and dubs the study of the process an 'epidemiology of representations'. During the process of transmission, the original information is affected by the hearer's memory and extant assumptions and becomes transformed. In this way ideas can become misunderstood or half-understood. The ideas that we hold are not always prepositional. There are prepositional representations that correspond to fully understood ideas, and semiprepositional representations which correspond to ideas that are not fully understood. Moreover, humans are capable of mentally representing not only facts, but also their own representations and mental states. These metarepresentations are not factual beliefs, but representational beliefs. The first are based on perception and inference, and the latter are, so to speak, put 'in quotes', in a symbolic form, for future understanding (cf. Sperber 1975). They are believed because they are embedded in factual beliefs. Factual beliefs can be either prepositional and have strong criteria of rationality, or semi-prepositional. Similarly, representational beliefs can have either prepositional content (like scientific assumptions) and be weakly rational, or semi-prepositional content and be held as if they were rational. The latter category includes mysteries and other cultural beliefs, as well as religious beliefs (Sperber 1985a: 58, 1996: 73; 91-92). It is perhaps worth noting that instead of factual and representational beliefs, Sperber 1996 and 1997 talks about intuitive beliefs (common-sense beliefs derivable from perception) and reflective beliefs. But this amendment does not have a significant impact on the approach. Now, the assumption of universalism and rationalism is required in order to talk about universal processes of assumption recovery.

Section 9

Semantic defaults

79

It is also useful for another reason. It demonstrates that there are degrees of strength of the hearer's beliefs tied to the degree of understanding of the speaker's utterance.28 Coming back to intention recognition, its study is problematic for two reasons. First, there are sources of assumptions utilized by the hearer which have not yet acquired an analysis. Second, there are psychological processes that affect intention recovery which may be too situation-specific to allow for a theory of this interaction. There can be no doubt that anger or fear of the speaker or the hearer may affect the utterance interpretation. Such psychological states or processes which cause the state of anger, amazement, fear, or non-emotional states such as incredulity, scorn, admiration, regret, or satisfaction, can help or hinder intention recovery. Generally, this influence can be classified as acceleration or impairment of intention recovery (cf. Jaszczolt 1996b). These forces depend on various aspects of discourse such as the source of assumptions, the place and time of conversation or the personal characteristics of the speaker or hearer. Peter's question in (22) may be recognized as a friendly offer, an attempt to distract the hearer from an important situation, or even to poison her. It is to be doubted, however, that a theory of such an interaction of psychological states with assumption recovery is ever to be attained. This example clearly demonstrates that the process of utterance interpretation cannot be seen as a straightforward recovery of assumptions but has to be given some flexibility instead. This conclusion is not incompatible with relevance theory. According to Wilson and Sperber (1986b), pragmatics is not a module in Fodor's sense,29 it involves the interaction of grammar, logic, and memory. Pragmatic processes are informationally unencapsulated, they have access to contextual information and can allow the hearer to assess the assumptions and intentions. Hence psychological states and processes including intentions can, in theory, be incorporated into a theory of discourse interpretation. In practice, they may not yield to generalizations. Assumptions have the following sources: perception, linguistic decoding, encyclopedic memory, and deductive process. The choice of context depends to some extent on the choice of these assumptions. Now, the hearer, knowing the source of assumptions, may also know possible problems with assumption recovery. For example, if it is linguistic decoding, how does it proceed? Sperber and Wilson (1986b) say that discourse interpretation is incremental, constrained by the central processes, but how does the hearer know what constitutes a unit for parsing? For example, in the case of a prepositional attitude expression, how does the hearer know whether the attitude is held 28

The weaknesses of this approach are discussed in Jaszczolt (1996b: 707 and Recanati 1997) but they are not relevant to the present discussion. 29 See Fodor 1983, 1989.

80

Discourse, beliefs, and intentions

Chapter 2

about the state or event as a whole represented by the embedded sentence (f/j0r-clause) or about an individual represented by a referring term and its having a certain property (i.e. de dicto or de re)? Parsing into smaller or larger constituents will depend on this.30 Bearing all these difficulties in mind, let us try another solution. Utterance interpretation does not consist of the hearer's recovery of the speaker's assumptions but rather is an interactive process. The hearer plays an active role in changing the speaker's assumptions about the world during the conversation and it is so because the speaker assumes that these assumptions are subject to change. This suggestion is compatible with Sperber's idea of metarepresentations (ideas about other people's and one's own representations) and it well explains the lack of clear assumptions to be recovered from Peter's utterance in (22). The suggestion finds support in Levinas (1961) according to whom meaning is created between the interlocutors. It is not the whole truth to say that the aim of the hearer is to recover and enrich the speaker's assumptions, they can also be created. The hearer's interpretation affects the conveyed information to a greater or lesser extent and it is so due to, among other factors, the intervention of psychological states. This explanation seems to be compatible with Sperber's idea of epidemiology of representations, the theory of how cultural beliefs propagate. Sperber and Wilson correctly observe that there is no algorithm for recovering the intended interpretation of an utterance. Hypothesis formation is based on non-demonstrative inference and this is where the theory ends. Nonetheless, it is possible to go a little further in spelling out the principles for this inference. Firstly, it is so because there are defaults of interpretation guaranteed by the intentions in communication, and second, because the speaker's and hearer's psychological states allow for some predictions as to the degree of departure from standard utterance encoding and processing. In addition, the authors of relevance admit that there may be no definite demarcation point between assumptions intended by the speaker and those derived independently by the hearer. The next question to ask is whether the speaker takes this into account and whether we can call these independent inferences utterance meaning. Now, as I argue elsewhere (Jaszczolt 1996a and Chapter 6 of this book), beliefs are conveyed by means of the following vehicles: language, mental images, and non-linguistic actions, all with the help of background knowledge. Intention recovery uses the same sources of information, with the added obvious condition that the linguistic vehicle has to be used. The interaction of these vehicles, as well as their interaction with psychological 30 Cresswell (1985) talks about different sensitivity of the complementizer that to the structure of the embedded sentence. See Chapter 8 below.

Section 9

Semantic defaults

81

states producing acceleration or impairment, constitute a general outline of an answer to (i) and (ii). The prepositional content of the utterance may have to be enriched with information coming from background knowledge in accordance with the view of semantic underspecification. But the influence of psychological states and personal differences is harder to generalize. Concerning the latter, people seem to be prepared to go to different lengths to recover speaker's intentions.31 All in all, the selection of context for utterance interpretation may not be solely determined by the search for relevance as proposed by Sperber and Wilson. It was pointed out to me by an anonymous reviewer of one of my articles (Jaszczolt 1996b) that the idea of psychological states impairing and accelerating intention recognition bears some affinities with Lindsay and Gorayska's (1993), Lindsay et al. (1993) and Gorayska and Lindsay's (1993) proposal of broadening relevance to accommodate non-linguistic plans and goals. The latter say (ibid.: 319) that relevance theory accounts only for the communicative-cognitive goal, consisting in improving the hearer's knowledge of the world. Goals are said to be "[c]ognised motivations or ends" and encompass aims, objectives, wishes, desires, interests, intentions, and wants (jbid.: 303). They are immanent in the producing and recovering of assumptions. Now, instead of broadening relevance, I propose seeing motivations as inherent in intention production and intention recovery. Goals are motivations for action that are a type of psychological states impairing or accelerating intention recovery, so to speak, 'from within'. It is also unclear whether goals mean what the speaker, or the hearer, wants to achieve. Gorayska and Lindsay are committed to the hearer's perspective, the hearer who performs the interpretation. This would be correct if the hearer's sole role was to recover what the speaker intended. But as we have seen, assumptions and intentions do not always go hand in hand: assumptions can be created rather than recovered, and they can be intended to be created. The hearer's perspective is also undermined by the fact that now, in view of the recognition of various vehicles of thought, we have to see relevance as based on the intention to perform a certain action, not necessarily linguistic. And if the goal is not linguistic, the role of the hearer can be far less significant: the hearer may not be required for the fulfilment of the plan. Hence, the result is that Embedding communicative intention (when present) in the over-arching intention to act shifts the perspective from the hearer to the speaker in utterance interpretation. 31

This was recognized by Sperber and Wilson 1986b.

82

Discourse, beliefs, and intentions

Chapter 2

If linguistic actions are only a subclass of actions, then the speaker does not necessarily try to be relevant in Sperber and Wilson's terms. He/she only attempts this when the hearer is required for the fulfilment of the plan. So, Communication is neither totally dominated by the hearer and his/her assumption recovery, nor by the speaker's communicative intention relative to his/her present goals. The agent's assumptions are created either for the addressee or without regard to the observers. After all, there may not be any assumptions to recover. The balance of the interactants' roles is further affected by the presence of emotional and other psychological states which affect assumption creation and the lack of assumption creation on the part of the agent, as well as assumption recovery, assumption creation, and the lack of assumption recovery or creation on the part of the addressee. All in all, the balance of responsibilities for assumptions that relevance theory places on the speaker and the hearer has to be redrawn. Communication is not like dancing where one partner leads and the other follows. Assumptions are not simply created and recovered but rather created between the two in a dynamic manner. Naturally, this broadening of the perspective to non-linguistic actions and goals requires a rethinking of the modularity hypothesis. But with the recent revival of non-modular approaches to language such as parallel distributed processing (Rumelhart and McClelland 1986), a plausible alternative to modularity has arisen. The choice between the two standpoints will not be discussed here. The limitations in processing effort differ from person to person. The question to pose at this point is to what extent the speaker takes into consideration the role of the hearer in his/her life while coding the message. It is a widely held philosophical view that the other person is perceived by the ego as an object in the ego's world, an object with a place and function in this world. Levinas (1961) challenges this idea in his approach to verbal communication. He argues that we coexist with 'the other' but also leave this 'otherness' intact. We do not superimpose our own picture of the person's characteristics onto our reality. This is possible thanks to language. Now, the proposal does not boil down to an ethical or social rule of a preferred linguistic behaviour. Levinas merely says that in order to survive, to coexist with other humans and to acquire knowledge, we have invented self-defence mechanisms such as conversation. Speaking allows us to reveal our world and experience to the hearer without endangering ourselves. Words can be interpreted in many ways, are not always planned in advance, and, first of all, they are not the genuine T but rather its 'revised' presentation to the other person. The idea of infinity is the core of Levinas's view. Infinity means that

Section 9

Semantic defaults

83

thought cannot be assessed through adequation with an object. Thought is creating meaning through events and actions, i.e. through infinity. The rough application of the idea of infinity, contrasted with totality, goes as follows. Totalizers are people satisfied with themselves, seeking power and control in life, whereas infmitizers strive for freedom, creativity, for what is 'other', different from oneself. Varying compositions of these tendencies are also possible. Infinity, striving for something other than oneself, is a precondition for intentionality, the directedness of acts of consciousness towards objects, developed in the phenomenological tradition.32 Hence, infinity is a theory of intentions, also intentions in communication: "... intentionality, where thought remains an adequation with the object, does not define consciousness at its fundamental level. All knowing qua intentionality already presupposes the idea of infinity, which is preeminently non-adequation!' Levinas (1961: 27). Levinas sees the relation between interlocutors as dynamic, involving 'creating one another' in the process. This claim seems to be congenial with the relevance-theoretic denial of mutual knowledge and the claim that the speaker makes things mutually manifest in the process of conversation. Levinas says that conversation always means going beyond the limits of the self and this reaching beyond is infinite, does not stop where the hearer's assumptions or knowledge end. Neither the speaker nor the hearer are responsible for the infinity. Rather, it is the conversation that brings it about. As a result, the speaker may not be so important for recovering meaning: meaning is created instead (cf. ibid.: 76). Thoughts are dynamic, are conveyed, uttered, rather than entertained in advance and exchanged. I called Levinas's view on conversation doubly-dynamic (Jaszczolt 1996b). The hearer consults the available sources of information for intention recovery, including psychological processes of what I called acceleration and impairment, and assigns attributes to the speaker, or 'creates him/her anew'. This 'creation' is not recovery: "Language does not exteriorize a representation preexisting in me: it puts in common a world hitherto mine." Levinas (1961: 174)

and: The phenomenological idea of intentionality is the topic of Chapter 3 below.

84

Discourse, beliefs, and intentions

Chapter 2

"In speaking I do not transmit to the Other what is objective for me: the objective becomes objective only through communication." Levinas (1961: 210). Communicating is co-temporal with changing the state of the speaker's knowledge. So, the speaker may not presume what interpretation the hearer will assign to the utterance but instead rely on this dynamism. This would explain the lack of recoverable assumptions in example (22). This dynamism is best seen in the example of referentially mistaken descriptions. If the speaker utters (23) making salient the individual present who is in fact the woman's brother, the hearer does not normally adjust his/her referring expression to that of the speaker but instead provides additional information as in (24): (23)

Her husband is kind to her.

(24)

Yes, her brother, the man over there, is always kind to her.

(from Linsky, quoted after Donnellan 1966). This is a general tendency of referential truthfulness, spelled out in Grice's maxim of quality and it operates even when the intelligibility has to be sacrificed. Now, why doubly-dynamic! Sperber and Wilson's (1986a) metaphor of conversation as dancing is singly-dynamic: they say that context is created, not given, and it is not established before relevance is assessed. Double dynamism means that interpretation is created in-between the interlocutors, rather unlike dancing which requires a leader and a follower.33 So, utterance interpretation is not assumption recovery and is largely co-temporal with speaking as (a) interpretation is incremental and (b) context is constructed in situ. To conclude, let us retrace Peter's steps in the processing of Mary's response in (22). First, he evokes the sources of assumptions such as perception, linguistic decoding, encyclopedic memory and deductive processes. Next, he responds to Mary's (observable or inferrable) psychological states which may accelerate or impair normal34 intention recovery. He also exhibits psychological states other than intention himself and may take some into consideration. The second and third points are suggested by Levinas's idea of dynamic'self-creation' in conversation. All in all, communication appears less dependent on assumption recovery; the speaker gives the hearer relative freedom of interpretation. 33 According to Levinas, the objective becomes objective through communication rather than through imposing on the hearer what is objective for the speaker. 34 I.e. ceteris paribus.

Section 10

Semantic defaults

85

This, not uncommon, intuition of the active role of the hearer can be pronounced with support from other philosophical traditions. Recently, Fredsted (1998) backed it with arguments from Kierkegaard according to whom indirect communication, i.e. communication through implicatures and ambiguous discourse, aims at leaving the 'creation of meaning' to the recipient (cf. ibid.: 540). The interpretation of the utterance is deliberately left open, to be created by the interlocutor. This proposal supports my doubly-dynamic approach based on Levinas's view of conversation. 2.10. Conclusions The analysis in this chapter adhered to the rejection of semantic ambiguity but one further step was taken to demonstrate that the level of underdetermined semantics can be disposed of as well. This level is unnecessary due to the influence that intentions exert on the semantic representation. The main message is that the ambiguity-underdetermination dilemma is only the result of a wrongly posed problem. Degrees of intentions help classify readings of referentially ambiguous constructions as to their accessibility to the hearer. The referential intention, embedded in the communicative one, is present in any of the full spectrum of degrees from its absence to unproblematic referring. People arrive at an interpretation by default reasoning. Which interpretation is it? If there is a systematic answer, it comes from the juxtaposition of the two ideas discussed here: [1] types and degrees of intentions on one hand, and [2] default reasoning on the other. As I have demonstrated, putting the two together, one is led to conclude that the referential interpretation of utterances involving a referring expression (including a definite description) is the more salient one. This interpretation is accompanied by the strongest intentions and the effort required on the part of the hearer to recognize these intentions is the smallest. When the intentions are weaker, the object of the utterance is less exposed to interpretation, as in the case of referential mistakes. In the case of the lack of the referential intention, the effort involved is the greatest: the intentionality of mental states makes the hearer look for the referential interpretation and the latter is not found. It has also been signalled that the degrees of intentions may help decide on the default readings of other expressions where the primary intention is not applicable. Finally, it was stressed that utterance interpretation is a creative process to a degree much higher than is standardly admitted in pragmatics. This claim was discussed with the example of relevance theory.

This page intentionally left blank

CHAPTER 3

Intentionality and Propositional Attitudes "... there are many descriptions of happenings which are directly dependent on our possessing the form of description of intentional actions. It is easy not to notice this, because it is perfectly possible for some of these descriptions to be of what is done unintentionally. For example 'offending someone'; one can do this unintentionally, but there would be no such thing if it were never the description of an intentional action." Anscombe (1957: 83). Contents 3.1. 3.2. 3.3. 3.4. 3.5. 3.6. 3.7. 3.8.

'Directedness' of Acts of Consciousness: The Phenomenological Tradition Intentional Relation and Defaults Intentionality and Voluntariness Intentions and Intentionality De Re, De Dicto, and the Horizons Speech Acts: Double Level of Intentionality? Are Intentions in the Head? Intentionality and Ambiguity: Concluding Remarks

DISCOURSE, BELIEFS, AND INTENTIONS Current Research in the Semantics/Pragmatics Interface, Vol. 2 K.M. Jaszczolt © 1999 Elsevier Science Ltd. All rights reserved 87

88 92 94 96 99 104 Ill 119

88

Discourse, beliefs, and intentions

Chapter 3

3.1. 'Directedness' of Acts of Consciousness: The Phenomenological Tradition Default Semantics and intentions on which it is founded are derived from the intentionality of mental states. In this chapter I present the connection between intentions and intentionality. In Section 2.9 I looked at the process of utterance interpretation from the perspective of the 'double dynamism' of Levinas's view. I concluded that assumptions are created, as well as recovered, by the hearer and both processes are compatible with intention recovery. Now, if intentions accommodate creation of the assumptions by the hearer and the construction of meaning between the speaker and the hearer, acts of consciousness must also depend on this dynamism. It is so because they are intentional and intentionality of acts of consciousness, as Levinas says (1961: 27, quoted in 2.9 above), presupposes this dynamism. Bearing this in mind, I shall now discuss this intentionality as represented in the phenomenological tradition of the nineteenth and early twentieth century, with particular reference to belief ascription. The Latin word intendere means to aim in a particular direction, to direct the thoughts to something (Simpson 1959), similar to drawing and aiming a bow at a target (Harland 1993: 71). Traditionally, forms of 'being'are divided into esse naturale, natural objects, and esse intentionale, such as thoughts and mental images (W Lyons 1995: 1). Intentionality means here the property of mental acts of having content, being about something. Mental acts include such attitudes as belief, desire, want, need, expectation. Taken from Aristotle through Avicenna and the medieval doctrines of knowledge and experience, intentionality was later developed in nineteenth-century phenomenology. At present, intentionality is also known in some circles under a guise of representationality (cf. Woodfield 1994: 1705). Aboutness' or 'direction' mean 'orientation' of mental acts towards objects. Intentional acts provide the meaning of expressions. This is best seen in the example of indexicals that exhibit acts of demonstration as part of their meaning. For example, the meaning of the pronoun 'he' is not entirely present in the word but requires an additional act of the assignment of meaning to this open symbol (cf. Osborn 1934: 95). In contemporary terminology, this meaning is called procedural and is distinguished from conceptual meaning (Wilson and Sperber 1993). The idea of intentionality was brought to the fore by Brentano (1874,1874a) and his students, most notably Husserl (1900-1901,1900-1901a), Meinong and Twardowski. Phenomenology is the study of forms of conscious experience, including the ways things ('phenomena') are presented in consciousness. It is also sometimes regarded as a philosophical method used by Husserl or Heidegger but it is not the sense in which the word is used here. Brentano proposed that objects of consciousness are all in the mind. When John sees a red square, intentionality does not amount to a relation

Section 1

Intentionality and prepositional attitudes

89

between John and an object (here: a red square), but rather to the act of seeing whose feature is spelled out by the adverbs 'redly' and 'squarely' as in (1) and (2): (1)

John sees redly.

(2)

John sees squarely.

(from Smith and Woodruff Smith 1995: 14). This adverbial theory of intentionality did not account for the fact that mental acts are about things in the world and that we can talk about someone else's mental acts.1 It was subsequently replaced with a relational account, according to which intentionality is a relation between the experiencer (thinking subject) and an object.2 Alternatively, it was also defined as a relation between an act and its object. Husserl went further and rejected immanent, mental objects, replacing them with real things as correlates of perception. The idea of intentionality presupposes a distinction between subjective ideas in our mind and objective ideas, attributed to Bolzano (1837, 1837a). Thinking means having an idea of something, or having subjective phenomena present in the mind. These phenomena are imperfect counterparts of the objective 'meaning'. Objective ideas are constituents of propositions. The latter exist independently of the mind and can be entertained, brought to the mind in a judgement: "Existence belongs only to propositions thought, and also to those held to be true, i.e. to judgements, but not to the propositions in themselves which are the material a thinking being grasps in his thoughts and judgements." Bolzano (1837a: 167). Bolzano distinguishes objective entities which are ideal, present 'outside the mind' and their counterparts present in the mental processes. In Husserl's (1900-1901) terminology, there are objective meanings or thoughts, and acts of meaning, or acts of thought that bring these thoughts to consciousness. Consciousness is understood in the Brentanian tradition, as a totality of experience. Meaning is given to linguistic expressions by language users in mental acts (see Smith 1987: 211). This view of meaning was later modified by Husserl and replaced by the idea that meaning is contained in the noema, the objective content of consciousness (Husserl 1913: 304; 1913a: 368). 'Cf.: "The embarrassment for adverbial theories is (...) that they are not well placed to account for the objective aspects of intentional mental acts and states." Bell (1990: 134). 2 This step is attributed to Kazimierz Twardowski, extended by Alexius Meinong. See Smith and Woodruff Smith (1995: 15) for references.

90

Discourse, beliefs, and intentions

Chapter 3

The noema is an intentional object, the content of an act, it functions as the meaning or sense. In virtue of this content an act achieves an intentional relation to an object. In other words, an act is directed toward an object through a mediator called noema (cf. Woodruff Smith and Mclntyre 1982: 87). This direction or intention is always definite, aimed at a specific object determined by the noemata.3 Now, linguistic meaning, i.e. meaning expressed in language, is a sub-type of noematic meaning, i.e. meaning attached to relevant mental acts. This is how language becomes one of the carriers of meaning, and sense becomes one of the kinds of noemata. The idea of carriers of meaning is developed further in Chapter 6. The property of directedness of thoughts and other mental acts was a natural consequence of Bolzano's distinction between thoughts and acts of thoughts. It was developed by Brentano (1874a: 324) as the idea that "[a]ll our thinking has a thing as an object". This 'thing' can be the subjective or the objective idea and this dilemma was well discussed in the phenomenological literature of the period, with the natural development towards the acceptance of real instead of mental objects.4 In the post-Brentanian tradition, intentionality has been regarded either as a feature of conscious states (Searle 1983,1992), or as really present in the head, in the computations in the brain, as in Fodor's realism (Fodor 1975,1981; see W. Lyons 1995). According to the latter stance, there are no mental acts independent of the brain: there are only brain states with informational content (representations), and intentionality is the feature of the brain that allows for this informational content to be present. Approaches to intentionality are ample in the literature and they will not be reviewed here. I shall accept the dominant realist stance that originated in Brentano's intentionality and derive the theory of intentions in communication from this account.5 I shall also accept Brentano's confinement of intentionality to consciousness. This confinement is by no means universal. For example, W. Lyons (1995) puts forward a theory of how intentionality developed in evolution and can be classified on three levels: (i) the primitive intentionality of the brain where information is not processed by means of representations but rather as an 'analogue imprint', similar to a footprint in sand; (ii) 'proto-interpretive' intentionality on the level of sensory experience, where experiences are 'about' something, they have content. This is the prelinguistic 3

But see Section 3.2 on the content-object distinction applied to indexical expressions. Indexical expressions can lead to different objects. 4 See Smith and Woodruff Smith (1995: 21-22) for the status of objects. 5 For example, Millikan naturalizes intentionality in saying that intentional acts are comparable to the activities of organs and limbs. Dretske naturalizes it in saying that humans are physical information-processing systems. See W. Lyons 1995.

Section 1

Intentionality and prepositional attitudes

91

level of intentionality, for example of associating sounds with situations or objects at the 'pre-speaking' stage of the end of infancy. Finally, there is (iii) linguistic intentionality and intentionality of prepositional attitudes. Prepositional attitudes are intentional in the (a) linguistic sense, i.e. with respect to the properties of that-clauses in attitude reports such as the lack of substitutivity, as well as the (b) non-linguistic sense. The latter comprises the perspective, angle of looking, or guise under which things reveal themselves when we perceive them and form concepts of them, as well as a disposition, the totality of the characteristics attributed to the holder of the attitude (see W Lyons 1995: 217-219). Now, this expansion of intentionality to the three levels is rather contentious and unnecessary. For example, on the third level of intentionality and assigning dispositions to the believer, it seems plausible to conflate linguistic and non-linguistic intentionality, or, in other words, it seems plausible not to distinguish them. Language, as a standard vehicle of propositional attitude, is normally present in processing the output of perception and awareness of the 'angle'of looking at things. It is also present in attributing beliefs to people; representations are frequently formed in the linguistic means and externalized, or brought to consciousness, to awareness (as phenomenologists would say) by the linguistic vehicle. They are normally further conveyed between individuals also by means of language. Treating beliefs as dispositions to certain behavioural manifestations is also a result of this broadening of intentionality beyond consciousness. This result is, however, rather harmful and confining. Beliefs need not be dispositions, they need not produce gestural, linguistic, or other physical results.6 Deriving intentionality of beliefs from the two lower levels of intentionality distorts the picture and necessitates this confinement of the account to those attitudes that result in something physical. The extension of intentionality is therefore either harmful or otherwise at least not needed. Not all mental states are intentional: feeling dizzy, for example, is not. Also, not all intentional mental states are conscious: there can be unconscious beliefs or desires. These will not be discussed here. The category under discussion is conscious intentional states, sometimes called experiences (Woodruff Smith 1989), such as perception, recollection, thought, or conscious desire. For example, when I look at Botticelli's La Primavera, I 6

Lyons tries to dissociate this standpoint from behaviourism by pointing out the difference between dispositions and their resultant observable effects. But retaining the dispositional account does not seem to be defensible. Cf: "A belief is a disposition. It is the whole parcel of perceptual input, of inner states or processes, and their output in terms of behavioural, linguistic, facial, gestural, and any other manifestations. But dispositions are not their manifestations - or even their behavioural manifestations in the light of input - for to say that would be to fall back into behaviourism." W Lyons (1995: 231).

92

Discourse, beliefs, and intentions

Chapter 3

perceive not only patches and shapes of different colour, but I have a visual experience of a painting of an orange grove with a goddess-like figure in the centre, a beautiful woman in a flowery dress to her right, all surrounded by nymphs and other figures. So, perception is about something, represents something, or is intentional. This is the sense of intentionality that is, firstly, the least contentious, and, secondly, useful for the investigation of the semantics/pragmatics interface.

3.2. Intentional Relation and Defaults Prepositional attitudes that are intentional experiences have an object and a content. The object is that which the attitude is about. When I think of a painting of Botticelli's La Primavera, the object is the painting of beautiful nymph-like figures in an orange grove. The content of an experience is not the object; it is rather the way the object presents itself to the experiencer, it is a 'mode of presentation' (henceforth: MoP). For example, when I think that Botticelli was a great painter, the content of thought is not Alessandro di Mariano di Vanni Filipepi himself but rather, the author of the beautiful paintings I saw in Florence. The distinction between object and content is very important. For instance, if the belief is about Santa Claus, it has no object but it clearly has content. Or, a belief about the Morning Star has the same object as a belief about the Evening Star while their contents clearly differ due to the difference in the mode in which the planet Venus is presented. Or, if I think I am tired and you think you are tired, the content of our beliefs is the same while the object differs (cf. Woodruff Smith 1989: 8-9). A similar distinction between the object and content is forced by cases where the object of experience changed unbeknownst to the experiencer: the content of experience remains the same while the object has changed, like, for example, in the case when I see a French one franc coin and, after someone has replaced it with an identical coin while I looked away for a couple of seconds, I again see a French one franc coin. Experiences with content and no object and pairs of experiences with different content and the same object have been the focus of phenomenological debates since Husserl. They are important in the semantics of intensional contexts because they illustrate how meaning is possible without reference to a real object and how meaning can differ although the referent does not in sentences with coreferential expressions which are otherwise identical. The case of same content and different objects (cf. the coin example) is more contentious and pertains to the discussions on the status of indexical expressions in semantics. This issue will not be discussed here. Now, MoPs will be tied here to referring expressions. In other words, a referring expression has one object revealed

Section 2

Intentionality and prepositional attitudes

93

to the speaker under different MoPs, just as the object is revealed to the believer under different MoPs. This part of the phenomenological MoP is of interest for the semantic ambiguity debates of prepositional attitude sentences. Consequently, Woodruff Smith (1989:10) distinguishes intentional character of experience, i.e. its having content, and intentional relation, i.e. being about a particular object. Intentionality means having intentional character, and, if successful, also intentional relation to an object. The expression 'if successful' carries an enormous weight for the discussion of semantic ambiguities. It supports the idea that being about something is the optimal type of experiencing, and, when experiencing involves language (language of thought), it is also the ultimate type of using language. Now, by transitivity, it is also the optimal type of use of language in communication. It follows from this that utterances with referential intention are the optimal type of communicating and hence if there is a choice of interpretations of an utterance, the referential interpretation, pertaining to the presence of the strongest referential intention, is the most optimal interpretation, or the default. Content is very important for intentionality. The content of my thinking that the author of La Primavera was a great painter is the thought that the author of La Primavera was a great painter, where the definite description is represented as a concept pertaining to this description. This content makes it possible for the thought to be about an object: content picks out the object (see Woodruff Smith 1989: 20). Again, the route through content to the object as an ultimate, successful intentionality signals the default status of 'being about an object', referring to a real entity exhibited by a thought or an act of language use.7 It is so in spite of the fact that phenomenology abstracts from the real world and a mental act has an object, in a sense, even if the object does not really exist. But the intricacies of phenomenological reduction will not concern us here.8

7 Woodruff Smith (1989: 13) points out that in addition to the traditional Husserlian content theory of intentionality there are other theories, for example an object theory of intentionality such as Russell's theory of acquaintance where belief is a 'prepositional attitude' sensu stricto: it is a relation between a believer and a series of objects. 8 See Simons (1995: 116). For the exegesis of Husserl's idea of the real, common-sense world see Smith 1995.

94

Discourse, beliefs, and intentions

Chapter 3

3.3. Intentionality and Voluntariness The widely accepted property of intentionality of beliefs raises a question as to whether the believer has the belief voluntarily, whether he/she is aware of having it, and whether believing that p means accepting it in further reasoning. Are there involuntary beliefs? Or are beliefs always voluntary and amount to an acceptance of certain propositions? Does the speaker's utterance give evidence to the speaker's belief or acceptance of the corresponding proposition? In this section I review Cohen's (1992) arguments for the involuntary character of beliefs and for the need to postulate two separate mental acts that underlie sentences, namely beliefs and acceptances. The decision on this issue is essential for the interpretation of belief expressions and reports on beliefs. Beliefs have been traditionally considered to be either a voluntary act of judgement or an involuntary feeling. Beliefs are also taken to be dispositions to feel something because they can be assented to or dissented from without the concurrent feeling being present. Now, distinguishing beliefs from acceptances is quite straightforward at first glance but going beyond the obvious folk feelings on the subject may prove to be a more complicated exercise than the gains would justify. Beliefs are normally reasons for a mental act of acceptance but not everything believed ends up being accepted by the believer (cf. Cohen 1992: 19). So, if beliefs are intentional, must they be voluntary or should we employ the concept of acceptance for the latter purpose instead? Unfortunately, we cannot employ it. The reasons for accepting a proposition may have nothing in common with believing it. These reasons can be ethical, social, or practical in some other sense (cf. ibid.: 20). But contrasting the two terms is useful. It demonstrates that while accepting a proposition means accepting its logical consequences (deductive closure of propositions about acceptances), believing a proposition has no such necessary property.9

9

Cf.: "Acceptance is objectively closed under deducibility if and only if necessarily, for anyp!, p2... and pn and for any q that is deducible from the conjunction of p}, p2 ... and pn, anyone who accepts pi, P2 ... and pm accepts q. Acceptance is subjectively closed under deducibility if and only if necessarily for any p\, p2 ... and pn and for any q, anyone who accepts both the conjunction of p\, p2 and pn and the deducibility of q from that conjunction, also accepts q." Cohen (1992: 28-29). Acceptances are subjectively closed: "Of course, where q is deducible from p and you accept/?, you ought always to accept q. But whether you actually do so or not depends on whether you also accept that q is deducible from p" (ibid.: 29).

Section 3

Intentionality and prepositional attitudes

95

Beliefs may not, after all, be directly expressed in assertions: "... it is hardly deniable that, in quite a wide range of familiar indicative-mood speech-acts, saying that p does not imply belief that p. Rather it implies there a mental act of acceptance that p? Cohen (1992: 72). So, utterances give evidence to acceptance rather than to a belief. This is Cohen's view. Now, it is also widely accepted that an assertion is an expression of belief and this is the view we have adhered to so far, seeing linguistic meanings as sub-species of the genus called noema. The broadening of the perspective to include acceptances is important. For example, a sentence such as (3) can express either an involuntary disposition (Cohen's 'belief') or a voluntary acceptance of a judgement or of principles for further reasoning: (3)

People should always tell the truth.

And, "... we should expect that the semantics of the natural-language sentence will be adequately understood only when the underlying mental duality is properly grasped." Cohen (1992: 77). This duality is important when one views beliefs, expressions of belief or reports on beliefs in the context of, respectively, other beliefs, expressions or reports. While beliefs can be inconsistent in that they are not subject to deductive closure, acceptances cannot. Knowledge can rely on either beliefs or acceptances and thus does not exhibit deductive closure either: logical consequences of beliefs may not be believed and consequently are not present in belief-based knowledge (cf. ibid.: 99). Now, it seems that the overall consistency of beliefs is a matter of degree. Cohen makes the following two claims: "It is certainly impossible for the same person at the same time to believe firmly both that p and that not-/?." 10 (ibid.: 85). and But it can happen when beliefs are weak. See ibid.: 132.

96

Discourse, beliefs, and intentions

Chapter 3

"Since beliefs are not deductively closed, they cannot be intrinsically subject to any holistic requirement for overall consistency (...). And at the heart of this willingness to tolerate inconsistencies of belief, in certain cases, is the implicit awareness (ibid.: 105). that belief is involuntary." Since beliefs are beyond the believer's choice, they need not undergo rationality. But the direct contradiction/? & —>p is excluded, it normally does not occur. Hence, there is some degree of rationality and the boundary of this rationality seems to lie somewhere in the infinite set of less and less obvious logical consequences of beliefs. In other words, beliefs are rational to a degree and when one holds a belief, one holds some of its logical consequences. Furthermore, considering that "In ordinary cases of perception and memory, a person's possession of a belief that p is a presumptive reason for him to accept that /?." (ibid.: 130), the distinction between belief and acceptance seems to boil down to a common-sensical boundary between 'standard' beliefs that lead to acceptance and are closed under deduction, and 'nonstandard' beliefs that do not exhibit these properties. Since, in addition, the overt contradiction p & -*p is excluded, we shall continue discussing language expressions as expressions of belief, bearing in mind the departures from this connection.

3.4. Intentions and Intentional ity The next step is to demonstrate that intentions in communication cannot be separated from the intentionality of cognitive processes. Intentionality will be shown to be best suited to explain the significance of the degrees of strength of intentions11 (pace W. Lyons (1995: 1) who challenges its connection with intentions in action or communication saying, rather short-sightedly, that intentionality has little in common with people's intentions to do something). To repeat what was briefly stated in Chapter 2, the relation between the referential intention and intentionality proceeds as follows. The referential 11

Cf.: "The theory of intentionality can be compared with the theory of linguistic sense and reference in the tradition of Frege and Husserl. As every meaningful expression in a language has a sense and if successful a referent, so every intentional experience has a content and if successful an object to which it 'refers'." Woodruff Smith (1989: 11).

Section 4

Intentionality and prepositional attitudes

97

intention is directed at an object (cf. Bach 1987a: 52), just as intentionality of the underlying mental act is directed at an object. Intentionality of, let us say, a belief is its essential property: a belief is always formed as intentional in this sense, that is it necessarily has the feature of 'aboutness'. Now, the 'aboutness' may not stand up to scrutiny when the linguistic expression of belief is confronted with the situational context. In other words, although the speaker's belief 'intends' to be intentional, it does not always succeed in being so. Similarly, utterance interpretation 'sets off' as if the utterance conveyed a strongly intentional state, a belief directed at a particular object. Only further evidence to the contrary changes this default to an interpretation according to which the referential intention is not present. Or, it is present, but the speaker committed a referential mistake by using an incorrect name or description. This is also a departure from the default (see Jaszczolt 1997a, 1998d). About the use of proper names, Husserl (1900-1901a: 496) remarks the following:

"Using the proper name significantly, we must present to ourselves the subject named (...), and as endowed with some definite content or other. However impalpably, defectively, vaguely, indefinitely we may think of [it], presentational content cannot be wholly lacking."

There is some degree of inevitable vagueness in the use of a proper name because the way in which real things present themselves to the mind is always incomplete. But this incompleteness can be further completed, the object can be further determined to make the content of this presentation more adequate. This statement taken from Investigation IV is typical of Husserl's approach to referring. It suggests that the object of intentionality can be revealed to a greater or lesser extent in thought and communication. In other words, the referential intention is present to a greater or lesser degree. The possibilities of further determination of the object of experience culminate in the absolute correspondence of the act of meaning and the object itself which, once again, demonstrates the privileged, unmarked, default status of this full correspondence and the strongest referential intention (see ibid.: 497). Now, in Investigation VI, Husserl talks about intuitions and calls them intentions that require fulfilment (cf. ibid.: 699-701), as in the case of a melody that is perceived with intuitions slowly fulfilled as the music progresses. And,

98

Discourse, beliefs, and intentions

Chapter 3

"An intentional act may intend its object emptily or with evidence. If the evidence presents the object intuitively, precisely as it is described or meant, then the act is 'fulfilled'. Paradigmatic of the latter sorts of acts are those in which an object is seen as such and such." Mohanty (1983: 4). Meaning-intentions and acts of meaning-fulfilment belong to so-called 'objectifying acts', or meaning-giving acts. An act is objectifying if it can play a part in the acquisition of knowledge (cf. Mohanty 1964: 46). These acts can be expressed when the meaning-giving intention is fulfilled in 'intuition', realized as perception or imagination.12 Objectifying acts can be placed in the wider perspective of intentional acts as follows. An intentional act in the widest sense pertains to an object given (later called noema, see Husserl 1913, 1913a) in the act of intending (later called noesis).13 In the more specific sense, intentional acts comprise (i) meaning-giving, objectifying acts which amount to thought (pignitive acts) or perception or imagination (both called intuitive acts) as well as (ii) not-objectifying acts, such as will, desire, or hope, that do not establish meaning (cf. Husserl 1929, 1929a, and the diagram in Mohanty 1964: 47). So, meaning is composed of meaning-intention and meaning-fulfilment. And, when this intention to represent has been fulfilled, notably in perception, the adaequatio rei et intellectus is achieved. The intention is directed at a real object and is fulfilled, the degree of referentiality is the highest. The object is 'given', referred to in the strongest, unambiguous way (cf. Husserl 1900-1901a: 762). All thoughts are carried by these acts.

12 Intuition (perception or imagination) can be either of real objects (sensory) or of universals (categorial). It can also be adequate or inadequate. Perception and imagination are opposed to 'mere thinking', 'signifying' meaning (cf. Osborn 1934: 107). 13 Cf.: "The noema of the judging process, i.e., of the concrete experience of the judgment, is the 'judged content of the judgment as such'. (...) The judged content of the judgment is not to be confused with the matter judged about." Husserl (1913a: 272). And, "The uttered word, the actually spoken locution, taken as a sensuous, specifically an acoustic, phenomenon, is something that we distinguish from the word itself or the declarative sentence itself, or the sentence-sequence itself that makes up a more extensive locution." Husserl (1929a: 19-20).

Section 5

Intentionality and prepositional attitudes

99

3.5. De Re, De Dicto, and the Horizons Meaning-giving acts have been tried in the literature as a criterion for distinguishing de re and de dicto acts, and in particular de re and de dicto beliefs. The distinction will be properly attended to in Chapter 4. At present I shall look at its relation to intentionality in phenomenology. Mohanty (1983: 5-6) entertains the following possibility: "... an act is de dicto if and only if it is unfulfilled, empty, merely signitive, and an act is de re if and only if it is fulfilled. Fulfilled acts, as it were, put us in touch with the real thing itself." In the next move Mohanty entertains the possibility that an act is de re when it is verified, confirmed, or 'fulfilled', that is when the experiencer 'knows that/?'. This equating of de re with knowing proves to be unsatisfactory, though. 'Knowing that p' is a factive attitude, an attitude that presupposes that it is true that/?. But there is a whole gamut of non-factive attitudes such as beliefs that can be either de re or de dicto. In other words, I can believe something about somebody de re without knowing this thing about the person. Hence, there are two binary distinctions involved: factive/nonfactive, and de re/de dicto, and only one theoretical tool in Husserl to deal with them, namely fulfilment of intention. Mohanty correctly concludes that the 'empty/fulfilled' distinction does not help with the de re/de dicto one. However, he forgets that knowing someone may also be fulfilled, for example the intention to refer to Bernard J. Ortcutt can be fulfilled in perception. So, although the fulfilled act is not a de re belief, fulfilment in perception normally guarantees beliefs de re, whereas fulfilment of intention in imagination renders beliefs de re but about an individual that is difficult to identify intersubjectively. Expressions of belief that are based on such acts can contain referential mistakes. De re can be 'about someone else' (or, no real living person for that matter). Mohanty resorts to 'perceptual sense' instead (ibid.: 7): "... an intention whose sense is articulated in a definite description does individuate its object for the speaker, provided the sense also contains a perceptual component such as 'The person over there...'. We thus come back to the phenomenological idea of 'perceptual sense'." Now, the perceptual sense advocated by Mohanty will not do either. De re belief can be based on perception or imagination (in the latter case including 'de re about someone else', opaque de re), but it need not be so based. The believer may be able to individuate the referent thanks to some reliable

100

Discourse, beliefs, and intentions

Chapter 3

knowledge of it. So, instead, it seems necessary to amend HusserFs classification of objectifying acts and add the referential intention which is derived from the original knowledge-belief (factive-nonfactive) distinction. It is derived by way of specifying that referential intention is fulfilled when the referent is known, or mutually manifested (in Sperber and Wilson's 1986a terminology) to the speaker and the hearer. Moreover, the fact that Husserl eliminates mental objects from his model and regards the real world as a correlate of our factual experience constitutes an important contribution to the semantics of referring expressions. Perception is directed at objects rather than sensations. We see things, we see something real. And this is what gives de re beliefs their privileged, unmarked status. The referential intention is a part of the informative intention and is itself satisfied either through perception or through other mental acts that sum up to some knowledge. This totality of mental acts that are involved and that constitute background knowledge for subsequent acts, is attended to by Husserl in his proposal of various 'horizons', possible states of affairs, distinguished with respect to objects, acts, people, and times. Background knowledge of the holder of a belief helps explain the contradiction in belief expressions such as in (4) and (5) where the definite description in (4) and the proper name in (5) are coreferential and both (4) and (5) are uttered by the same speaker: (4)

The man in the brown hat is a spy.

(5)

Ortcutt is not a spy, he is an honest citizen!

In the light of (4), an utterance of (5) may be taken to mean that someone else, e.g. Mr Smith, is a spy and the speaker attributed an incorrect name to him. Similarly, in the light of (5), an utterance of (4) may be taken to be about someone else, for example due to a mistake caused by poor visibility of the person in the interlocutors' visual field. The hearer is aided here by the background information, by contextual assumptions. Juxtaposing the old and the new information yields a potential clash that can only be avoided by adhering to the underspecification of content. This underspecification, or, rather, possibilities of further specifications, Husserl sums up in the idea of the horizon developed in his later works (1939, 1950). He says that our experience of objects is always perspectival and is surrounded by a horizon or other. Further information about the object comes to light incrementally, as experience progresses, and the gained information has to be compatible with the one already added to the context. Every mental act has a horizon of other possible acts directed at the object of the original act. These acts may provide further information, further determination of the object. In addition to the act-horizon, there is also a horizon of the object which is

Section 5

Intentionality and prepositional attitudes

101

associated with the original act, a horizon of further properties and characteristics of the object (cf. Woodruff Smith and Mclntyre 1982: 227, 240). The totality of the speaker's experience, including the totality of beliefs, is present in every particular act of experience of the speaker's (see Jaszczolt 1996a: 127). It follows, for example, that a belief is de re not because it is founded on clear, unproblematic perception but because in addition it fits in with the network of beliefs. Beliefs are consistent, we can draw on our past knowledge, beliefs, perceptions and other experiences. This act of perception is a part of reality; the latter constitutes a context called a'co-present margin' (Mitgegenwdrtige, Husserl 1913). What people see is always fragmented, incomplete, and surrounded by the unperceived or imperceptible things, states, events and processes. As Husserl (1913a: 102) says, "What is actually perceived, and what is more or less clearly co-present and determinate (to some extent at least), is partly pervaded, partly girt about with a dimly apprehended depth or fringe of indeterminate reality". The temporality of this world is infinite, the world has its temporal horizon.14 This horizon aids a particular experience: knowledge gained from other acts and possibilities of other acts is accessible. Generally, the horizon of an object is thus the sum of possible experiences of an object. The act's horizon is the sum of expected experiences, those which are founded on the experiencer's background beliefs about similar objects. The range of possible horizons is itself very large and diversified because they are theoretical constructs that are to account for the penumbra of meanings with reference to the experience, other experiences, the experiencer, other experiencers, the object of experience, other objects, and time. In Experience and Judgment, Husserl (1939: 27-29; 1939a: 32-33) says that every experience of a particular object has a horizon of other experiences of the same object. This is an internal horizon of experience. It comprises the possibilities and capacities of the experiencer. This horizon is not a 'real' horizon of any mental act. Instead, it is a device of talking about clustering of future experiences and about their being affected by the previous ones. The internal horizon of experience is a rather weak construct because it merges the real context and the possibilities of context: things co-given and things anticipated (cf. Jaszczolt 1996a: 129). This merger is particularly unwelcome when the discussed possibilities of the experiencer pertain to the external horizon of the object. In other words, the problem arises when we talk about the possibilities of the person's experiencing different objects. These possibilities seem to be infinite, we can encounter new things every moment and have 14

In HusserPs phenomenology, the existing time is the immanent time of the flow of consciousness. The objects of perception appear in this subjective time (see Husserl 1928 and 1928a; 1950 and 1950a). It seems, however, that the horizon is infinite no matter what measure and ontology of time we accept.

102

Discourse, beliefs, and intentions

Chapter 3

mental states about them. Husserl recognizes this infinity as the "total horizon of the world in its infinity" (Husserl 1939a: 36 and 1939: 33), approached in our experience of real things throughout the lifetime. In addition, every experience is placed within a class of experiences aiming at other objects. In other words, other objects may affect the original experience. This "aiming-beyond to other objects of which we are aware at the same time" (1939a: 33 and 1939: 28) is an "external horizon of objects cogiven". These objects are only co-present rather than intentional but the believer may focus his/her attention differently and experience these things which, in the former belief, were only co-present. It is an over-arching horizon which enables the internal limits of experience called the internal horizon. It is also called a 'second-level' horizon as it is built upon the internal one. The example with proper names used in Section 3.4 is relevant here. Every experience has its content and its possible further determinations of meaning and reference, which also constitute a horizon. Every experience contains the possibility of obtaining new information about the object of experience. The amount of this information depends on the experiencer and the direction in which this revealing process progresses is determined by the reality, the object itself (cf. Husserl 1939: 27-28; 1939a: 32). Every experience has its internal horizon in that the experiencer may either stop or continue the experience in the subsequent, more precise experiences of the same thing. In his later work, Husserl develops this horizon in discussing potentialities, possibilities of perception, the ways perception would have happened had we turned our eyes in a different direction. In Cartesian Meditations, Husserl (1950: 81-82; 1950a: 44-45) talks about the forthcoming parts of the particular experience, an intentional horizon of reference (cf. also Bell 1990:190). This is a horizon of anticipating, for example, seeing the whole object after having caught a glimpse of its part. Both the object's internal and external horizons contribute to the system of experiences that determine the particular belief. However, the distinction between them seems of limited importance to Husserl. He talks about possible experiences on the one hand, and the horizon of recollections and past experience on the other, without specifying their exact role in forming the context. It is easy to see how the possibilities such as a different point from which we perceive an object or a different set of circumstances can be used in semantics: after all, this is nothing else but an early, primitive version of the possible-worlds approach to meaning. So, while the real experiences constitute the context, the possible ones delimit the semantics of the expressed content of an experience in question: they provide the default. This interpretation is also in agreement with Husserl's early view of Logical Investigations:

Section 5

Intentionality and prepositional attitudes

103

"... certain objects or states of affairs of whose reality someone has actual knowledge indicate to him the reality of certain other objects or states of affairs, in the sense that his belief in the reality of the one is experienced (though not at all evidently) as motivating a belief or surmise in the reality of the other? Husserl (1900-1901a: 270). The idea of real and possible context is also congruent with Husserl's (1939 and 1939a, §16) discussion of the association of beliefs where one eventuality recalls another. This storage of information is what in terms of modern computational linguistics would be called the knowledge base. So, the meaning is unravelled through the multiplicity of different acts and every belief is determined by other beliefs held by the person. The act of perception, such as the act of seeing the person called Ortcutt, presupposes some other mental acts which constitute its background. The background is constitutive of the meaning: it is what makes seeing someone seeing him as Ortcutt, as Smith or as someone or other. Applied to beliefs, it is the totality of the experience that decides whether the belief is de re, de re about someone else (a case of a referential mistake15), or de dicto. The distinction on the level of sentences (belief reports) can be drawn by extension. The idea is best demonstrated when applied in building the discourse representation structures of DRT in Chapter 7. To account for the context-dependence of departures from default interpretations of belief reports and expressions of belief, one has to delimit the horizon of possible experiences to the expected experiences of other objects that may affect the particular belief on which the speaker wishes to report. In other words, we need a link between a belief about Ortcutt and the background information concerning Ortcutt as well as other potential candidates for the referent. This is what I called the external horizon of the act, a horizon not distinguished by Husserl but very useful as a device for delimiting reference (see Jaszczolt 1996a: 129). This device is, in fact, used in DRT as the choice of placement of discourse referents in particular representation structures (see Chapter 7). In general, the similarities between the horizons on the one hand and possible worlds semantics on the other are self-evident. There are possibilities of going beyond the act of meaning, further experiences, possible states of affairs that, so to speak, provide 'more meaning', further determinations. These similarities also signal a very useful application of the horizons to the semantics of prepositional attitude reports. 15

To be discussed at length in Chapter 4.

104

Discourse, beliefs, and intentions

Chapter 3

All in all, what it means is that the thing, the real object, is always in the centre of belief or other experience. It can be thought of as known, transparent, to a smaller or greater extent, and this boundary of knowledge or 'acquaintance' with the object differs with time, just as the Earth's horizon moves forward when we walk. But a real thing is always presented to consciousness as one thing, referential mistakes do not occur on the level of mental experiences. Intentionality may simply be weak enough to render referential mistakes in utterances but this misfiring of intentions is in fact the fringe of information that is hidden beyond the horizon. It is not a different mental object without a real correlate: misfired intentionality simply means weaker intentionality and the first degree of departure from the default interpretation of an utterance discussed in Chapter 2. To sum up, meaning-giving acts, supplemented by a gamut of horizons, seem to point to the availability of a wide range of information to the holder of the belief because this belief belongs to a network of other experiences - real and potential, of various objects. Meaning-giving acts are fulfilled in thought, perception, or imagination and the horizons (or: potentialities) allow for further experiences of this and other objects that sum up to further development of the experiencer's knowledge, should this be necessary for utterance interpretation. As a result, the belief normally exhibits a strong intentionality and a clear identification of its object, so to speak, by default. This clarity is passed on to linguistic expressions. The way in which it is passed on is discussed in the following section.

3.6. Speech Acts: Double Level of Intentionality? The intentionality of mental states has been commonly regarded as more basic than, and prior to, the intentionality of linguistic expressions.16 It has been argued that the non-linguistic intentionality operates prior to language and hence leads to more basic meaning. Hence, the meaning of speech acts can be analysed in terms of intentional states such as belief, desire, or intention. It would seem that since the intention to represent is prior to the intention to communicate, there is some ground for this conjecture, which was in fact entertained both by Husserl and later by Searle, bearing also some affinities with Grice's account of speaker's meaning. Allegedly, the mind imposes intentionality on linguistic expressions that are not intrinsically intentional. Searle (1983: 165-166) distinguished two types of meaning intentions: an intention to represent, and an intention to 16 I shall utilize now more contemporary terminology and talk about conscious, intentional mental states (psychological states). The term 'mental act' will be reserved for stressing the act of aiming at an object.

Section 6

Intentionality and propositional attitudes

105

communicate. Speaker's meaning is inherent in the intention to represent and the latter reflects the intentionality, or aboutness, of its mental correlate, a mental state. For Searle (1983), the parallel between the referential intention and intentionality is sufficient for constructing speech act theory based on the idea of the inheritance of the conditions of satisfaction of a speech act from the intentionality of the underlying mental state. Utterances have derived intentionality, whereas beliefs have intrinsic intentionality (Searle 1991: 84). This 'derivation' proceeds as follows: "The mind imposes Intentionality on entities that are not intrinsically Intentional by intentionally conferring the conditions of satisfaction of the expressed psychological state upon the external physical entity. The double level of Intentionality in the speech act can be described by saying that by intentionally uttering something with a certain set of conditions of satisfaction, those that are specified by the essential condition for that speech act, I have made the utterance Intentional, and thus necessarily expressed the corresponding psychological state. I couldn't make a statement without expressing a belief or make a promise without expressing an intention because the essential condition on the speech act has as conditions of satisfaction the same conditions of satisfaction as the expressed Intentional state. So I impose Intentionality on my utterances by intentionally conferring on them certain conditions of satisfaction which are the conditions of satisfaction of certain psychological states." Searle (1983: 27-28, my emphasis) In other words, Searle's account conforms to the so-called 'double intentionality', the utterance is intentional because conditions of satisfaction of a speech act make it express a psychological state and the psychological state is itself intentional. So, intentionality of linguistic expressions is inherited from the intentionality of corresponding psychological states by means of the imposition of conditions of satisfaction (see Searle 1983: 27-28; Harnish 1990: 188). According to Searle, mental states (specifically: intentional states of consciousness) such as wishes or hopes, determine the conditions of satisfaction for speech acts. These conditions help establish the meaning of a speech act. For example, a mental state of desire can imprint conditions of satisfaction on the command uttered by the holder of this desire. Inheritance, however, or imposition for that matter, is an enigmatic magic wand here. The problem is that of transfer: how does the mind transfer 'aboutness' from mental acts to sentences? How does the mind transfer conditions of satisfaction? Harnish (1990: 189) entertains several possibilities and all of them fall seriously short of providing a satisfactory answer.

106

Discourse, beliefs, and intentions

Chapter 3

One cannot intentionally transfer the conditions just by intending to do it because it should be possible to intend and fail. If so, what would it mean to fail to transfer the conditions from mind to language? This view of transfer leads to an obvious trivialization:"... can one intentionally confer conditions of satisfaction on anything, a dog, on the birth of a dog?" (ibid). Arriving at a parallel does not suffice. Perhaps one should not envisage transfer or conferment as a movement in time or space, or even a move from the mental domain to the physical, because, firstly, no Cartesian dualism of mind and body is assumed here and, secondly, Searle talks about the same conditions (see emphasis in the quotation above). Moreover, the double level is present in the speech act rather than between the mental and the linguistic. The obvious conclusion to draw is that 'transfer' or 'conferment' are just unfortunate terms signifying identity, sharing, rather than inheritance or derivation. This possibility is entertained in the account of language as subsumed under the mental, as one of the vehicles of thought, sketched below and elaborated in detail in Chapter 6. The act of uttering a statement and the content of the statement are the two entities that were confused in philosophical texts for a long time, with a strong emphasis placed on the latter aspect (see Smith 1990: 31-32). The aftermath of this confusion is still present. It is only in Bolzano's, Brentano's and Husserl's writings that an act of saying and a concept became distinguished on psychological grounds. As Husserl says in his early work (1900-1901), meaning is created in intentional acts. So, in our terminology, the prepositional form depends on the physical and psychological acts of utterance production and comprehension. Intentions and other psychological states are, so to speak, read off"the world, as Reinach says (Smith 1990: 54) and they provide vital information for the construction of prepositional form. The world' includes language use. In other words, properties of language are not derived from properties of mental states. Instead, linguistic meaning is social and primary, established in language use. As a result of Searle's proposal, speaker's meaning is derived from the intentionality of mental states. On the one hand, this proposal is common-sensically correct. One has to have a desire first, and subsequently one can express it. An intentional link between the two can be easily surmised. But on the other hand, as Apel (1991: 41) suggests, the meaning of a desire, conviction or any other state is intersubjective and hence the conditions of satisfaction of the mental state may to some extent depend on the meaning of the associated linguistic expression. This is so because of the a priori of language: the representation of the world through language, through the propositions with a structure, is unavoidable. This thesis permeates Wittgenstein's Tractatus Logico-Philosophicus and is also the foundation for Tarski's semantic theory of truth (see e.g. Platts 1979). This reversed dependence, i.e. the dependence of the meaning of mental states on linguistic

Section 6

Intentionality and prepositional attitudes

107

meaning, is also manifested in the conventions of language use and in early HusserFs social theory of concepts. Apel proposes that the meaning of speech acts (the prepositional content and the illocutionary force) can be dependent on the conventional use of sentences and phrases in a language. This conventionalization can be either linguistic, e.g. an utterance of (6) corresponds to a promise, or it can be social, e.g. an utterance of (7) counts as a resignation: (6)

I hereby promise never to use swear words.

(7)

I hereby resign from the membership of the Committee of Global Happiness.

Now, just as speech acts depend on linguistic conventions, so do mental states. Normally, hearers have no difficulty with associating a heard speech act with the speaker's mental state. This association may not yield to a one-to-one correspondence, as, for example, a wish may correspond both to a request and a command,17 but no real ambiguity ensues here. This reversed dependence can be easily explained when we think of the role language plays in bringing about the mental state. Intentional mental states are regarded as psychological equivalents of utterances, they express the 'directedness', the relation toward the world that Searle calls the 'direction of fit': the situation adjusts to the speech act or the speech act follows the situation, as in the case of a promise and a complaint respectively. Utterances are composed of prepositional content and illocutionary force, just as intentional mental states are composed of representative content and the psychological MoP (cf. Baumgartner and Klawitter 1990: 213). Now, instead of postulating double intentionality, we can plausibly infer from the similarity of the two types of acts to the hypothesis that language plays a role in intentional mental states. Desires, for example, can be held in language, they can be represented to the holder of the desire in a linguistic form. They need not necessarily be linguistic but they are commonly so. The same holds for other mental states, and hence, what concerns us most, for propositional attitudes and specifically beliefs. It is so because intentionality means directedness at the world. Intentions are, so to speak,'read off the world', they concern situations. They do not take a trip from the mental to the linguistic, they are mental and linguistic in the first place. This view has strong advantages over Searle's approach. First, there is no convincing evidence in support of the duality of level of intentions. Second, by Occam's razor, unitary intentionality is preferred over a multiplication 17

This is an important strand of Apel's argument by which he demonstrates that mental states are too coarsely-grained to provide adequate conditions of satisfaction for speech acts.

108

Discourse, beliefs, and intentions

Chapter 3

of levels. And finally, language is one of the vehicles of thought and one of the carriers of meaning of mental states and hence should not be separated from them. The latter argument is attended to in detail in the discussion of vehicles of thought in Chapter 6. Subsuming language under the mental is suggested by the similarity of intentional states and speech acts. For example, Moore's paradox normally extends to them. Normally, though not always, one would not utter (8) successfully: (8)

Please open the door but I do not wish the door to be open.

Sentence (8) seems anomalous because the speech act of requesting comes together with the mental act of wishing. Searle (1983: 9) makes an even stronger claim in saying that "[t]he performance of the speech act is eo ipso an expression of the corresponding Intentional state". The word 'corresponding' may require a qualification here. Mental states are not sufficiently finely-grained to allow for a one-to-one correspondence. States do not individuate speech acts and there are occasions when it is not obvious what exactly the underlying state is, as, for example, in the case of congratulating a person on something. This strong connection between the mental and the linguistic signals that the linguistic may in fact be a component of the mental, a carrier of meaning-intentions, a vehicle of thought.18 All in all, if one accepted Searle's double intentionality and his arbitrary (though conventional) constitutive rules, one would have to, firstly, explain the transition of intentions, and secondly, prove the reality of this transfer or a methodological advantage in postulating it. Neither has been achieved, the postulate is largely ad hoc, it is epistemologically dubious and methodologically costly. In addition to flouting Occam's razor, Searle's postulate does not comply with the POL principle either. Searle created a superfluous level of explanation between reality and the act of speaking. Neither is his 'intentionality as a whole' consistent with POL. It takes a long route in making coarse ontological commitments to the mental and separately to the linguistic, which Husserl successfully avoided by locating meaning in the mind and at the same time in the world. Real objects are intended, and this objectivity gives mental states their intersubjective status (see also 3.7 below). The conclusion seems to be that, instead of merely comparing intentional states and linguistic expressions and postulating arbitrary conditions, one should specify what role language plays in intentional states. It is the role of a vehicle of thought as spelled out by Husserl and subsequently by Frege and Dummett. 18

For a comprehensive summary of similarities between speech acts and intentional states see diagrams in Baumgartner and KJawitter (1990: 213-214).

Section 6

Intentionality and prepositional attitudes

109

Compelling evidence for subsuming language under the mental comes from demonstratives that are used to refer to perceived objects. The use of demonstratives, according to Searle, is only possible because the meaning of the demonstrative combines with the intentional content of the perceptual act (cf. de Mulder 1994). Instead of the widely accepted view of direct reference according to which the referent is part of the proposition expressed (as advocated by Kaplan and Perry), Searle proposes a combination of the linguistic and the perceptual. Together they fulfil the meaning-intention, the intention to represent. Since demonstratives are often, so to speak, 'completed' by perceptual experience, the proposition expressed contains the intentional content of the perceptual experience (cf. de Mulder 1994: 21-22). No direct reference, no presence of entities in the semantic constructs has to be evoked. Meaning, as the Fregean sense or the linguistic sub-type of Husserlian noema, is given by the complex of the lexical meaning and the context. So, instead of direct reference, we have a cognitive correlate of demonstratives - a noema, sense, or discourse referent of DRT. This cognitive construct is possible when we subsume the linguistic under the mental as one of the ways in which thought is represented in the mind and, in the case of speech acts, also conveyed to the hearers. It has to be noted, however, that intentionality is not necessarily linguistic: intentional states can be displayed by babies (e.g. wanting milk) or dogs (e.g. wanting to be let out, examples from Searle 1983: 5). In fact, humans who are in possession of language can also display non-linguistic intentional mental states, for example a driver's sudden terror caused by a child running in front of his/her fast moving car. These states seem to rely on a pictorial, rather than a linguistic, vehicle of thought.19 To sum up, the intentionality of a speech act depends on the intentionality of the corresponding mental state because, for example, normally one has to have a wish before one utters a request. But also, conversely, the intentionality of the mental state depends on the meaning intention (intentionality) of the linguistic expression in that language makes it an intersubjectively recognizable mental state. The dependency here is a dependence on a language system rather than on the physical act of speech. In effect, it is a dependence on the linguistic a priori, on the fact that language reflects the states of affairs (as in truth-conditional semantics) and there is nothing beyond that which can be expressed in language: there is no way round the representation of the world through language. In addition, the meaning of a mental act is intersubjective because it reflects the meaning of illocutionary force indicators such as 'I promise that...' present in a language

The types of vehicles of thought are further discussed in Chapter 6.

110

Discourse, beliefs, and intentions

Chapter 3

system. This dependence makes the speaker's promise a promise to the hearer, and the desire a desire. The hearer reconstructs a mental state of the speaker as if there was such a link from the type of a speech act to the type of mental state. We can also infer this dependence from Searle's principle of expressibility (Searle 1969: 19-21) which says that anything that can be meant can be said. The dependence of the linguistic and the mental is clearly mutual and it seems to be so because they are not two separate species: language should be subsumed under carriers of meaning in mental states, or, in Husserlian terms, as a component of an objectifying act. It is a component because objectifying acts are either linguistic themselves, or operate with horizons. We perceive a shape as a tail of a cat whose body is hidden from sight behind a bush, we perceive three sides of a figure as a cube although the other sides are hidden. There are other acts that function as background knowledge and enable this synthesis of information. Perception is not just a separate act, it is a component of a knowledge dossier and this dossier standardly, or at least more often than not, involves language as its means of representing and storing information. Horizons, and in particular the external horizon of the act, are also present in Searle's account of collective intentions, for example an orchestra's intention to perform a piece of music. This collective intentionality is not reducible to intentions of individual players or to the mutual knowledge of their respective intentions and yet it is present in the individuals' heads as an intention to do something by means of the individual's contributing a component of the activity (cf. Searle 1990b: 407). Collective activities require some idea of the other person, a "preintentional sense of 'the other'" (ibid.: 413). This idea seems to be a version of Husserl's horizon, potentialities of actions performed by other persons all aimed at one object. So, collective intentionality, similar to background knowledge, is an example of integration, synthesis of information, or meaning holism. In his later work, Husserl (1913, 1913a) stresses that intentionality, or the orienting at the object, happens automatically to the experiencer. Searle (1980) talks about intentionality as a biological phenomenon which also seems to indicate its automatic character. If directedness is automatic, then it is difficult to divide it into stages or components, the mental and the linguistic. Searle's main argument for this separation seems to be the difference between speech acts and mental states, or expressions and beliefs. He also stresses problems with the connection between mental states and behaviour: behaviour (including linguistic) is not a sure guide to mental states (Searle 1992: 22). If intentionality is biological, this connection should be diaphanous. On the other hand, since for Searle intentionality is biological, he stresses its import and classifies intentionality of speech acts as derived from this biological phenomenon, trying to overcome the fact that the relation between mental states and behaviour is contingent.

Section 7

Intentionality and prepositional attitudes

111

I extended here Apel's argument that mental states are affected by language because they are intersubjective to the proposal that language is one of the media in which beliefs are represented in people's minds. Beliefs and other mental states need not be linguistic but they very often are. And whatever language they are represented in, this language is either a natural language or a language of thought that, in principle, shares the principles of propositional structure laid out in the language of first-order logic. This generalization is all that matters for the present discussion of how beliefs are represented; the intricacies of the discussion of the language of thought are of no importance here.20 Now, since the intended object is by default a real object, the referential intention of the speech act achieves independent support from the intentionality of the mental state. By the same move of subscribing to directedness at the world, semantic ambiguities between referential and nonreferential interpretations of definite descriptions and de re and de dicto interpretations of belief reports are merely cases of stronger and weaker intending and undergo an explanation in terms of default intentionality. And, by the same move again, they require an explanation in terms of default intentions that contribute to the semantics of these constructions and render ambiguities (or underspecifications for that matter) superfluous. Since mental states can be presented in language to their holder, their intentionality ends up as the strongest intention that can be assigned to the corresponding speech act, that is as a communicative intention that subsumes an informative intention that subsumes a referential intention, if applicable. Q.e.d.

3.7. Are Intentions in the Head? Intentionality appears to be a convenient term for a connector between mental states and real situations, including states, events, processes and their component parts. For Husserl, intentionality is non-natural, it concerns consciousness rather than physical world and real situations: it is directed at objects but it is directed 'indirectly', through a mediator of sense or noema. For Heidegger, it is part of the world. Husserl overcomes the dualism of the mental and the physical by saying that the human being includes something from nature, something from humanity, and something from consciousness. This allows him a non-committal approach. In other words, meaning is in the mind but it is also in the world (cf. Woodruff Smith 1994).21 20

The literature on the topic is ample. See e.g. Fodor 1975; Gumperz and Levinson (eds) 1996. The construct of the human being is heavily theory-driven, though. On the status of 'ego' see Kotekowski 1975.

21

112

Discourse, beliefs, and intentions

Chapter 3

Intentionality cannot be called a process, it is an instantaneous 'firing at', 'targeting' objects. Hence, it is not a process of passing through sense to the referent, it is not arriving at the meaning through the way the objects present themselves to the mind. Great care must be taken not to reify intentionality in this way so as not to multiply beings beyond necessity. The use of intentionality in semantics can be best assessed in the example of options of interpretation of referring expressions. The property of 'aboutness' allows for degrees of this aboutness, as well as for the degrees of sensitivity to the real situations with all their inhabitants, their properties and relations between them. Weak aboutness concerns a situation (or: eventuality as a whole); strong aboutness picks out constituents of this eventuality, and in particular it reaches real people, animals and things. For example, (9) can be believed as a, so to speak, whole proposition or as a property predicated of a known or otherwise identifiable painter: (9)

Botticelli painted Flora.

Similarly, (10) can be uttered, and for that matter interpreted, as a prepositional attitude report about Tom's belief in a past activity of the known painter or about a past event concerning some person or other (called 'Botticelli') who painted some character or other (called 'Flora'): (10)

Tom believes that Botticelli painted Flora.

This sensitivity to the content of the embedded clause is better seen in sentence (11) which states on one reading that Johnny believes a proposition remembered from a science-fiction story, without knowing what 'red giants' and 'white dwarfs' are. Alternatively, on the other reading, he believes, on the basis of extensive reading of astronomy books, that one particular, familiar type of star transforms into another. (11)

Johnny believes that red giants become white dwarfs.

A formal semantic account of this sensitivity to the constituents of the proposition is developed in Chapter 8. What is of interest for the present analysis is the dependence of these interpretative options on the degrees of 'aboutness' and the strength of intentions, including the referential intention. Now, the question to ask is, how does the mental correlate of experience, that is sense or noema, end up being about these real eventualities. And, indeed, can it succeed in being so? There are two avenues to pursue here. Firstly, it seems to succeed because, in intensional contexts such as (11), there are ways, modes of presentation (MoPs) in which the object (here: Botticelli) is known to the holder of the belief, and these MoPs rely on intentions

Section 7

Intentionality and propositional attitudes

113

and intentionality. In other words, the referential intention individuates the particular painter, known, mutually available for the speaker's and the hearer's knowledge dossier. The lack of this intention corresponds to the speaker's entertaining a representation or other which makes him/her capable of using the proper name. So, MoPs, or their 'fineness of detail', aid the referring process (see Schiffer 1992; Jaszczolt 1998a).22 On the other hand, MoPs need not fit the person or object named, they can correspond to a different individual, be mistaken, or they may not fit any individual at all. So, perhaps MoPs and intentions never leave the mental after all? Perhaps they render referential, de re, and otherwise strongly transparent readings only because of the dossier, background knowledge, or horizons with which they interact? In fact, Husserl, and before him Kant, are commonly regarded as precursors of cognitivism in the study of meaning. Phenomenology, with its insistence on contents of mental acts (or: noemata), succeeds in relegating real objects and eventualities to a secondary role; reality is cognized through mental correlata. This is essentially the idea of phenomenological reduction which was a direct predecessor of methodological solipsism associated with arguments laid down by Hilary Putnam and Jerry Fodor (cf. Dreyfus 1982). Methodological solipsism is an assumption that no mental state presupposes an existence of any individual other than the holder of that mental state - or even, other than the 'disembodied mind' of this holder (cf. Putnam 1975b: 220). According to methodological solipsism, attitudes are 'in the head' and are individuated by objects and propositions internal to the organism, by the so-called narrow content. Searle's intentionality can be reconstructed as solipsistic if one follows a strong interpretation of his double level of intentionality, i.e. the separation of mental states and speech acts. As was discussed above, this strong interpretation is unfounded, though. Firstly, in the long passage quoted in the preceding section, Searle says that the same conditions of satisfaction pertain to both domains. Secondly, I suggested that language 'belongs' to the mental by being one of the vehicles of thought. Language establishes the intersubjectivity which would otherwise be lacking. In other words, if not for language, solipsism would ensue in the manner spelled out by Apel (1991: 40): "My language-independent laying-down and testing of the conditions of satisfaction of my fact-intention would in a certain sense be carried out in terms of the methodical (sic) solipsism of the pre-linguistically oriented evidence-phenomenology of Edmund Husserl." 22

Schiffer's theory of types of MoPs is discussed and further developed in Chapter 4.

114

Discourse, beliefs, and intentions

Chapter 3

There would be no mediation of language, no intersubjective component of mental states. Moreover, the solipsistic stand does not explain attributing beliefs to people, neither does it help with explaining cases of the lack of bi-uniqueness (one-to-one correspondence) between the real referent and the mental equivalent, as in the case of the ancients' belief in the existence of the Morning Star and the Evening Star while in reality they both correspond to the planet Venus, or, in the mismatch going in the opposite direction, calling a substance 'water' while in reality there is H2O on Earth and XYZ on Twin-Earth as in Putnam's example discussed in the following paragraphs. In other words, it is the object that causes the mental state that brings about the differences through a mental mediator, MoP or sense. Hence, methodological solipsism is called into question. Now, before proceeding with the argument, two essential terms require a brief introduction. I have referred so far to intensional contexts as sentences in which substitutivity of coreferential expresions does not hold. Further examples and properties of intensional contexts will not become relevant until Chapter 4 but the term intension, as opposed to extension, is necessary. Generally, the extension of an expression is what the expression refers to. The intension is the manner in which the referent is presented. Hence, intension in its wide sense has close affinities to MoPs. The extension of a sentence is often equated with its truth value. In extensional semantics, the meaning of a sentence consists in its truth conditions, i.e. the conditions which the eventuality has to satisfy for the sentence to be true. For problematic contexts, like the ones of prepositional attitudes, we also need intensions as they play a part in meaning. Standardly, in Montague semantics, intensions are defined as functions from contexts to extensions in a context. So, the intension of a common noun is a function from possible worlds to sets of individuals, while the proposition, or the intension of a sentence, is a function from possible worlds to truth values. In other words, we take a sentence,'check' if it is true or false in different possible states of affairs and arrive at a classification of these states of affairs with respect to this sentence. In Montague semantics and its current derivatives, intensions constitute the core of the meaning of expressions (cf. van der Does 1994a: 1699-1700; 1994b: 1702). Searle (1983: 203-204) adheres to the slogan that meanings are in the head because intensions, which are in the head, determine extensions. This is his response to Putnam's (1975b) objection. Putnam presented a well known thought experiment to prove that it is possible for a person to experience the same mental state on two occasions and yet a different object is involved in each. Knowing the meaning is not just having the corresponding concept or intension, it is also knowing that this concept is a correlate of a particular object. Now, mental states may not determine concepts (intensions), or

Section 7

Intentionality and prepositional attitudes

115

concepts may not determine objects (extensions). Putnam's famous example of water having the chemical composition of XYZ on Twin-Earth demonstrates that an Earthling and a Twin-Earthling who are not aware of this chemical composition have the same thoughts concerning the relevant substances, even though their thoughts are directed at two different objects. Hence, mental states do not determine objects. In a different thought experiment, I may not be able to tell the difference between an elm and a beech tree, have the same concepts for both, and yet hold that 'elm' means a set of elm trees, while 'beech' means a set of beech trees. So, "[c]ut the pie any way you like, 'meanings' just ain't in the head!" Putnam (1975b: 227). We know that there are elms and beeches because we can consult the experts, that is we know it through the 'division of linguistic labour', and we know the meaning of 'tiger' when we know what the stereotypical tiger is like (see also Putnam 1970; for criticism see e.g. Crane 1991 and Beattie 1993). Now, Searle advocates instead a Fregean internalism about meaning. A speaker and a hearer have mental states with some intentional contents and this enables them to understand linguistic reference (cf. Searle 1983:198). Searle opposes the predominant idea of direct reference of indexical expressions with its claim of the redundancy of Fregean sense in the case of indexicals, de re beliefs and other attitudes, as well as referential use of definite descriptions. In other words, he opposes the claim that no mental intentional states are needed for their explanation. Searle's (1983: 201) argument for meanings in the head is that intensions are more powerful than Putnam admits. Intension or Fregean sense is abstract and can be perfectly or imperfectly grasped by the speaker and hearer. Hence, not knowing the intension is not an argument against the route from intentions of mental states to extensions (objects): intentions, so to speak, override the lack of a concept (intension) and provide meaning anyway. In fact, intentional mental states do distinguish between H2O and XYZ in that intentional states are directed at a particular object causing the experience. Water is whatever is identical with the stuff that causes the particular experience, hence either XYZ or H2O, not both. By the same argument, intentionality is sufficient to individuate beliefs. Beliefs can be about a real object or not, but they always involve a MoP, sense, or noema, i.e. a mediator, content of the mental state.23 Now, if language, a public, intersubjective system can be part of the mental, or, as Apel's discussion indicates, be a pre-condition for some mental states, then methodological solipsism is avoided. There is an objective reality at the objective target of the 'aboutness'. 23

That is why for Searle (1983: 215) all beliefs are de dicto, and some of them are additionally de re. We shall not adopt this terminology.

116

Discourse, beliefs, and intentions

Chapter 3

Arguments against methodological solipsism are in fact ample. The role of the 'broad content', the real world, is acknowledged more and more often in the literature. Fodor (1994) takes the intentionality of mental states further and reconciles it with the computational character of mental processes. The content of mental states is intentional and does, somehow or other, depend on situations and objects in the real world. This relation is of particular importance to Fodor. He uses it as a tool against the idea that meaning arises out of relations between thoughts, i.e. against semantic holism. Instead, he defends atomism which, in semantics, amounts to the idea that small units of meaning can be conceived of and acquired in separation from other units. The equivalent in psychology is the idea that experiences are composed of such small, indivisible mental entities. What is important here is that the real world matters for meaning. In the past Fodor (e.g. 1980) maintained that the relation of thoughts to the world is not of any importance to psychology: truth and reference are not psychological categories. This is known as the idea of narrow content. He subsequently abandoned narrow content and admitted that truth and reference do matter for psychology and for mental states: the content of mental states is broad. Mental processes are computational and yet they have access to the real world. In other words, the content as information and thinking as computation are reconcilable, although the relation between them is contingent on the laws and structure of the world. The way the world is organized determines this relation. The main argument in support of this view goes as follows. Mapping from referential content to computation is essentially unproblematic, the relation is bi-unique. There are occasional departures from this common-sensical state of affairs in that, for example, the ancients believed that two different occurrences of the planet Venus were instead two separate planets (the repeated ad nauseam Morning Star and Evening Star). On the other hand, in theory, two different chemical substances, H2O and XYZ, can be both referred to as 'water' prior to the sufficient advances in chemistry. So, Fodor also recognizes the role of intentionality in semantics, denied in Fodor 1975. Sometimes the broad content and computations are not bi-uniquely mapped. But this anomaly is very rare, it is controlled by the way the world is organized and, in any case, it only seems to cause theoretical rather than practical problems because at a given time /, it is not known that 'both planets' are Venus or that 'water' is in fact two natural kinds. The world is very well organized, computations and content 'stick together', so to speak, by nature. Putnam's 'water' example and Frege's Morning Star - Evening Star are, en masse, prohibited. There isn't any XYZ, there isn't any half-Venus because prepositional attitudes involve MoPs of objects, neither is there any 'elm-beech' species (cf. Jaszczolt 1995a). In a sense, Fodor fights with straw men. Intentionality pertains to mental states and, a fortiori, also to language. It does reach real objects. Meaning pertains to linguistic expressions that have

Section 7

Intentionality and propositional attitudes

117

this intentionality and hence to deny it is to put the cart before the horse. It is to deny the objective component of mental states in order to preserve the computational character of the processes in the mind. Fodor's readmission of broad content can be seen as a commendable move to put the horse back before the cart again. So, are intentions in the head? And are meanings there as well? In spite of Searle's convincing internalism, the answer is probably 'yes and no'. They are between the mind and the world: they are on the path of intentionality, through the noema, ending with the real objects when those are available. And, if the mind (or mental phenomena) are a feature of the brain, then they are in the brain and henceforth in the world. This chapter should thus bear a sub-title: "How to Putnam a Searle and how to retaliate". This conclusion has serious consequences for the role of intentions in semantic theory. As was suggested in Chapter 2, sentences that have more than one interpretation and that were traditionally described as semantically ambiguous and more recently as semantically underdetermined, are better viewed as allowing for default interpretations that are achieved through the proposed approach of Default Semantics. This semantic theory relies on the contribution of intentions to sentence meaning. Now, intentions are neither in the world nor in the head because mental states have an over-arching property of intentionality directed at the real objects. This directedness gives them a high degree of objectivity and as a result it imposes this degree of objectivity on sentences expressing these states, as well as on sentences reporting on them. Directedness at objects, realized as a referential intention, also renders default interpretations of such sentences. Intending to communicate, to inform, and to do so about something are crucial components of utterance meaning, but also, and what is more important, of sentence meaning. This is how semantics is intention-based, while intentions are in-between 'the head' and 'the world'. A word of caution is due at this juncture. Intention-based semantics is not a new concept, in contemporary pragmatics it originated in Grice's work on meaning and the present proposal is not the only development of Grice's approach. Schiffer (1982), for example, derives intention-based semantics as follows. First of all, semantic and psychological facts are all entirely dependent on physical facts and the semantic facts are likely to be reducible to psychological properties, in line with Grice's (1957) reductionism of meaning to intentions (cf. Schiffer 1982: 119-120). Schiffer's intuitions, however, receive only limited support in his account. Intention-based semantics should rely on psychological properties founded on belief, desire and intention. In order to demonstrate that these are not in turn determined by semantic properties, one has to have a successful theory of belief and other attitudes. Such theory has not yet been proposed (and hence had not been

118

Discourse, beliefs, and intentions

Chapter 3

proposed by 1982),24 as can be seen from the discussion of the state-of-the-art in prepositional attitude research in Chapter 4. So, Schiffer's task is confined to arguing that the putative correct theory of belief has to be compatible with intention-based semantics. One of the main problems with such a semantics seems to be the following. A linguistic act is successful when both the speaker and the hearer agree to some large extent about word meaning and syntactic structures, and as a result about sentence meaning. In other words, their knowledge of the above is similar, they know the language. Intention-based semantics has to account for this knowledge and, a fortiori, for the principle of forming and understanding new sentences which is founded on (albeit not exhausted by) the principle of compositionality. In other words again, the speaker believes that the hearer believes that the word 'dog' stands for or means in English any member of the class of dogs. The speaker also believes that the hearer would interpret an utterance 'dog' a such. This reductionism has sometimes been misunderstood in the literature. For example, Grandy (1982: 331) identifies it with the postulate of 'behavioral dispositions' that identify the mental state without the mediation of language. He rejects it and says that in fact the content of belief can frequently be fixed only through the mediation of language. For example, if one believes that Cambridge is a beautiful town without ever having seen the city, this belief is likely to be held as a result of a heard or read linguistic expression rather than any other form of acquaintance. This behavioural interpretation of reductionism that leads Grandy to its critique is unwarranted, though. A belief can be arrived at, held and expressed through various carriers or vehicles, of which language is only one type. The fact that some beliefs are language-based need not yet mean the lack of reducibility of the linguistic to the psychological, nor the reducibility of the psychological to the linguistic. Grandy (ibid.) says that "[i]n many cases the state that is being portrayed in a belief report can only be fully identified by reference to language". This is plainly false. Even beliefs de dicto, beliefs about unknown individuals, about whoever or whatever happens to have a certain property, or beliefs with a content that is not fully understood as in (12), are not fully analysable as expressions of belief by reference to language alone: (12)

Red giants become white dwarfs.

Sentence (12) requires some knowledge of astronomy to be at all understood. Reports on such beliefs are potentially ambiguous, have at least the de re and de dicto interpretations, and this choice of interpretation is resolved 24

For a dicussion of Schiffer's recent work see Section 4.2.2.

Sections

Intent tonality and prepositional attitudes

119

thanks to the reliance on intentions in communication. More precisely, it is resolved because the speaker and the hearer both know how intentions work and the meaning of an expression of belief is established thanks to this reliance on the knowledge of this process. For example, in (12), on one scenario the default referential intention is absent since the speaker does not know what red giants are, neither does he or she know what the property of becoming white dwarfs means. And this overriding of the default presence of the referential intention is already a partial reductionism of the semantic to the psychological. It is in fact better to use a different term than 'reductionism' here. In Chapter 2,1 used the term 'contribution of intentions to the semantics' which is more non-committal while allowing for this partial reductionism. So, the fact that some beliefs are linguistic does not preclude that the linguistic in them, i.e. their vehicle, has an intention-based semantics. Although this conclusion may at first glance look like infinite regress, it is only a move of stepping down one level to the properties of that carrier of belief, the vehicle of thought. Expressions of beliefs de re are even more straightforward. If one believes something de re, the reduction of the meaning to intentions is automatic: one believes of something (or someone) that it (or he or she) has certain properties. The linguistic is partly reduced to the physical so to speak automatically, through the representation of the object. But the problem with discussions around reductionism is even deeper than that. In fact, whether this reducibility is total or partial is beside the point. It was only an insignificant theoretical exercise of the past to try to reduce meaning to psychology. The true question is how they interact and how intentions contribute to meaning and to semantic theory. There are linguistic as well as non-linguistic thoughts, beliefs and desires but this is not an interesting fact. So, Schiffer's (1982) reductionism is only one side of the enterprise of intention-based semantics. It is the side which is tangential to the present investigation. It develops only one of many possible questions that stem out of the Gricean programme developed in his seminal article 'Meaning' (1957). To sum up, semantics can be intention-based without being totally reducible to psychological states or, what is more important, without this question having to be addressed.

3.8. Intentionality and Ambiguity: Concluding Remarks The arguments presented above seem to seriously undermine Searle's distinction between intrinsic intentionality and as-if intentionality. Searle argues that if we abandon this distinction, everything becomes mental: water flowing downhill seems to display intentionality by 'trying' to get to the bottom of the hill and 'seeking' the easiest route to do so (example after Searle

120

Discourse, beliefs, and intentions

Chapter 3

1990a: 274). This is a non sequitur. Speech acts are intentional because they use language which is also a standard component of mental states and enables their intentionality. Water flowing in a stream is not intentional in this sense. Neither water as a natural kind nor a stream are pre-conditions or constitutive parts of mental states, unless we want to push biology extremely far and look at the chemical composition of the human body and at intentionality as a biological property of the brain. But then we can invent an example of a rubber ball thrown up in the air and rest safe. Besides, mental states are conscious or potentially conscious (cf. Searle 1992, 1994; Fodor and Lepore 1994) and so are speech acts, while a stream and a rubber ball can be safely classified as devoid of this remarkable property - at least in the possible world in which I am sitting at my desk in Cambridge and writing these words. So, the conclusions of this chapter are as follows. (i) Our beliefs and other experience are 'about' something, are 'directed at' objects, and this 'aboutness' is mediated by a mental content before reaching the real object. 'Before' is not to be understood in a temporal sense. If the object is there to be targeted, intentionality is the strongest. (ii) Intentionality plays an important role in the semantics of natural language. It underlies (and enables discourse of) intentions in communication. As a result, it provides interpretations of utterances, ranging from the default to various cases where this default cannot be reached. Finally, a tentative conclusion that will be further tested in Chapter 6: (iii) Intentionality of mental states subsumes intentionality of speech acts because language constitutes one of the possible vehicles of thought.

CHAPTER 4

The Default De Re Principle "If two things don't fit, but you believe both of them, thinking that somewhere, hidden, there must be a third thing that connects them, that's credulity." Eco (1988a: 49).1

4.1. Prepositional Attitudes: A Close-Up 4.1.1. Prepositional Attitudes, Attitude Sentences, and Attitude Reports 4.1.2. The 'Ortcutt Puzzle' and Its Interpretations 4.1.3. Factives and Non-Factives 4.1.4. The Object of Belief 4.1.5. Defaults for Belief Reports or Defaults for Referring Expressions? 4.1.6. Varieties of De Re and De Dicto 4.1.7. Substitutivity? 4.2. Whose Meaning? On Sense and Mode of Presentation 4.2.1. In Defence of Speaker-Dependent Senses 4.2.2. Types of Modes of Presentation and Default Semantics 4.2.3. The Persistence of Sense 4.2.4. Sense-Based Semantics? 4.3. De Re, De Dictoi, and De Dicto Proper 4.4. Red Giants and White Dwarfs: Context-Dependence of Attitude Ascription . . . . 4.5. Default De Re 4.6. Interim Conclusions 4.7. Loose Ends: Context and Presupposition Projection

122 122 124 130 131 133 135 137 142 142 147 158 161 166 176 183 192 194

DISCOURSE, BELIEFS, AND INTENTIONS Current Research in the Semantics/Pragmatics Interface, Vol. 2 K.M. Jaszczolt © 1999 Elsevier Science Ltd. All rights reserved 1 "Di due cose che non stiano insieme, crederle tutte e due, e con 1'idea che da qualche parte ve ne sia una terza, occulta, che le unisce, questa e la credulita." Eco (1988: 47).

121

122

Discourse, beliefs, and intentions

Chapter 4

The aim of this chapter is to present and defend the view that the interpretation of belief reports is governed by the referential intention that assigns the default status to the de re interpretation. In the preceding chapters I made frequent reference to intensional contexts, substitutivity of coreferential expressions, MoPs, and the de re/de dicto distinction. These terms are of primary importance for any analysis of reports on people's beliefs and require a more detailed discussion at this point.

4.1. Prepositional Attitudes: A Close-Up 4.1.1. Prepositional Attitudes, Attitude Sentences, and Attitude Reports A propositional attitude is a relation between a person (or other organism) and a piece of information in the language of thought, be it natural language as Chomsky suggests, or a universal language of the brain that is used for mental computations, as Fodor maintains. This relation can be regarded as computational and holds between the organism and a formula coded in the system used by this organism. The role of real world situations in individuating such mental states or processes is still a moot point, although methodological solipsism is found untenable, as the discussion of Fodor's (1994) recent research in Chapter 3 points out. It was also established in Chapter 3 that beliefs have direct affinities with linguistic expressions and these affinities pertain to their shared property of intentionality. Beliefs are intentional states (or: experiences, see Woodruff Smith 1989: 11) and intentional states have content. They can also have a referent, an object to which they refer. Similarly, every linguistic expression has meaning (or: sense, albeit not necessarily Fregean) and can also have a referent. In fact, intentionality as a theory is not remote from the theory of sense and reference: mental states and linguistic expressions are both directed at objects and have meaning: "Intentionality theory may thus be viewed as the 'semantics' of thought or experience, encompassing the theory of intentional content, or 'sense', and the theory of intentional relations, or 'reference'. (But the concern is experience, not language, which is the subject matter of semantics in the usual sense.)" Woodruff Smith (1989: 11). As a result, one can formulate conditions of satisfaction for mental states which parallel truth conditions for sentences in Tarskian semantics. Now, Woodruff Smith (ibid.: 16) admits that there is a problem at this juncture of experience and language because the latter may express the first with

Section 1

The default de re principle

123

different degrees of accuracy. This problem disappears when language is regarded as one of the carriers of thought. This idea is developed in Chapter 6 below. All that matters for the present discussion is that by looking at the properties of beliefs one can reveal some relevant properties of sentences that express them - a platitude often neglected in the research on expressions of propositional attitude. Beliefs have also other features in common with linguistic expressions. Utterances contain, so to speak, or carry in themselves certain attitudes to their contents. Belief is the most basic of these attitudes. No other clues are necessary to attribute a belief to a person than the mere fact that the person uttered a sentence in the indicative mood. In other words, there is no need to express by lexical means that the statement represents a belief. An assumption to this effect is derivable in a process of inference from the mere syntactic form of the uttered sentence. Sperber and Wilson (1986a: 73) say that belief is a fundamental factual assumption present in the mind. Another example of an attitude conveyed by syntactic means is desire which corresponds to the indicative mood. So, the scope of analysis is large indeed: a belief expression is an expression in the indicative mood.2 The main focus of this chapter is the relation between beliefs, such as for instance the belief that the author of David was the best Italian sculptor, expressions of belief, such as (1), and reports on beliefs, such as (2):

(1)

The author of David was the best Italian sculptor.

(2)

Kasia believes that Michelangelo was the best Italian sculptor.

The appropriateness of (2) will, naturally, depend on many factors. The impact of these factors on the semantics and pragmatics of propositional attitude reports is, in a great approximation, the topic of what follows. The topic is vast because there is no one-to-one correspondence between mental states and speech act types; there are many ways of expressing a belief (cf. Mey 1993: 156). It is also central to current debates because attempts at incorporating the context-dependence of attitude ascription in the theory have not been entirely successful (see Richard 1990, 1995; Soames 1995).

2

However, in the tradition derived from Plato (cf. Lee 1953), belief has also been regarded as a state weaker than knowledge, or as failed knowledge, an attitude that lacks the sufficient certainty in order to be knowledge. See also Wierzbicka (1988: 132-136) on the primary role of the verb 'know'.

124

Discourse, beliefs, and intentions

Chapter 4

4.1.2. The 'Ortcutt Puzzle' and Its Interpretations It seems that an investigation of attitudes and reports can only be successful if it commences, so to speak, at the chronological end of the belief—> expression -> report scenario, that is with sentences of the type (3), as exemplified in (2): (3)

A believes that B Eap (where Eap stands for 'the person a believes that />'), as long as the subject of the proposition p is presented to him/her under different MoPs and the holder of the belief does not realize that the beliefs concern the same individual. The difference between proper names, which are commonly considered to be directly referential, i.e. have individuals as their semantic values, and definite descriptions is not significant in this respect, they both create problems of reference in the interpretation of attitude contexts (cf. Burge 1977: 342).8 The properties of expressions that are used to refer are attended to in Chapter 5. The problem with prepositional attitudes is tri-partite. Firstly, there is a problem of belief ascription or, in the above example, with determining, on the basis of his belief expression, what Ralph believes. Secondly, there is a problem with reporting on this belief without distorting the original belief. Thirdly, there is a problem with the interpretation of a belief report. This is the setting of the scene for semantics and pragmatics, large as it may seem. The tradition in the philosophy of language narrowed the field substantially and concentrated on the formal semantic properties of sentences reporting on beliefs. It is perhaps this narrow focusing and a virtual neglect of For a discussion of some problems with direct reference see Section 4.2.3.

130

Discourse, beliefs, and intentions

Chapter 4

pragmatics, as well as the neglect of addressing the issue of the interaction of semantics and pragmatics that have led to repeated failures in providing a successful account of the meaning of prepositional attitude sentences. This chapter signals how to remedy this state of affairs and, at the same time, demonstrates how attitude sentences yield to the Default Semantics sketched in Chapter 2. 4.1.3. Factives and Non-Factives

In Chapter 3, the de re/de dicto distinction was contrasted with the distinction between factive and non-factive attitudes and attitude verbs. To repeat briefly, factives are those verbs which generate propositions whose truth entails that the embedded proposition is itself true. For example, A knows that />' presupposes that p is true. 'Know' is an example of a factive verb. On the contrary, non-factives do not require that the embedded proposition is true. 'Believe' and 'think' are examples of non-factive verbs. Moreover, 'wish', 'hope', or 'imagine'are sometimes called contrafactives or counterfactives since they can be said to presuppose the falsity of the expressed embedded proposition (cf. Crystal 1980). Other classifications are also present in the literature, for example Partee's (1973) distinction between emotives (e.g. 'regret'), verbs of inference (e.g.'deduce'), verbs of communication (e.g. 'say'), manner-verbs of communication (e.g.'whisper') and epistemic predicates (e.g. 'believe', 'know'). They have different grammatical properties but, in the case of the latter group, the distinction into factives and non-factives is not preserved.9 It has been common practice in the studies of attitude ascription to stress the problems posed by non-factive attitudes, and in particular belief. The same principle is adopted in this work. This practice is dictated by the fact that in the case of belief there is no commitment on the part of the holder of the 9

Some other typologies of attitude verbs have been proposed. One of the more interesting ones is that of Asher (1987), established on the basis of (1) exhibiting by these verbs anaphoric properties, and (2) observing the rules of logical inference. The property (1) means that certain attitude verbs such as believe contain a logical barrier to anaphoric relations. Sentence (i) is obviously problematic. (i) Tom believes that unicorns exist. One of them is behind the sofa. He suggests parallel divisions of attitude verbs into: (1) factives ('know') and non-factives ('believe'); (2) negative and positive; (3) definite and indefinite (definites require some background justification, their truth conditions require another attitude as their justification). This leaves e.g.'believe'and 'think'as positive indefinite non-factives, and e.g.'expect" and 'decide' as positive definite non-factives - along with seven other categories which are of lesser importance for the present concern.

Section 1

The default de re principle

131

belief or the reporter on the belief to the truth of the proposition expressed or incorporated in the report, respectively. Hence, these attitude expressions are the most interesting ones through being the most difficult ones to analyse.

4.1.4. The Object of Belief The object of belief has been a subject of controversy at least since Frege brought prepositional attitudes to the centre of-attention. In Chapter 3, it was accepted that beliefs are directed to real objects through mental constructs called noemata. They have also been regarded as relations between the believer and a proposition, a sentence, statement, fact, state of affairs, or some other entity.10 In fact, states of affairs are compatible with phenomenological directedness at an object. I suggested that when the referential intention is weak or non-existent, the belief is directed at the situation as a whole. This plurality of views should not be a hindrance, though. If beliefs are intentional and intentions, according to the DI principle, come in degrees, then the object is either (i) a proposition as a whole or (ii) an individual referred to, depending on the strength of the referential intention. Construing propositions as objects of the attitudes has been rather unfortunate. It may be useful for discussing properties of the embedded that- clauses, especially as far as substitutivity of coextensive terms is concerned, but from the point of view of intentionality and the psychosemantics of beliefs, 'proposition' is a confusing term. It has been defined, among others, as (i) a state of affairs; (ii) a sentence; (iii) sentence meaning; (iv) Fregean sense; (v) mental representation; or (vi) a set of possible worlds.11 Moreover, opinions differ as to whether propositions are constant, that is whether their truth value can change with time and location. At least this issue has been resolved in pragmatics by postulating the widely acknowledged intrusionism of pragmatics into the propositional form of a sentence. In other words, it is widely accepted that propositional forms of sentences require further pragmatic specification, including indexical reference, before they are truth-evaluable (see Section 1.3 above). All in all, it can be safely assumed that propositions are theoretical constructs made up for semantic purposes and have no ontological status. The embedded proposition contributes some meaning to the whole expression. But it is not the smallest, indivisible entity which can legitimately be called an object 10 There are a variety of standpoints on this issue and some of them are discussed below. For classical arguments for and against a proposition as an object of attitude see e.g. Prior 1971,1976; Harman 1968-69, and Quine 1960. 11 It has also been claimed that an object of propositional attitude is a property of inhabiting a possible world where the embedded proposition holds. Cf. Lewis 1979.

132

Discourse, beliefs, and intentions

Chapter 4

of attitude. The name 'prepositional attitude' seems to be retained in the technical parlance for historical reasons and is, strictly speaking, a misnomer.12 Acts of consciousness are directed at objects which may or may not exist in reality. In other words, we always believe something. Brentano (1874:124-125; 1874a: 88) calls it 'reference to a content', 'direction toward an object', or 'immanent objectivity'. And every, as he calls it, 'mental phenomenon', contains something of an object. In thinking we grasp a sense, and in the theoretical analysis it is justified to say that we think a proposition or a sentence. It is, so to speak, a shortcut in interpretation which should indeed be regarded merely as a shortcut.13 Propositions and their parts are constructs that function as mediators between objects of attitudes (states of affairs and their inhabitants) and linguistic, syntactic units (sentences and their parts). On this precarious looking basis, 'believe' can be called a propositional verb. For the same reasons, the standard form of propositional attitudes is that in (3), where the verb is followed by a //iaf-clause. In other words, all sentences reporting beliefs are translatable into 'A believes that /?' constructions (see Montague 1973). Numerous attempts have been made to dispose of a proposition as an ultimate object of attitudes and introduce its 'parts'as objects. These attempts conform to Frege's principle of compositionality of meaning which says that the meaning of a sentence is composed of the meanings of the words and the structure in which they are involved. Russell (1912) claimed that we are acquainted with parts of propositions. He later added that beliefs are many-termed relations (Russell 1984), and subsequently dubbed propositions 'psychological occurrences'of complex images when they appear in attitude contexts (Russell 1940). More adequately, Cresswell (1985) suggested a semantic account of belief reports based on 'structured meanings', the sensitivity of the complementizer that to the structure of the following f/za/-clause. This account is used in Chapter 8 for the contrastive, cross-linguistic analysis of belief reports. But breaking a proposition into pieces will not suffice. A proposition is a semantic object

12

Fodor (1978: 520) points out a different problem connected with accepting proposition as an object of attitudes. Propositions do not have forms, they are 'sheer contents', the contents of what is conveyed through different lexico-syntactic forms. In other words, there are various ways of saying the same thing and these ways are disregarded in propositions. The problem arises when one tries to compare these contents with mental representations. Propositions are public, representations are private, and hence the public meanings have to be acquired and made private, or grasped like Platonic ideas. According to Fodor, the system of such representations constitutes a language of mental operations. 13 Brentano and Frege rely on a mysterious procedure of 'grasping' sense. This idea will not be developed here. See Dummett 1991b.

Section 1

The default de re principle

133

whereas intentionality requires a physical one, mediated by a mental representation. That is why we shall need Husserl's noema (see Chapter 3) and sense or MoP discussed in Section 4.2. The view that the verb 'believe' expresses a relation between an individual and a proposition can only be defended when the structure of this proposition is recognized as part of this proposition, i.e. as part of the informational content of the that-clause (cf. McKay 1991: 726). Attitude reports with personal pronouns as subjects of the that-clause that are coreferential with the expression in the subject position of the main clause are good proof of this, and so are that- clauses with reflexive pronouns. In addition, the contrast between de re and de dicto reports also relies on the degree to which the that- clause is transparent. In other words, the contrast depends on (i) whether the referent is assigned, and if so, whether it is a correct one, and (ii) whether the expression is understood by the speaker or merely repeated 'whatever it may mean', as in the example with red giants becoming red dwarfs used in earlier discussions (cf. Chapter 3, example 12). This information from the structure of the that- clause varies from one interpretation of attitude report to another and from one situation to another. Sometimes more information is relevant, sometimes less. For example, a de dicto belief report exhibits a small sensitivity to the structure of the that- clause as compared with its de re equivalent. By 'the equivalent' I mean a belief report that has the same linguistic form as the first one. This topic is developed in more detail in Chapter 8 and in Jaszczolt 1993. To sum up, prepositional attitudes, like other intentional mental states, are relations to real objects, as was established in Chapter 3. They are relations mediated by noema or sense, but in semantic discourse they can be conveniently regarded as attitudes to propositions or their parts. But, until we know how to classify the possible objects of attitudes as to fineness of grain (states of affairs, construed objects and real objects), the interpretation of prepositional attitudes remains a clandestine enterprise. This is where unmarkedness and default in interpretation come in.

4.1.5. Defaults for Belief Reports or Defaults for Referring Expressions? The options of interpretation of propositional attitude reports have close affinities to the choices of interpretation of expressions that are used to refer such as proper names and definite descriptions. Other expressions in the subject position of the embedded sentence do not create such problems. For example, sentence (15) can be taken to be about a specific cat, or the indefinite noun phrase can be taken in a nonspecific sense as in (15a) and (15b) respectively:

134

Discourse, beliefs, and intentions

(15)

Ralph believes that a cat talks.

(15a)

There is a cat about which Ralph believes that it talks.

(15b)

Ralph believes that there is a cat such that it talks.

Chapter 4

The readings of (15) differ in the relative scope of the indefinite noun phrase and the belief operator. The readings of reports with proper names and definite descriptions are slightly different from (15). They also involve the instances of referential mistake, and hence the choices of interpretation involve the referential reading, the attributive reading, and referential including a referential mistake. So, we can conclude that the choices of interpretation of belief reports follow the choices of interpretation of the embedded sentence. Following Burge (1977: 342), definite descriptions and proper names are regarded as posing similar problems of individuation of the referent. Despite their acclaimed differences attended to in Chapter 5, they, in a sense, constitute one category of what Mill calls individual names, in that they are ideally affirmable truly of only one thing. However, whereas, according to Mill, proper names are non-connotative, i.e. they do not carry information as to the attributes of the individuals (which is why they may be said not to carry meaning), definite descriptions do more than merely denote an object. For Frege (1892), definite descriptions belong with proper names, the definite article makes a proper name out of a concept (Begriff).14 It seems that a moderate solution in-between the two is preferable. Namely, proper names and definite descriptions exhibit similar problems of interpretation but neither is a sub-class of the other. Denotation, reference (or Frege's Bedeutung, see Section 4.2 below) makes them similar. Both definite descriptions and proper names are, according to Frege, what we can call terms in logic. They are expressions referring to an entity, they also have sense, as well as its subjective correlate, MoP. Sense, or meaning, determines reference but it is possible to know the meaning without knowing the reference. Both proper names and definite descriptions can have different referents (if any) in various situations.15 They are not rigid designators, they can have different referents in different possible worlds (pace Kripke 1972 on proper names). It is so due to their context-specificity. In short, proper names have MoPs which may explain the apparent inconsistency of beliefs. The MoPs may be mistaken and lead to referential mistakes in the use of proper names, just 14 In languages which do not have articles the role of the definite article is said to be fulfilled by suppositio ^implex orpersonalis). However, this solution seems not to conform to Occam's razor: supposition is not a constituent of the sentence. 15 This is independent of the speaker's intentions according to Frege. Sense is objective, not individual. See Section 4.2 below.

Section 1

The default de re principle

135

as they may do in the case of definite descriptions. The believer's MoPs in oratio recta (the belief expression) may not be the same as the reporter's MoPs in oratio obliqua (the belief report). Similarities are important for the present concerns because we are interested in the ways people refer rather than the ways expressions refer. The differences are attended to in Chapter 5.

4.1.6. Varieties of De Re and De Dicto The distinction between de re and de dicto readings of belief reports is standardly tied to the difference between talking about a person, so to speak, independently of the names the person possesses and the description he/she undergoes, and talking about a person with respect to one particular name or description, whoever it may be. In other words, a belief report is de re when it is uttered about a particular thing fyes) and its having a certain properly, and it is de dicto when the believer is merely able to assert that a certain proposition Dictum) is true. The distinction also holds for beliefs: there are beliefs de re and beliefs de dicto, although the distinction pertaining to beliefs is a matter of controversy in the literature.16 Searle (1983: 208-209), for example, claims that all beliefs are constituted by their intentional content and their psychological mode and since both of these are in the head, all beliefs are de dicto and the distinction is redundant. All there is to the distinction is that some beliefs are additionally about real things and hence de re. De re beliefs are a subclass of de dicto beliefs. This objection to the distinction is, however, a result of redefining the purpose for which the distinction is used. It is used primarily for distinguishing between holding a belief about a known, salient individual and holding a belief, so to speak, 'second-hand', about whoever undergoes a name or description used in the source of the belief such as a newspaper article, information delivered by a friend, a teacher, heard on the radio, etc. It is true that all beliefs contain MoPs of their objects but the role this mode plays in the individuation of the object differs. Similarly, the MoPof the individual referred to in the subject position of the embedded sentence of a belief report can be more or less vital for the content of the report. And this is what the de re/de dicto distinction is supposed to grasp. It does not mean, however, that one is a belief concerning a thing and the other a belief concerning a proposition. Instead, it is the amount of information concerning the possible guises of the 16 The standardly accepted definitions of de re and de dicto come from Hintikka 1969a, esp. fn 18 on p. 443: "One thing at which this old distinction aims is obviously the distinction (...) between statements about whoever or whatever meets a description and statements about the individual who in fact does so."

136

Discourse, beliefs, and intentions

Chapter 4

individual about which/whom the belief is held that constitutes the difference. This emphasis on the amount of information is reflected in the definition of de re and de dicto in terms of substitutivity of coreferential expressions discussed in the following section. There have been many other attempts to undermine the usefulness of this distinction. De re beliefs have frequently been reduced to de dicto. Dennett (1982: 82-84), for instance, argues that there cannot be any de re beliefs because in many cases the believer cannot identify the object although he/she is under the impression of being able to do so. He gives an example of believing that a certain penny coin is a lucky charm when unbeknownst to the believer the coin has been replaced by another identical one. Similarly, Schiffer (1978: 203-204) talks about there being no interesting de re/de dicto distinction due to the role MoP plays in attitude reports. Assuming that an object of belief is always given to the believer under a certain MoP or Fregean sense, this is not surprising. On the other hand, Bach (1982: 121) observes that if people had no de re beliefs about objects, people's knowledge would be acquired merely by description and this is certainly not the case. We can think of objects directly. We can, so to speak, have an object in mind rather than merely think in concepts associated with it. In other words, people are capable of pointing to an object of their belief when they see it and this requires de re beliefs. In a similar spirit, Burge (1977: 349) argues for the necessity and the fundamental character of beliefs de re as follows: "... having de re attitudes is a necessary condition for using and understanding language - in fact for any prepositional understanding - and for acquiring empirical knowledge." Beliefs de dicto are not equally necessary. Now, even if beliefs de re are a sub-species of de dicto, they are so only on one rather stretched understanding of de dicto, namely as having MoP or sense. Since Fregean senses are public and objective meanings anyway (see Section 4.2 below), this is not a serious obstacle. De dicto means that belief is a sentential rather than predicate operator and hence the difference lies not in the presence or absence of MoP or sense, but rather in the damage this presence causes to the transparency, clarity and intersubjectivity of this belief. In order to elaborate on this technical sense of de re/de dicto, some historical facts are required. The distinction as such comes from medieval logic, it is attributed to Abelard (see Honderich, ed. 1995: 188) and was originally applied to modalities. When modal terms such as 'it is possible', 'it is necessary' apply to the subject and its possessing a certain attribute, the modality is called de re ('concerning the thing'). When the modal term

Section 1

The default de re principle

137

applies to the whole statement (proposition), the modality is called de dicto ('concerning the statement'). When we think of sentence (16) as of a de re one, the sentence is true because it amounts to saying that nine is necessarily nine. (16)

The number of planets of our solar system is necessarily nine.

However, on the de dicto reading, (16) is false. It is not true that this number of planets is necessary. In the de re reading substitutivity of coreferential expressions is possible, whereas in the de dicto reading it is not.17

4.1.7. Substitutivity? The problem with belief constructions is usually classified in the literature as a problem with the applicability of Leibniz's Law. Leibniz's Law states that two things are identical with each other if they are substitutable preserving the truth of the sentence, or, in Leibniz's original formulation, "Eadem sunt, quae sibi mutuo substituipossunt, salva veritate" (quoted after Frege 1892: 64). This formulation of the law is based on the more general claim dubbed the identity of indiscernibles: if things have all properties in common, then they are identical, they are one thing (see Lacey 1976: 124). Two observations have to be made before we proceed to discussing the applicability of this law. Firstly, when we talk about substitutivity in belief reports, we talk about words, terms, or expressions that are substituted for each other rather than 'things'. In other words, if all that can be said of one thing can be said of another, then it is said about one thing under two guises. Secondly, there is another part of what is nowadays called Leibniz's Law which is in fact due to Quine. It states that if things are identical (i.e. if they are one), then they have all properties in common. This law is dubbed the indiscernibility of identicals. So, if we take one thing described in different ways, all that can be said of the thing under one guise can also be said of it under the other guise. Leibniz's Law is frequently confined to the latter principle, with no regard for the historical facts. In other words, "... if A and B are identical, then anything that is true of one is true of the other: Vx Vy (x = y -> (F(x) -> F(y)))"

Neale (1990: 124).

17 There is also a third group of beliefs, distinguished e.g. by Castaneda and Perry, called beliefs de se. These are beliefs about oneself which are not merely beliefs de re about oneself. We shall not be concerned with this category. See e.g. Lewis 1979 and Cresswell 1985.

138

Discourse, beliefs, and intentions

Chapter 4

Left between Scylla and Charybdis of historical truth and contemporary use, we shall adopt the wide sense of the law, allowing for its extension to the indiscernibility of identicals and to the applicability to the substitution of terms, remembering, however, that this was not the intent of the original law put forward by Leibniz. Sentences ascribing prepositional attitudes have sometimes been called hyperintensional in that expressions that can be substituted salva veritate outside prepositional attitude contexts and direct quotations cannot always be substituted for one another in attitude constructions (see Richard 1997: 198). These expressions have the same possible-world intension and yet are not substitutable. For example, proper names of the same individual are substitutable salva veritate in modal contexts but not in the context of attitude reports. Even apparent synonyms, words which seem to have identical semantic properties, are not always substitutable.18 However, instead of multiplying types of intensional contexts by adding 'hyperintensionality', it may be more economical to look into intensionality more carefully and distinguish degrees to which the MoP of the individual is relevant for its semantics. This is what I attempt to do below. For the purpose of semantic analysis of prepositional attitude expressions, the de re/de dicto distinction began to be redefined, rather unfortunately, as distinguishing between the cases where substitution of coextensional terms holds and those where it does not.19 Quine (1956) saw de re beliefs as relations between believers and real objects, and hence as relations that can be captured by the logical form of the report in which 'quantifying into' the belief context is possible, as in (17a) (cf. also Searle 1983: 210). (17)

Anne believes that Einstein is a film director.

(17a)

3x (Einstein (x) & BelA (Film Director (x)))

This is a de re reading. Anne believes something about Einstein. In contrast to (17a), the reading in (17b) is de dicto. Anne believes something about somebody whom she calls 'Einstein' in her belief world. 18

Richard (1997: 199) gives the following example: (i) Tyler thinks that some doubt that all who shout yell, (ii) Tyler thinks that some doubt that all who shout shout. Sentence (i) may be true while (ii) is false. This also presents a problem for the compositionality of semantics. 19 It is sometimes claimed that substitutivity holds in attitude reports but this fact is obscure to the believer. Cf. Barwise and Perry's (1983) semantic innocence, and the original account of 'saying that' in Davidson (1968-69: 142-143). Davidson proposes that the embedded sentence has no semantic connection with the main sentence expressing the act of saying. For criticism see Platts 1979; Cresswell forthcoming.

Section 1

(17b)

The default de re principle

139

BelA (3x (Einstein (x) & Film Director (x)))

Naturally, on this account, names are not rigid designators. The same analysis concerns definite descriptions such as in the example (18): (18)

Anne believes that the man whom she saw in the coffee shop is Tom's uncle.

The referring expression may refer to the person whom she in fact met or to the man whom she thinks she has met. Equating the de re/de dicto distinction with the difference of scope will not do, though. There are three interpretative options for beliefs and belief reports, namely de re, de dicto, and de re about someone else in the case of referential mistakes. This tri-partite distinction cannot be captured by differences in scope. It is a well known fact that truth-preserving substitutivity of coextensional terms does not hold for intensional contexts including belief reports. But it is frequently forgotten that the law works to some extent, that there are belief reports that allow for such substitutivity. Unfortunately, equating these cases where the law holds with de re reports is a frequently made mistake and a cause for confusion. Reports can be de re without being transparent with respect to substitution. If substitutivity was the criterion for assigning the de re or de dicto status to belief reports, the case of a referential mistake called here de re about someone else would have to be renamed a sub-type of de dicto because substitutions cannot guarantee preserving the meaning of the original report. They cannot guarantee preserving the truth value, neither can they guarantee preserving the believer's consent to the report. An alternative account, taking these three options of interpretation onto consideration, is developed in Section 4.3 below. To sum up, the problem of substitutivity salva veritate differs from the problem of context-dependent substitutivity in discourse. The interpretation of belief reports is a much wider problem than the narrowly focused intra-theoretic problem posed by Frege, Carnap, or Quine. Philosophers frequently presuppose that it is clear whether or not the speaker can truthfully say "A believes that p" or "A believes that q" in the circumstances which they describe. They are interested in the relation between p and "A believes that p", and p and "A believes that q" where p and q have the same truth value or, at best, predicate the same thing about the same object. But the conversational (and hence semantic and pragmatic) problem of interpreting belief reports is not founded on the search for reasons of the failure of substitutivity of coextensional terms. Rather, it is founded on the criteria for distinguishing between the readings where the speaker talks about the salient, known individual, commits a referential mistake, and talks about

140

Discourse, beliefs, and intentions

Chapter 4

whoever fits the name of description. Substitutivity salva veritate comes in later. It comes in as a result of the transparency of the report, that is as a result of the situations where the speaker is capable of individuating the object of belief to the extent in which substituting coreferential expressions would still trigger the believer's consent to the report. To the extent' is vital here because "... there are sentences where substitutivity fails at the surface level, but which are nevertheless de re" (Burge 1977: 341) - in the sense of being about a particular res. In fact, not only the conversational substitutivity, but also the substitutivity salva veritate has to rely on a certain degree of context-dependence. This context-dependence is by now a standard feature of dynamic semantic theories (see Chapter 8 on DRT) and accounts of attitude ascription (see Richard 1995; Soames 1995). That is not to say that it does not cause problems. The degree of this semanticization of contextual information still poses problems for the latter (see Soames 1995: 540) and requires a theory - perhaps the theory of degrees of MoPs from Section 4.2 below. In brief, the goal of the semantic-pragmatic analysis is not to specify conditions under which "A believes that p" is synonymous with "A believes that q", neither is it to produce a principle of substitution for oblique constructions to replace Leibniz's Law. The data is constituted by an expression of belief and a report on a belief, rather than by a pair of belief reports which differ with respect to the form, but not reference, of the expression in the subject position of the embedded sentence. It also has to be emphasized that the expression which leads to options of interpretation need not be in the subject position of the embedded sentence. Sentences (19) and (20) yield to the de re/de dicto duality of reading equally easily:

(19)

Ralph believes that he can see Ortcutt.

(20)

Ralph believes that Jane and Mary met at John's.

Hence, defining the puzzle in syntactic terms is overly confining. This confinement stems out of the preoccupation with Tarski's theory of meaning and truth (see Platts 1979; Wqjcicki 1994) and the conviction that the structure and properties of neat and regular formal languages open a window for an analysis of unorderly and fuzzy natural languages. The structure of sentences of both types of languages is delineated by the distinction between predicates and their arguments. Sentences are composed of, first of all, predicates and arguments. Predicates take a certain number of arguments, and arguments are realized as logical variables or constants

Section 1

The default de re principle

141

that stand for objects. Further, they are also composed of quantifiers and connectives.20 The theory of truth of the sentence relies on the correspondence of the sentence with reality. This theory is the basis of the theory of meaning. Truth defined for a particular formal language is a point of departure for a semantic analysis of truth: a sentence p of a language L is true iff it is the case that P, where P is the translation of/? from L into a metalanguage ML which is at least as rich as L. This is Tarski's Convention T (cf. Tarski 1933, 1944; Wqjcicki 1994: 430). Now, formal languages differ significantly from a natural language such as English. They are extensional in that the meaning of the terms can be identified with their referents and hence substitutivity of coreferentials is truth-preserving, it preserves the logical value (truth or falsity) of the sentence. In natural languages, the sense (or meaning) of an expression depends on the context of utterance. This meaning can sometimes be vague or underdetermined, as was discussed at length in Chapter 1. In addition, natural languages can be used self-reflexively, for talking about themselves. As a result of these characteristics, the syntax of formal languages is useful for the analysis of English and other natural languages only to a certain extent. There is meaning that can be reached only through an analysis of mental states: beliefs, desires, or intentions. Context-dependence plus (or: including) meaning derived from what is 'in the head' have been frequently dubbed by logicians logical pragmatics and ignored as inaccessible, not yielding to any formal analysis. Philosophers of language in the Fregean and neo-Fregean tradition made significant progress by assigning an important role to sense or MoP. But even then context in the wide sense of a situation was not assigned a clear semantic role. This role of MoP is the topic of Section 4.2. By vindicating mental states to the role of a component of semantic theory, logical pragmatics becomes an overly coarsely-grained term. It becomes evident that it comprises in fact two separate stages of utterance interpretation: (a) part of sentence meaning in the theory of Default Semantics as sketched in Chapter 2, and (b) implicatures. This state-of-the-art is not a departure from Tarski's semantics, though. It means rather that Tarski's theory of meaning and truth is assigned a more appropriate, albeit more modest (and ever decreasing), place and role. It is so because the point of departure of semantic analysis is not language and its reflection of reality, but rather representations that belong to the language user and the way these 20

On the preference for generalized quantifiers see Barwise and Cooper 1981; Neale 1990, and Kamp and Reyle 1993.

142

Discourse, beliefs, and intentions

Chapter 4

representations reflect reality. In this account, the role oflanguage is modest: it is one of the vehicles of thought. And if language shapes representations, then it moulds our understanding of the reality in the Whorfian sense (see Whorf 1956: 214; Gumperz and Levinson 1996: 21).21 But this is a separate topic discussed in Chapter 6.

4.2. Whose Meaning? On Sense and Mode of Presentation 4.2.1. In Defence of Speaker-Dependent Senses People always see things under an aspect. The famous duck-rabbit figure (cf. Wittgenstein 1953:194) is the best example of this feature of our perception. The intentional object of our seeing is either the 'duck aspect'of the drawing, or its 'rabbit aspect' (cf. Searle 1983: 52). The same concerns beliefs and other mental states. We always believe something under an aspect. As was established in Section 3.2, propositional attitudes have content, or MoP, which differs from their object. It was noted that two mental states can allegedly have the same content but different objects as in the case of indexical expressions but that this case will not interest us here because this approach to indexicals is not totally satisfactory and because in the present analysis we want to look at the differences in content that one linguistic expression can bring about to the content of the belief. In other words, we shall confine the study to MoPs of a referent, or concepts associated with a single referent. The first question to ask is whether the speaker's MoPs can be known by the hearer, i.e. whether they are intersubjective. For Frege, they are and they are called senses. The sense of a sentence is that part of its meaning that determines its truth value; reference alone does not suffice. It is the way in which the referent is given, or known, to the speaker (cf. Evans 1981). So, sentences have both reference and sense. Reference is equivalent to extension and sense is a way of thinking about the referent, or its intersubjective MoP. As a result, there is a representation of the object in the speaker's mind. According to the compositionality principle, the hearer has to determine the sense of each word in a sentence. In extensional contexts, the reference of a proper name, for example, is the object designated by using it. Speakers also have subjective ideas associated with the use of proper names. But sense is neither a subjective idea nor the object; it is somewhere 21

Cf.: "We are thus introduced to a new principle of relativity, which holds that all observers are not led by the same physical evidence to the same picture of the universe, unless their linguistic backgrounds are similar, or can in some way be calibrated." Whorf (1956: 214).

Section 2

The default de re principle

143

in-between and is an objective MoP attached to the object. Sense does not vary from context to context. According to Dummett's interpretation, it is a property of a word itself (cf. Dummett 1973: 268). Since sense is a property of the word, it does not vary from speaker to speaker. Instead, speakers have private'replicas'of senses, their representations or ideas. And, as Frege clearly warns, it is impossible to compare the images of sense (or: ideas) of two different people. Every representation (Vorstellung) has one owner, and no two people have the same representation (Frege 1918-19: 42; 1918-19a: 335). One word may have different senses and there are three degrees to which words may differ from one another: they may differ in ideas, sense, or sense and reference. So, it is possible for two expressions to have the same referent but differ in sense. Coreferential expressions cannot be substituted in some contexts precisely because of this difference. In Frege's words, sense, or thought, usually becomes the (indirect) reference in such constructions. Thoughts are complex and not very clear entities but they seem to be objective in that they can 'inhabit' various people's minds: "... thoughts are neither things in the external world nor ideas. A third realm must be recognized." Frege (1918-19a: 336-7). They are objective correlates of representations, or objective meanings in the head, essentially similar to those advocated later by Fodor (1981). They are the objective content of thinking and can be shared by several thinkers. On the Fregean semantics, attitudes are relations to propositions which are extralinguistic, abstract entities that have truth conditions and constitute contents of attitudes (cf. Schiffer 1982: 137). It is difficult to reconcile this picture with an account of belief as a relation to some internal representation of the believer's. The notion of sense would have to be individualized, objective sense will not do. But it has to be remembered that people can share MoPs; MoPs are conditioned by language, society, circumstances, and mental background. Hence, they are not entirely subjective. On the other hand, MoPs can, so to speak,'attach themselves to wrong objects' in the case of referential mistakes. For both these reasons, Frege's sense has to be made more finely-grained and more speaker-oriented to be useful in semantic theory and theory of communication. As it is, given a sample of conversation where a report must be made on somebody's belief, the notion of sense proves to be a non-starter. Substitution of referring terms performed in conversation is a problem which can and should be approached with respect to the linguistic practice of how people normally do it, not merely with respect to Leibniz's Law and substitutivity salva veritate. And in order to know how people commonly understand a sentence structure corresponding to reports on

144

Discourse, beliefs, and intentions

Chapter 4

attitudes, we require a theory of interpretation that accounts for the relative priority of available interpretations. This is where intentions fit in and where the theory of Default Semantics sketched in Chapter 2 is required. To sum up, Frege's theory of sense could not be a direct precedent for a theory of attitude ascription for at least one basic reason. It concerns substitutivity salva veritate and ignores the cases of referential mistakes or using the same name without preserving sense or reference between the expression of belief and the report. In order to incorporate these cases, we need a believer-oriented approach rather than the language-oriented one. In other words, we have to reduce as much as possible of the linguistic to the psychological. Referential mistakes were not interesting for Frege. Although he says that the verbal expression of thought is incomplete and must be supplemented by the specification of context, the procedure of recovering sense with which the speaker used the expression is not provided. Presumably, Frege sees the role of context as a source of further information concerning the way in which the referent is given to the speaker or holder of the attitude, rather than as a source of information for the assignment of reference to the expression by the hearer. There is one further obstacle to the proper treatment of context. In Frege's account, the specification of the speaker, time and place of utterance, audience, and other parameters is subordinate to one vehicle of thought, namely language. The uttered proper name, e.g. 'Smith', is the starting point on the basis of which the reporter on the attitude has to select the MoP of the referent. But would it mean that uttering 'Smith' can be associated with a certain way of picking out a person who is in fact, say, Bernard J. Ortcutt? Can 'Ortcutt'and 'Smith' be said to be substitutable salva veritate once the referential mistake of the believer's has been revealed? Even if so, it is an uninteresting question. It only serves as an illustration of the point that binding senses to referents is overly restrictive. If one continued to use Frege's theory having located such instances as the one above, one would have to say that the reporter may use the name 'Smith' in his report although he knows that the believer meant Ortcutt. Frege says that the sense of the belief expression often becomes the reference of the report. But the sense is tied to the referent and there is no principle that would guide the change of the proper name from 'Smith' to 'Ortcutt'. Moreover, trading off reference for sense is not what people do in practice. Firstly, the rule for substituting 'Ortcutt' for 'Smith'cannot come from Frege's theory, and neither can the rule which makes the believer use the name 'Smith' to mean Ortcutt. Secondly, there are belief expressions and belief reports where sense is not relevant and preserving sense is not a necessary condition for the report to be successful; de re beliefs are the best proof of this. In the case of de re beliefs, no reference to MoP is needed. The reporter has no difficulty, so to speak, in recognizing who the believer spoke about. Finally, confining

Section 2

The default de re principle

145

the analysis to one vehicle of thought, namely language, seems to be the over-arching source of Frege's problems. Language is neither the unique, nor the epistemologically prior carrier of thought, albeit perhaps it is the most important one. Let us take stock. Frege held the view that proper names, demonstratives, and other indexical singular terms are descriptional, i.e. they have sense. Sense has three roles to play. It is a conceptual MoP, a way of securing a referent, as well as an informational value of the expression. This view has been challenged since the mid 1960s by many defenders of direct reference such as Donnellan, Kaplan, Kripke, Putnam and others (see Salmon 1986: 65). For instance, a version of Putnam's Twin-Earth argument discussed in Chapter 3 demonstrates that the conceptual content is not the total meaning of proper names. The next question to ask is whether there are objective senses at all. Salmon (1986: 121) argues that, in Fregean theory, in using a name, the reporter refers to his or her own sense or MoP. But this does not seem to be correct. We refer to our MoPs only if we take them to be intersubjective senses. This is confirmed by the presence of the communicative, informative and referential intention present in communication and acknowledged in Default Semantics. This intersubjectiviry is also in agreement with common sense intuitions concerning the practice of discourse. The intended meaning is conveyed successfully because of the presumption of this intersubjectivity. But frequently, there is no 'typical', 'conventional' sense attached to an object. Sense varies intersubjectively. This seems to be the main drawback of Frege's proposal. There are two obvious ways out of it. Either we bind sense to the speakers and hearers rather than objects, or we bind it to objects but allow for context-dependence of reference (cf. Anduschus 1997). Since Frege made only cursory remarks on context, it is difficult to assess his view. He talked about context in his discussion of connections between thoughts and how thoughts can affect one another (Frege 1892). But later on, sense as a way of thinking about an object was developed as a more detailed account of context e.g. by Forbes (1990; for criticism see Crimmins 1993 and Richard 1993. See also Forbes 1997 on logophor). Forbes says that people receive de re information and their operating system creates a dossier where this information is held and where further information is added. This idea seems to be akin to dynamic semantic theories such as DRT (Kamp and Reyle 1993 and Chapter 7) and File Change Semantics (Heim 1988) which take the contribution of context to semantics much further thanks to the idea of pragmatic intrusionism and context- updating. However, instead of making reference heavily context-dependent, we should make the notion of sense more flexible. Firstly, it is a device used by speakers and hearers to arrive at correct referents and it should remain as such. Secondly, its objectivity is a matter of degree. It is shared by speakers insofar as language is shared and

146

Discourse, beliefs, and intentions

Chapter 4

conventions of social behaviour including linguistic behaviour are shared. They are shared while also allowing for idiosyncrasies, individualism, unpredictability and creativity. To put it in terms of the Default Semantics of Chapter 2, while there are default interpretations, these defaults can be overridden when the situation gives us clues to do so. All in all, there have been numerous attacks on Frege's notion of sense from the front of truth-conditional semantics and direct reference. The main drawback of sense seems to be that sense is tied to the referent, it is supposed to determine the referent, but no way back from reference to sense is admitted. Now, in utterance interpretation, there are no objective senses that fix a referent. There are subjective MoPs that may even be mistaken. This was not Frege's concern and Frege's theory of sense and reference should not be stretched to account for utterance interpretation. The concern in truth-preserving substitutivity leaves aside the whole range of situations in conversation where substitutions are made in reports and go through unnoticed, irrespective of the consent or its lack on the part of the believer. It seems that neo-Fregeanism in semantics and pragmatics simply means that there is a need for some mental representations that would fix the reference in problematic constructions. The role of the speaker of a belief report is to preserve reference rather than sense. Sameness of sense seems to be a sufficient condition for substitution of coreferentials but it is not a necessary one. Sense may be a useful device to talk about such qualitative differences as that between the Morning Star and the Evening Star on the one hand, and Venus on the other. The situation is different when there is an unspecified referent such as, on one reading of Situation B, not knowing who Ortcutt is, or not knowing the name of the object of belief as it may be the case in Situation A. Frege's account is too general: in Situation A, the name 'Ortcutt' in John's sentence has to have the sense of 'the man in the brown hat' of Ralph's expression as reference in order for the report to be true. Now, sameness of sense, although sufficient, is not necessary for substitutivity. There are other types of cases of attitude ascription than merely those relying on sameness of sense, i.e. sameness of the way of identifying the referent. It is obvious that some version of sense or its psychological equivalent is needed to account for belief reports and whatever it is, it plays a role in the semantics of these constructions. But Frege's sense is too restricted in one way, and not sufficiently pliable in another. It cannot be more or less finely-grained, neither can it be speaker or believer-oriented, with its binding to the referent sacrificed. MoPs, mental correlates of sense, have been extensively tried for this role, mainly with an aim to preserve the compositionality of semantics. So, the main objection to Frege's senses comes from their being too finely-tuned. It is perfectly possible to form a correct report on somebody's belief without preserving the original MoP of the holder of the belief. I shall now deal with this often repeated objection (see e.g. Richard 1997: 203)

Section 2

The default de re principle

147

by presenting my development of Schiffer's account of types of MoPs. Here the priority of the de re reading of belief expressions and reports seems a plausible surmise. If senses can be as finely-grained as is required, they can also be ignored when required. In agreement with the Default Semantics, as long as there is no evidence to the contrary, preserving reference will do. But this cannot be achieved in Frege's theory in which, so to speak, de dicto is the default. It is a default in that sense is always a route to reference; no suspension of sense is allowed. In this respect, conversational routines lie outside Frege's domain, they require a speaker-oriented approach.22 In other words, we want to know how to extract sense only when sense is necessary for the specification of the referent.

4.2.2. Types of Modes of Presentation and Default Semantics My solution to the lack of pliability and orientation to speakers of Fregean notion of sense is an elaboration of Schiffer's idea of the type of MoP. This account was first introduced in Jaszczolt (1998a). It can be best summed up as a further voice in the discussion between Schiffer (1992, 1995a, 1996) and Ludlow (1995,1996) concerning the hidden-indexical theory of meaning of belief reports. The first of the main problems mentioned by Schiffer is the so-called meaning-intention problem. I shall focus this discussion on that problem. According to the hidden-indexical theory, belief reports of the form in (3) contain an unarticulated, hidden constituent whose content is established by the practice of discourse interpretation. (3)

A believes that B cps.

The hidden-indexical theory assigns a different semantic role to MoP than the Fregean solution of substituting sense for reference in attitude reports. According to the theory, the embedded sentence refers to the same proposition as does the same sentence when unembedded. The proposition is the same, but the belief report also contains an unarticulated constituent. This constituent is a way of thinking about the proposition, a way which is implicitly referred to. So, the proposition expressed is the same, but the 22 Nota bene, Cohen (1962: 13-14) argues in favour of a temporal concept of meaning. The meanings a word expresses are historical phenomena and one does not need the Fregean concept of eternal meaning and objective sense. Similarly, semantics is temporal: language should be seen as a combination of events and states rather than a system of rules. Then various sources of information can be taken into consideration in order to deal with non-extensional contexts.

148

Discourse, beliefs, and intentions

Chapter 4

price to pay is compositionality of the expressions of the type (3): no successful syntactic equivalent of the unarticulated constituent has been proposed to date. I suggest that only on some occasions of use belief reports of this form contain a hidden constituent. Only when the identification of the referent poses a problem for the hearer does the hearer resort to the way the referent is given to the speaker. An external justification of the restricted role of this constituent comes from the default semantics and the proposal of salient interpretations based on the degrees of intentions in communication, as developed in Chapter 2. As was discussed in Section 4.1, belief ascriptions are infamous for not conforming to the compositional theory of meaning and truth. In order to vindicate compositionality, one has to demonstrate that sentences expressing reports on beliefs have a logical form which makes their meaning compositional. And this is where hidden indices fit in. Schiffer (1977, 1992) suggests that if natural languages have a compositional semantics, then belief reports require a hidden-indexical theory. His version of the theory is based on the tenet that believing is a three-place relation of the form Bel (x, p, m) where 'x' stands for the believer, 'p' is the structured proposition, and 'm' the MoP under which x believes/?. In agreement with Frege, the MoP of a proposition is determined by the MoPs of the objects and properties of the proposition as well as by the place they occupy in the structure of the sentence by which the proposition is expressed. So, for example, sentence (21) has the logical form as in (22), where O* is "an implicitly referred to and contextually determined type of mode of presentation" (Schiffer 1992: 503), a property of MoPs. (21)

Ralph believes that Fido is a dog.

(22)

(3m) (s' that differ only in that 'Cicero' has been replaced by 'Tully', the propositions expressed by them are the same. (b) If 'C s.

This duality of interpretation has normally been regarded as a semantic ambiguity and has inspired extensive research in the areas of philosophy of language but less so in linguistic semantics and pragmatics. As was

Section 3

The default de re principle

167

conceded in Section 4.2, the meaning of a referring expression is too often 'tied to the referent', so to speak, to be of use for explaining referential mistakes. And yet, such mistakes are frequently performed by speakers in conversation. Let us come back to situations A and B from Section 4.1.2. Suppose that using the description 'the man in the brown hat' in (4), repeated below, Ralph intends to refer to Smith: (4)

The man in the brown hat is a spy.

If the hearer of (4) realizes the mistake, he will report on (4) as in (34), perhaps adding (35): (34)

Ralph believes that Smith is a spy.

(35)

He thought that the man wearing a brown hat was Smith.

Similarly, in Situation B, Ralph may be thinking of Smith but mistakenly call him 'Ortcutt'. Putting all particular contextual clues aside, it is possible to classify in a systematic way the possibilities of interpretation of a use of a definite description or a proper name as to reference assignment. In what follows, I propose a new classification of belief reports which has a clear advantage over the extant ones in that it takes into account mental representations both of the holder of the belief and the reporter on it, including also referential mistakes and the lack of identifying knowledge of what/who the belief is about, pertaining both to the believer and the reporter. This exhaustive perspective yields an initially complicated typology which, however, will soon appear to be very simple when checked against discourse strategies. Fig. 1 presents the possibilities of interpretation that are open to the hearer of (4). A similar diagram can be drawn for a speaker's use of a proper name, such as Ralph's use of the name 'Ortcutt' in Situation B.

'the man in the brown hat' 1. a man known to Ralph a. Ortcutt

2. a man unknown to Ralph

b. Smith Fig. 1 (from Jaszczolt 1997a: 317).

168

Discourse, beliefs, and intentions

Chapter 4

Now, by using the name 'Ortcutt' in his reports on Ralph's belief, John can refer to a man of whom Ralph has the knowledge summarized by any of the three options (i.a), (i.b), and (ii): (i.a) a man whom Ralph knows personally and who, according to Ralph, is called Ortcutt; (i.b) a man whom Ralph knows personally and who, according to Ralph, is called Smith; and

(ii) a man whom Ralph does not know. The case (i.b) is possible when Ralph or John are mistaken. It shows that the referential mistake can be attributed either to the believer (the speaker of the belief expression) or the reporter (the hearer of the belief expression). Similarly, the lack of knowledge of the referent can be attributed either to the believer or to the reporter. As a result, the options (i.a)-(ii) can be repeated, when carefully amended, with respect to John's knowledge of the referent of Ralph's expression of belief. So far we have considered three readings of belief reports: de re, de dicto, and de re about someone else. In the latter case, we have not specified who the referential mistake should be attributed to: the holder of the belief or the reporter on the belief. It can belong to either. In both cases there is a discrepancy between the original expression of belief and the belief report as far as the referent correctly identified by the use of the name or description is concerned. I have called the interpretation (i.a) the de re reading, (i.b) was classified as de re about someone else, and (ii) de dicto. However, (i.b) seems to be a hybrid between the de re and de dicto properties. It concerns a particular res, an individual (albeit the one incorrectly matched with a name or description) and hence it is de re. On the other hand, it is not transparent to substitution. If the holder of the belief was referentially mistaken in his belief and belief expression, the reporter would normally substitute a correct name or description in his report. This substitution of non-coreferential expressions signals that the belief report is de dicto in that it repairs some essential lack of knowledge on the part of the believer and the substitutivity between the belief expression and the report does not hold. It corrects an attribution of an incorrect name or description to the individual. It is even more obviously de dicto when it is the reporter who is referentially mistaken. This hybrid de re/de dicto reading of (i.b) is very confusing but it will be simplified shortly. In order to clarify its properties and classify it with the least possible confusion, it is necessary to juxtapose expressions of belief with the classification of beliefs and belief reports. The properties of being de re or

Section 3

The default de re principle

169

being de dicto apply to various kinds of entities. I suggest that the distinction can apply to: (1) beliefs (or thoughts), (2) belief expressions, and (3) belief reports. Definitions of these terms in the literature are ample but not exhaustive (cf. e.g. Recanati 1993; Bach 1987a). Let us first consider beliefs. A belief is de re when it is about a particular thing and its having a certain property, and it is de dicto when the believer is merely able to assert that a certain proposition is true. It is sometimes said that all beliefs are de re because they are about somebody, or that all beliefs are de dicto because they all involve the MoP of the individual, or that such a distinction on the level of beliefs does not make sense (cf. e.g. Searle 1979). But usually, belief de re means a belief about a particular person or thing, independent of the particular way in which this person or thing is referred to in the utterance, whereas a belief de dicto means a belief held with respect to one particular expression, about whoever or whatever undergoes the name or fits the description. This definition is reflected in Fig. 1 as the distinction into 1 and 2. So, generally, there are two types of beliefs. For example, "[b]elieving, of God, that he is benevolent is different from believing that God is benevolent" (in Honderich, ed. 1995: 188). De re belief is also frequently characterized as a belief that is backed by the amount of knowledge which is sufficient to individuate the object of belief.37 The distinction on the second level in Fig. 1, i.e. that into l.a, l.b and 2, reflects the distinction drawn between the uses of belief expressions. In expressions de dicto, the belief is attributed to the whole assertion, whereas in a de re expression, it is attributed to a thing's possession of a certain property (cf. also Flew, ed. 1979: 235). This level of belief expressions is the most controversial one in the present classification. The distinction into de re and de dicto is normally said to concern modalities and prepositional attitude reports, and, at best, also attitudes themselves. Expressions of the form A q>s' exhibit problems with reference ascription which are attributed to the referential properties of definite descriptions, proper names, and other expressions that are used to refer and that can stand in the position occupied by A. On the other hand, these expressions, as a whole, can also be 'about the thing' or 'about whoever fits the name, description, etc.'. Hence, I shall incorporate them into the classification based on the terms de re and de dicto so as to avoid multiplying terminological distinctions without necessity. The advantages of this move will be seen shortly when the terms are juxtaposed with the terms that signify the properties of referring terms in Table 2. Now, in belief reports, if the process of referring ends up picking out the correct individual, the belief report can be called de re, as in (i.a). If the believer is referentially mistaken, as in (l.b) in Fig. 1, the belief report will 37

'Individuate' in Hintikka's (1962) sense. Individuation is a vast philosophical topic that need not be discussed for the present purposes.

170

Discourse, beliefs, and intentions

Chapter 4

either reiterate the mistake or it will repair it, as in (i.b). While the original belief was called de re about someone else, the corresponding report should be properly called a sub-type of de dicto because it introduces referential opacity. Similarly, if the reporter is referentially mistaken and assigns an incorrect name or description, or if both the believer and the reporter are mistaken, then there is referential opacity. Let us call it de dictO]. Finally, if the believer forms the belief without being clear as to the identity of the person or object whom this belief concerns as in (ii), or if the reporter does not know who or what the belief expression reported on was about, such a report is also de dicto.38 To differentiate it from the case of a referential mistake, let us call it de dicto proper. To sum up, I have drawn the distinctions in the following way. Beliefs are de dicto when they concern an unknown individual. Expressions of belief are de dicto when they are about an individual unknown to the speaker. Now, reports on beliefs are de dicto when the socially recognisable way of referring is not present due to either [1] the lack of knowledge on the part of the holder of the belief and his holding a belief de dicto or [2] a referential mistake on the part of the holder of the belief while holding a belief de re, or [3] and [4] the lack of knowledge on the part of the reporter analogous to [1] and [2]. The cases [1] and [2], if known to the reporter, are normally attended to in the report by a substitution of a correct expression, by additional hedging and qualifications, so they can be eliminated from the discussion. Other combinations of the believer's and the reporter's beliefs can also be eliminated. As a result, de dicto reports normally correspond to [3] and [4]. When the speaker is referentially mistaken and the reporter reiterates the mistake due to failing to recognize it, miscommunication occurs and the report is about an individual different from the object of the original belief. This point is very important as it eliminates some of the confusing sub-types of de dicto which could not be ignored earlier because the analysis is founded on mental representations. They can now be excluded as they do not pose a problem for discourse. Languages have ways of distinguishing them overtly. However, the situations that correspond to [1] and [2] are only known by the reporter if contextual clues suggest any such problems with the knowledge of the referent on the part of the holder of the belief. Otherwise, we assume the default which is the correct match between expressions and individuals, as well as some lde re acquaintance' with the individual about which/whom the belief is held. I shall say more about defaults in Section 4.5. Now, hedges can take many forms, two of which are exemplified in (36) and (37): 38

Cf. Evans 1982, esp. pp. 64-85 on individuation and 305-342 on invoking possessed information in utterance interpretation, and Dummett (1993c:143).

Section 3

The default de re principle

171

(36)

Ralph believes the fact that Ortcutt is a spy, whatever it means to him.

(37)

Ralph believes that Smith is a spy, although he mistakenly calls him 'Ortcutt'.

The report (36) exemplifies [1], while the report in (37) is an example of [2]. The nuances of the de re/de dicto distinction attended to above are standardly neglected in analyses of attitude ascription. In this account, the remedy conies with seeing the varieties of interpretation as a conversational problem that cannot be exhausted in the semantics of prepositional attitudes and the semantic problem of substitutivity salva veritate. The 'conversational' perspective means the concern with the difficulties that interlocutors encounter in understanding each other's utterances and in reporting on each other's beliefs. The ambiguity of utterances can be called a pragmatic ambiguity because it concerns reference assignment as performed by a speaker in a conversation. The speaker either refers correctly, refers incorrectly, or uses a name or description without being able to individuate the individual about whom he/she holds the belief. The semantic debates are thus merely tangential to the present concern. From our perspective, both proper names and definite descriptions are expressions used to refer. Even if descriptions refer only with the aid of context rather than directly,39 one should still pose a question as to whether this referring role is their salient (or default) role in conversation. This is attempted in Section 4.5, with the help of intentions and intentionality. I proposed that referential mistakes and referring to an unknown person, both on the part of the reporter as in [3] and [4], render de dicto reports, namely its two sub-cases. It is only in de dicto proper where res is not known to the reporter. But the tri-partite distinction in Fig. 1 requires a further cognitive justification. In order to present the relation between cognition and its objective correlate, let us assume for a moment, to simplify the discussion on representations from Chapter 3, that the reality is represented in the mind in the form of what Fauconnier (1985) calls 39

According to the theories of direct reference, proper names always 'refer' directly, i.e. their 'descriptive content' is irrelevant for truth conditions (cf. Recanati 1993: 297-298), whereas definite descriptions only do when the context (situation) makes them do so. Be that as it may, the speaker still uses both types of expressions to refer and to say something about a known individual, about whoever undergoes the description/holds the name, or about a misdescribed/misnamed individual. Descriptions can be used referentially and express a singular proposition. This is achieved by means of contextual information. So,"... descriptions can only be 'token-referential', whereas proper names and demonstrative expressions are 'type-referential'". (Recanati 1993: 31).

Discourse, beliefs, and intentions

172

Chapter 4

mental spaces. Mental spaces are linguistic representations, linguistic structures attributed to the speaker and to the hearer, so they will do well for beliefs formed with the use of the linguistic medium.40 Table 1 presents the reference assignment, where 'x' and 'y' are variables standing for two different referents, i.e. x^y. Table 1 (adapted from Jaszczolt 1997a: 320).

I II III

hearer's referent

speaker's referent

hearer's belief report

X

X

de re

y 0

X

de dictoi

X

de dicto proper

Only in (I) was the act of referring performed accurately; the other cases represent certain amounts of lack of knowledge on the part of the reporter. Now, I have promised to simplify the controversial application of the de re/de dicto distinction on the level of expressions of belief to the more standard distinctions that pertain to the expressions with referring properties. First of all, the ambiguity41 of beliefs is directly related to the ambiguity of expressions that are used to refer. Donnellan's (1966) referential and attributive uses of definite descriptions correspond in Fig. 1 to the two nodes giving 1 and 2. The choice seems to depend on the reporter's decision as to what the believer would attribute the predicate to on a particular occasion (cf. also Stalnaker 1970). The reporter assigns one of the two kinds of use of the definite description to the believer: a referential or an attributive use. In the case of proper names as in Situation B, the story is similar: the hearer has to answer the question "To what would [the speaker] on this occasion attribute the predicate 'is '?" (Donnellan 1970: 357). Proper names and definite descriptions are similar in that they both can assume MoPs. This is not the place to develop a comparison between them, they acquire a more detailed 40 Fauconnier's idea of spaces is used here merely as a technical device for representing the mapping between language and reality. Other commitments are not followed. The following quotation best summarizes the idea of spaces: "Language (...) is not merely interpreted with respect to worlds, models, contexts, situations, and so forth. Rather, it is involved in constructions of its own. It builds up mental spaces, relations between them, and relations between elements within them. To the extent that two of us build up similar space configurations from the same linguistic and pragmatic data, we may 'communicate'; communication is a possible corollary of the construction process." Fauconnier (1985: 2). 41 In the sense of options of interpretation.

Section 3

173

The default de re principle

analysis in Chapter 5. Suffice it to say that proper names serve the purpose of referring in two ways, as presented in Table 2. One is to pick out an individual spoken about, and the other to denote whoever holds that name. Speakers can be referentially mistaken and use an incorrect name or description. Now, the reporter also uses names and descriptions in the report. Since the referential use of a description or name may involve mistaken reference, the resulting belief report may be opaque. It is a de dicto report that corresponds to a de re belief, i.e. de dicto7, by force of the definitions accepted above. So, now belief reports exhibit a tri-partite distinction presented in the bottom row of Table 2. De dictoj report corresponds to a de re belief and hence is what we have so far called de re about someone else. The corresponding belief is indeed de re but, as has been argued earlier on in this section, the report is not transparent to substitutions, neither is it transparent to the identification of reference in conversational practice. As a linguistic expression corresponding to a mental state, it is thus better dubbed as a variety of de dicto. Table 2 sums up the distinctions on the level of beliefs, expressions of belief, and reports on belief introduced above. Table 2 (adapted from Jaszczolt 1997'a: 320). Ralph's belief

de re

de dicto

Ralph's expression of belief Ralph's use of the a. description b. name

de re

de dicto

referential (correct or incorrect) referential (correct or incorrect)

attributive pure denotation

John's report on belief

de re, de dictot

de dicto proper

John's use of the description or name exhibits the same characteristics as that of Ralph presented in Table 2. The break marked by dots depicts the lack of direct correspondence between the assignment of a referent by Ralph and by John. John has to infer Ralph's reference assignment and the principles of this inference are laid down in what follows. In other words, the dots will be filled in later on in this chapter. We shall confine de dictoj to the cases where the reporter, the speaker of the attitude report of the form in (3) repeated below, is referentially mistaken. (3)

A believes that B cps.

174

Discourse, beliefs, and intentions

Chapter 4

When the referential mistake is on the part of the holder of the belief and this mistake is recognized by the hearer, an ensuing report does not present any difficulty of interpretation. If the reporter realizes the mistake, he normally uses a correct name or description instead, frequently adding a further hedging clause or other clarification as in (37) above. Both a description and a proper name can be either disambiguated in a situation or, if it is not the case, connected in one way or another with the information provided by the previous discourse. De dicto proper pertains to the lack of individuating information on the part of the reporter. If the lack of knowledge of the referent can be attributed to the holder of the belief and the reporter realizes that this is how the belief is held, he normally adds a clarification to the form in (3), as exemplified in (36) above. But we have to beware that, in practice, if the reporter lacks the individuating knowledge of the referent, he normally also qualifies his use of the expression by, for example, 'someone called a\ 'somebody whom he calls d and other similar indicators. So, this pattern is less reliable in practice than the one pertaining to referential mistakes. The qualification is not always required for conveying the meaning successfully and can be omitted. If it is omitted, then de re and de dicto proper are indistinguishable. If discourse was governed by strict rules, we would be left with the form in (3) as a report exclusively on de re beliefs, by a reporter who has exclusively de re knowledge about the object of the report. But this idealization goes far too far. However, ambiguity of interpretation normally does not ensue. In (3), we can presume the existence of a possible law or discourse principle which would specify one of the readings as the more salient, or even unmarked, default one. On the level of beliefs, intentionality of mental states signals that de re beliefs are the default. As a result, on the level of utterance interpretation, the unmarkedness and priority seems to be attached to the de re interpretation. Example (38) can have all three interpretations presented in Table 2. (38)

Ralph believes that the person who vandalized the phone booth is an under-age criminal.

To repeat, the de dicto/ interpretation that pertains to the referential mistake on the part of the believer is eliminated as it is normally hedged. The de dicto/ interpretation that pertains to the referential mistake on the part of the reporter, on the other hand, poses the problem of interpretation. In theory, it can be analysed as the believer's referential use embedded in a de dicto report, or as the believer's attributive use embedded in a de dicto report. The first also incorporates a referential mistake of the believer's that goes through unnoticed by the reporter and renders miscommunication. So, the options of interpretation are numerous. And yet, sentences of the type

Section 3

The default de re principle

175

(38) are normally understood without a problem, they do not give rise to ambiguities. Depending on the situation, it may have been the case that it was appropriate for the speaker of (38) to substitute the description for the original first name and surname of the child, or he may have retained the original description. Reporters do perform such substitutions in conversation and it is difficult to give any rules as to when they happen. The only reliable rule seems to be the calculation of the communicative value of the expression and the ease of interpretation, or, in more familiar terms, the calculation of costs and effects of the process of interpretation. The standard assumption on the part of the hearer seems to be that the speaker is talking about a known individual. This assumption is signalled by the balance of costs and effects (or Sperber and Wilson's 1986a processing effort and cognitive effect).42 The information value is then the highest and the processing effort is smaller than the one hearers would have to put in if there was an ambiguity of referring, be it semantic or pragmatic. This appeal to intuition is best justified in the following test. Sentence (39) of Chapter 1 repeated here as (40) seems to be a natural answer to sentence (38) of Chapter 1, repeated here as (39): (39)

The best architect designed this church.

(40)

Really? Who was it?

The speaker of (40) takes it for granted that the speaker of (39) knows who he or she is talking about, i.e. is using the description referentially, and that he or she is also able to tell the name of the architect. Similarly, in the case of proper names, the referential reading of (41) and the de re reading of (42) seem intuitively to be more natural, more salient and primary in the order of occurrence to the hearer. Referring to a particular person is clearly intended. (41)

Jimmy is an under-age criminal.

(42)

Ralph believes that Jimmy is an under-age criminal.

This salience of de re is even more obvious than in the case of descriptions which are, after all, normally regarded as not directly referential. They refer directly only in some contexts and with the help of the context. The intuitions concerning the priority, naturalness and unmarkedness of the referential 'Cognitive effect' was called 'contextual effect' in the first, 1986 edition of Relevance.

176

Discourse, beliefs, and intentions

Chapter 4

interpretation of such belief expressions as well as the de re interpretation of belief reports have their theoretical confirmation. This question of defaults is addressed in more detail in Section 4.5. We should take stock. The de re/de dicto distinction on the level of beliefs differs from the tri-partite distinction on the level of belief reports. A belief de re may result in a de dicto report when the reporter is referentially mistaken or lacks individuating knowledge, i.e. forms a belief de dicto. Now, although the holder of the belief must know whether he or she holds a belief de re or de dicto, his or her distinction need not match the classification of the corresponding report as de re or de dicto because either the holder of the belief or the reporter on it can be referentially mistaken. Moreover, the believer's thoughts are not diaphanous to the reporter and they are interpreted by the reporter according to some pragmatic/cognitive rules that govern conversation. I suggested that these rules cannot be very strict but they must rely on the gradation of the status of various interpretations with respect to the default interpretation. If conversation was not governed by priorities and defaults but instead participants relied on context for recovering a semantically underspecified message, interlocutors would spend a vast percentage of their time and effort trying to decide what or who the conversation is about. Alternatively, a vast number of conversations would break down due to referential ambiguities, which, as we know, is not the case. Generally, it is precarious to claim that pragmatic processing either disambiguates or fills in the semantic representation in one way or another, depending on the context. It is frequently the case that the paucity of contextual information makes such disambiguation or 'filling in' unfeasible. So, rather than ambiguity or underspecification, defaults are a better bet.

4.4. Red Giants and White Dwarfs: Context-Dependence of Attitude Ascription It appears that in our account of the interpretation of belief reports in discourse the role of context is defined somewhat negatively. It is defined as the bulk of information that prevents default inferences from going through. 'Prevents' does not mean that the default interpretation is considered by the hearer and then rejected. It means rather that the contextual clues fill in the semantic representation in a direction which is different from the one that would have been assumed had the context not intervened. In other words again, sometimes hearers leap to conclusions as to the interpretation of an utterance: they rely on default reasoning. At other times, however, the context becomes significant and such default reasoning is prevented. I have suggested that the de re interpretation of belief reports is such a default

Section 4

The default de re principle

177

interpretation and this suggestion stems out of the intentionality of mental states and the resulting intentions in communication. But before this default value is further justified in Section 4.5, it is necessary to have a closer look at the inference process that uses this contextual information. For example, in Situation B, how does John infer that Ralph ascribed a wrong name to the individual about whom he held the belief that he was an honest citizen? It seems that Situation A provides the necessary context. In Situation A, Ralph talks about a person who is perceptually salient, who is in the visual field of both interlocutors. Hence, the identification of the individual is easier and more reliable than in Situation B where Ralph's reaction is based on having heard a piece of news from a newspaper article. But it has to be remembered that in Situation B it need not be the case that Ralph is thinking about a wrong individual; he may hold a belief de dicto instead. In fact, it does not really matter so much for reporting whether Ralph held a belief de re about someone else or de dicto in Situation B. In both cases, the report would have to include an enrichment as in (43) and (44). Nota bene, (44) can also be used for the situation where John is not in a position to individuate Ortcutt, so the mapping between the hedges and other explanations on the one hand, and interpretations of belief reports is not bi-unique. (43)

Ralph believes that Smith is an honest citizen but he mistakenly calls him 'Ortcutt'.

(44)

Ralph believes that somebody whom he calls 'Ortcutt' is an honest citizen.

A similar context-dependence affects the interpretation of John's reports as de re or de dicto. Now, how does this context-based inference fare with default interpretations? Our mind processes only what is useful. Proper names would not be processed and remembered if they did not lead to individuals, to acquiring more information about the objects they denote. Their connotative value is minimal (albeit not non-existent, their role is not exhausted in referring) and hence if they did not have default referential properties, they would be useless. As Ryle (1979) observes, there are factors preventing inference such as distance in time or low gossip value of information. Low gossip value, or information value, is associated with talking about whoever undergoes the name of description. High information value pertains to talking about people, animals, or things that can be individuated and shared between discourse participants. Similarly, a long time interval dividing two utterances may prevent inference if there is no incentive to hold the content of the first utterance in memory, and at the same time to make an effort to process the other utterance. This incentive, again, comes with

178

Discourse, beliefs, and intentions

Chapter 4

talking about things. The default aboutness of conversation is in fact another guise of the intentionality of mental acts discussed in Chapter 3. This aboutness is conveyed, which can be summarized as a common-sensical, presumed rule of discourse: "When the speaker refers to a specific individual, by whatever means, he tacitly accepts the convention that he will provide any information (not given in the context) that is necessary for the addressee to identify the individual in question." Lyons (1977: 655). Applied to Situations A and B, this generalization amounts to saying that since reference assignment is context-dependent, the hearer has to make use not only of linguistic means of conveying information, but also of all the available non-linguistic means. The primary one among them seems to be ostension and, as Lyons (1977: 63) suggests, the paralinguistic information (from ostension) takes precedence. So, if, for instance, Ralph points at Smith and says "Ortcutt is a spy", he will be taken to believe that Smith, rather than Ortcutt, is a spy. If in Situation A Ralph makes it obvious that he is talking about Bernard J. Ortcutt by some gestures or even by looking at a person in the perceptual field of both interlocutors, the belief is taken to be de re and about Ortcutt. It is quite possible that Ralph does not know the name of the man in the brown hat but this lack of knowledge is irrelevant and concealed: de re is the default assigned to the belief. Using the name 'Ortcutt' in the report on Ralph's belief may be going too far, but it is unquestionable that Ralph's belief is about Ortcutt. Reporting that he believes that Ortcutt is a spy would mean subscribing to semantic innocence (cf. Barwise and Perry 1983): Ralph may believe that Ortcutt is a spy although he may not know it. This type of parlance has been avoided as a result of the incorporation of MoPs to the semantics of belief reports (in Section 4.2). Just as de re is the presumed default interpretation of beliefs, the referential interpretation is the presumed default assigned by John to Ralph's expression of belief, and de re is the presumed default assigned to John's report on it. In other words, both beliefs, in Situation A and in Situation B, would have been initially tried as de re had they occurred to two different people or had they been reported on by two different people who, in addition, had no contextual clues that would prevent them from interpreting the belief as de re. Similarly, both reports would have been tried as de re. In some languages, there are linguistic means of differentiating between de re and de dicto reports. In Polish, for instance, in the belief reports in the form of negative sentences, the subjunctive is used in the complement clause and it is introduced by a complementizer which incorporates a mood marker that indicates that a subjunctive clause is required. This type of clause is

Section 4

The default de re principle

179

not unique to Polish, it also appears e.g. in French and Spanish. In Polish, it is a marked clause, it is used only when the reporter has some doubt as to the reference assignment to the referring expression used in his or her report. There is also a corresponding unmarked that- clause in Polish that does not undergo similar restrictions in use. When the marked clause is used, the de dicto reading normally ensues. So, the form of the report may suffice for the interpretation as de re or de dicto. This intuition also acquires a theoretical syntactic support in Chapter 8 where it is demonstrated that in the subjunctive clause the constituents of the proposition matter less to the overall meaning than they do in the corresponding indicative that- clause. And if they matter less, then the belief is not de re. The embedded sentence does not undergo substitutivity, it has to be taken as a primitive, i.e. as one unit that corresponds to a state or event rather than to an individual's having certain properties. Or, as it is argued in Chapter 8, the sentence has a higher 'degree of clausiness', the constituents are bonded more strongly than is the case in the corresponding clause in indicative mood. In English, the strength of bonding of constituents of the that-clause is a matter of an interpretative decision of the hearer. There are no syntactic clues that would help in the interpretation. There is a semantic-cognitive clue, however, and this clue comes from defaults based on default intentions. But let us now consider examples (12) and (11) of Chapter 3, repeated here as (45) and (46) respectively: (45)

Red giants become white dwarfs.

(46)

Johnny believes that red giants become white dwarfs.

It can be argued that there is no indication in (46) that the speaker understands what kind of entities he or she is talking about. Indeed, there is no indication that Johnny knows it while uttering (45). Now, in (47), similarly, there is no indication as to how to interpret the relative scope of the indefinite descriptions 'a farmer' and 'a donkey': (47)

If a farmer owns a donkey, he is usually rich.

(from Heim 1990: 150). This example, albeit originally used for a more technical discussion of the semantics of indefinite NPs and anaphoric pronouns, clearly demonstrates the need of having the picture of a situation before attempting to interpret it.43 In sentence (47), the adverb can quantify over the farmer-donkey pairs, over the instances of the first indefinite, or over instances of the second indefinite. When we imagine 99 poor farmers who own See also Lyons (1999: 31-32) on types of anaphoric links.

180

Discourse, beliefs, and intentions

Chapter 4

one donkey each and one farmer with 200 donkeys, the difference becomes obvious: 'usually'can be attached to farmer-donkey pairs or to donkey-owning farmers. It can also be attached to farmer-owned donkeys. One of these readings, namely the second, is intuitively more correct. Heim suggests that the topic-focus distinction helps with the choice of the correct interpretation. The situation has to be constructed before the semantic interpretation is provided. How exactly the interpretation is chosen remains only a vague suggestion concerning the role of stress, topic and focus. What is interesting, however, is that the context has to be construed before the interpretation is attempted.44 This necessity is avoided by Heim by postulating unmarked interpretations, albeit with only tentative theoretical support. The theoretical support for the default de re reading of (46) is easier to obtain, although it is not immediately obvious.'Red giants'and 'white dwarfs' are natural kinds, they are names of types of stars. As natural kinds, they are part of language system. The knowledge of language requires knowing their meaning. However, they are not words of a common, everyday parlance, it is perfectly possible to be a successful member of a linguistic community and yet not be able to use them with some essential degree of understanding. The situation is not even of Fodor's (1994) elms and beeches type. In Fodor's example, although the speaker cannot tell between elms and beeches, he or she knows that they are two different species of trees and that consulting a tree guide or an expert botanist will confirm this and equip him or her with more information. On the other hand, the speaker of (45) need not know this much. He or she may merely, for instance, repeat a piece of information after a teacher at school, and, to use Sperber's (1975,1985a, 1996) explanation, store it in memory in some 'semi-propositional form', so to speak, 'put in quotes', for future understanding (cf. Section 2.9 above). Yet it would be rather implausible to insist that it is a normal, unmarked move in conversation to put propositions 'in quotes', to store them in a semi-understood or totally undigested way, hoping that some time in the future they will become clear. It is certainly not the default scenario of utterance interpretation. Instead, it seems to be the last resort open to hearers who have no source of information available to resolve the meaning problem immediately when it occurs. So, it can be tentatively proposed that although (46) can be taken 44

Studies of implicature show how the referent can be specified by the context. About example (i), Hobbs (1987: 2) says that in order to find the referent of 'the engine', the interlocutor has to rely on contextual information and mutual knowledge that cars have engines. (i) I bought a car last week. The engine is already giving me trouble. This information is necessary to provide a referent for 'the engine'. Without such an implicature being available, reference would fail. Sometimes it is the cultural context that makes the referent accessible. According to Givon (1992), the 'cultural context' is the world view shared by members of a culture or a speech community. The referent can also be accessed from the shared cultural and contextual information. See Section 5.6.1 below.

Section 4

The default de re principle

181

as a report de re or de dicto, and at the same time a report on the de re or de dicto belief, the de re interpretation is much more likely to be the case, and possibly is the default one. Intentions in communication provide the strongest support for this status of de re. All in all, it can be suggested that, although the contribution of context is limited to the departures from the default reading, its relevance goes far beyond that because context includes intentions. Since intentions are one of the aspects of context of utterance and intentions account for the default reading as well as the departures from the default, this aspect of context performs a crucial role in the interpretation of belief reports. Finally, I would like to assess briefly some arguments for and against an extensive role of context in attitude reports that have appeared in some recent accounts. I shall confine the assessment to two suggestions: Bezuidenhout's heavy context-dependence and Devitt's mild context-dependence. In her recent work, Bezuidenhout (1997) uses Bach' s idea of semantic underdetermination of expressions to argue that belief reports require pragmatic enrichment of the meaning of the embedded sentence. This enrichment allows for the speaker's MoP to be recovered. Enrichment, and a fortiori context-dependence, are said to be a common, general feature of utterance interpretation. She resorts for this purpose to Sperber and Wilson's (1986a) account of the semantic content of utterances called explicature. Heavy context-dependence of interpretation is possible because enrichment is non-modular and can draw on other beliefs and knowledge. Now, while Bezuidenhout is correct about the access to background information and the need for some enrichment, at the same time she assigns too great a role to context. It has to be remembered that not all belief ascriptions require the MoP of the referent to be known and that sometimes it is not possible to pinpoint a particular MoP. In fact, the assumption that MoP is not relevant for truth conditions is the default idealization that governs discourse participants. If there is no evidence to the contrary, the hearer believes that the speaker knows enough about the object of his or her belief to be able to identify the object comfortably in various situations of discourse. Hence, Bezuidenhout's enrichment is too strong a condition. Types of MoPs graded from MoPs 'proper' to the value zero are more truthful to discourse practices. Bezuidenhout's heavy context-dependence of attitude reports is a reaction to Devitt's (1996) mild context-dependence, as he himself calls it (ibid.: 197). Devitt favours Quine's ambiguity instead. Sentences of the type (3) are established by the context, and notably by the speaker's intention, as either transparent or opaque. He argues that the context-dependence of that-clauses is limited by the literal meaning of the sentence:

182

Discourse, beliefs, and intentions

Chapter 4

"A t[hat]-clause can only ascribe a meaning that is closely related to one of the meanings of its content sentence; there is an 'intimate link' between the two meanings. ... [T]he link is often the closest Devitt (1996: 197-198). relation of all: identity." Just as Bezuidenhout's heavy contextualism was too strong, so Devitt's mild contextualism is too restrictive. About sentence (48), he says the following: (48)

It is raining. "... in understanding the utterance, the hearer must supplement it, guided by what is obvious in the context." (ibid.: 199).

And what is obvious is the location, e.g. 'in New York'. Further, "The morals to be drawn from this brief reminder are first, the meaning of a token utterance (normally) depends heavily on the conventional meaning of its type, and second, insofar as the hearer's understanding must be guided by some aspect of the context, that aspect is usually accessible to the hearer." (ibid.: 199). Devitt concludes that the more the meaning of a belief report departs from the meaning of the f/za 7-clause, the more difficult the interpretation. He also concludes that the more different meanings the f/ia/-clause has, the more difficult the interpretation. It is so presumably because the interpretation becomes heavily context-dependent. It is true that de re readings do not require much contextual enrichment of the meaning of the referring term, unless the term is itself an indexical or is otherwise context-dependent. De re relies on 'jumping to conclusions', it is the default interpretation. However, these default interpretations are not the literal meaning, the meaning of the sentence. Instead, all of the interpretations, de re, de diciol and de dicto proper, are semantic representations because the context, in the form of intentions in communication, contributes to the semantics of attitude reports. Hence the gradation of context-dependence is not a departure from the sentence meaning as Devitt has it, but rather a departure from the default sentence meaning. The difference is vast. It is a difference of the level of analysis. The claim that the more meanings the sentence has, the more difficult the interpretation is rather contentious. There is extensive evidence that utterance interpretation is incremental and proceeds, so to speak, 'bit by bit', using contextual clues in the process, rather than relying on the choice of an interpretation from a pool of possible readings of a sentence. Devitt is tied

Section 5

The default de re principle

183

to the latter because he advocates the ambiguity of belief reports. He looks into context for their disambiguation. This stance is methodologically unparsimonious, as was demonstrated in an extensive discussion in Chapters 1 and 2 (see also Bezuidenhout's 1997 comments). Devitt has problems of his own with his account. Namely, ambiguity does not yield easily to any syntactic or lexical explanation. So, the context has a role in disambiguating a sentence that is not even clearly 'unambiguously' ambiguous.45 To evoke Kripke's (1977:19) criticism of Donnellan, "[i]t is very much the lazy man's approach in philosophy to posit ambiguities when in trouble."And it is very much so when it involves adding new types of ambiguity that do not even yield clear definitions. There is an advantage of Devitt's account that is not negligible, though. He allows for MoPs to take the value zero, which is still shunned by the hidden-indexical theory (although to a lesser and lesser extent, see Devitt 1996: 202, fn 63). This move would, however, be better exploited if it was paired with an account that does not confine the role of context to resolving putative ambiguities but rather assigns to it a more positive role of pointing to the suitable interpretation. Default Semantics offers this improvement. All in all, the adequate account of belief reports seems to lie closer to the hidden-indexical theory than Devitt's 'sort-of-ambiguity' view. However, the hidden-indexical theory has to strengthen the account of the 'value zero' of MoP, or, in other words, of the cases where MoP does not matter to semantics. After all, these cases turn out to be the default, as is argued in more detail in the following section. Devitt's mistake is to think that the role of context is to lead the hearer to the relevant MoP (ibid.: 205). Then he encounters the problem because context cannot do that. The context need not fulfil this role. Context performs the semantic task of enrichment when it is required. But this is not required by default, unless the expression contains other context-dependent elements or features such as indexicals, ellipsis or syntactic or lexical ambiguity. It is part of the definition of the de re interpretation that MoPs are mutually known and need not be specified.

4.5. Default De Re The default status of the de re reading of belief reports is not a novel idea. For example, Klein (1981) talks about its unmarkedness in interpretation, Burge (1977) and Bach (1982) stress the fundamental role of de re beliefs in the 45

Cf.: "Is the ambiguity syntactic or lexical? (...)! do not know." Devitt (1996: 200, fn 61). Devitt resorts to assigning an ambiguity of a 'special sort' to attitude reports, which is, clearly, an escape rather than a solution, dictated by a conviction that ambiguity is definitely there.

184

Discourse, beliefs, and intentions

Chapter 4

acquisition of knowledge about the world. Two things have to be discussed here. Firstly, the claim that the de re reading is indeed the default requires stronger theoretical support, and secondly, the extent of the contribution of pragmatic processes to the semantics of belief reports has to be assessed. Solutions to both problems will be shown to rely on intentions in communication. In discourse, we normally regard constructions of the type (3) as expressing something about B. People normally report on people's beliefs as if Leibniz's Law held, unless they have evidence that the holder of the belief was referentially mistaken or otherwise lacking in essential knowledge about the individual. Substitutivity salva veritate is a common fact of conversation and as such is uninteresting. What is interesting is how it happens that the reporter ascribes so much relevant knowledge to the holder of the belief, so to speak, by default. It is also interesting what mechanisms underlie the decision to depart from this default situation and how exactly they function. It was suggested in Chapter 2 that there are three types of intentions that accompany people's utterances: the referential intention which is an intention to secure the referent of an utterance; the informative intention which is an intention to convey a certain set of assumptions to the hearer or make the hearer accept a certain set of assumptions previously held tentatively;46 and the communicative intention which is an intention to make it mutually obvious to the speaker and the hearer that the speaker has this informative intention. In Chapter 3, it was suggested that the intentionality of mental acts makes these acts, and a fortiori linguistic expressions, about certain objects, and normally about real objects, where applicable. This property of mental acts is reflected in the referential intention which accompanies utterances. So, the referential intention makes the referential interpretation of potentially referring terms the norm. There are numerous partial arguments in favour of the referential interpretation of definite descriptions and the de re interpretation of belief expressions as the unmarked, more salient, or default one, i.e. the one that comes first to the hearer's mind.47 Now, it cannot be denied that on occasions like the Situation B following the Situation A, the hearer can be faced with potentially more than one reading of the expression of belief. There is an interpretative ambiguity there, perhaps better dubbed a multiple choice of interpretation, although the existence of this choice does not exclude the possibility that one of the readings may be privileged as a more salient or default one. Let us retrace the arguments for default interpretations of Chapter 2 in the example of Bach's 46

In Sperber and Wilson's (1986a) terminology, this is an intention to make manifest or more manifest to the hearer a certain set of assumptions. 47 Cf. also Salmon (1986: 121) on what we called de dictoj.

Section 5

The default de re principle

185

claim of underspecification. Sentence (49) is standardly regarded as interpretatively ambiguous between the two readings: the transparent one, paraphrasable as in (49a), and the opaque one, paraphrasable as (49b): (49)

Mary believes that the present king of Sweden is bald.

(49a)

Mary believes of the present king of Sweden that he is bald.

(49b)

Mary believes that there is a king of Sweden and that he is bald.

Bach (1987a: 206-207) argues that the distinction holds only for sentence use rather than for the semantic interpretation. He argues that although sentences of the type exemplified here by (49) can be assigned two representations in first-order logic that differ in the scope of the existential quantifier applied to 'the king of Sweden' (ranging over the belief predicate and being within the scope of the belief predicate respectively), this fact does not suffice to infer that this sentence has these two logical forms and, a fortiori, that it is ambiguous. I am inclined to agree with the latter claim but I have also argued that this consent does not yet mean that semantic representation has to be underspecified as to the particular meaning and as to truth conditions. Before reaching the verdict on the semantic representation of such interpretatively ambiguous sentences, it is necessary to investigate whether there are any general operational principles that would render one interpretation as primary, unmarked, that would under normal conditions function as the standard semantic representation of the sentence. When an expression has two readings, such as a transparent and an opaque one in the case of (49), caused by two different occurrences of a definite description, the explanation of these two readings is not provided by the logical form, i.e. by the scope of the existential quantifier. The narrow scope reading can still be either transparent or opaque, depending on the particular occasion of its use. And, if there is no such logical explanation of the ambiguity, there is no semantic ambiguity either. Now, those who exorcize the ambiguity view take a big leap from there to the claim that the semantic representation of (49) has to be indeterminate, underspecified, unfit to be assigned truth conditions (see Bach 1987a: 206-212). It is also sometimes said that only the uses of an underdetermined sentence have truth conditions, and thus the study of reference belongs to pragmatics. My standpoint represents a middle ground between the ambiguity thesis and the indeterminacy of sense. To repeat the conclusions from Chapter 2, in my view a sentence of the type (49) has a unique semantic representation that corresponds to the standard, default reading of the sentence and is achieved with the help of intentions that 'intrude', so to speak, in the semantic representation. This representation

186

Discourse, beliefs, and intentions

Chapter 4

can be assigned truth conditions and is fully operational on the level of sentence use in some (default) cases. Other cases can be accounted for by the non-standard workings of intentions in communication. So much for the reminder. Now, in interpreting an utterance, the hearer is not concerned with the question as to whether somebody is being referred to by a referring term. After all, in the phenomenological explanation, objects invoke acts and give rise to ideas. This much is taken for granted. Rather, the hearer is interested in who this person is. Even this question is not normally asked. The speaker relies on the knowledge of language and the world in his or her use of names and descriptions, as well as on the presumption that this knowledge is common or shared. Expressions are said by Husserl (1900-1901) to be applicable to objects as their 'clothing'. Knowing the object depends on the fulfilment of the intentions in thought, perception, or imagination. They constitute acts of 'meaning formation' and perception is the strongest among them, it gives the strongest fulfilment of the meaningintention, or, in other words, the most reliable knowledge. It also leads to beliefs with the highest degree of objectivity, beliefs de re and hence also belief expressions with referential significance. The corresponding utterances about beliefs are thus de re. So, meaning is provided at its best in perception; perception gives the best warranty for a belief. When we connect this claim with the one of Ideen (Husserl 1913) that intentionality of experience makes it a consciousness of the real world (as a special case, so to speak, of possible worlds), then the priority of referring to a particular object becomes a default. However, it has to be noted that to recognize this role of perception is not the same as to claim that acts that are fulfilled in perception are the only sources of beliefs de re. Let us move on. Intentionality is said to be a social phenomenon. Meaning is not private, and yet it pertains to mental representations. It is so because language gives thoughts some objectivity, without precluding idiosyncrasies, referential mistakes, incorrect use of words, and the like. So, we can say that intentionality is social and individual at the same time. These features go neck and neck in the majority of cases but referential mistakes ocasionally happen and are a problem. Fig. 2 summarizes this case. The diagram juxtaposes the physical world and the mind in a way loosely modelled on Fauconnier's (1985) mental spaces. Language is a vehicle of thought that builds its own constructions called mental spaces. The opaque and transparent readings of intensional contexts are explained by the ability of language to construct different spaces, where objects are linked by 'pragmatic functions' (see ibid.: 4, 22). In other words, a name that refers to an element of one space, the world, can also refer to an element of the other space, the mind. The element of the mental is linked with its real counterpart by a connector. But a name can also set up an element in the mental if no corresponding element exists in the world. However, Fauconnier's approach does not explain the

Section 5

187

The default de re principle

ontological status of his constructs. Hence, instead of his mental spaces and a connector, I shall use the idea of intentionality which relates mental representations to the world, in accordance with the phenomenological tradition. In my diagrams, the ovals depict (i) the totality of the mental representations of the holder of the belief and (ii) the world, the common-sense reality of everyday experience. The connection between them has a clear epistemological status of intentionality of mental acts or corresponding states, and, in my account, it can be weaker or stronger. A mental act is about the word in, so to speak, two ways. For example, referring by the name 'Ortcutt' reaches Ortcutt because this is the correct individual that undergoes the name and this fact is mutually known to the interlocutors. It is caused by the initial act of naming the person as Ortcutt, spreading of this information in the community, etc. On the other hand, it is possible that the speaker can be referentially mistaken and then the belief is inherently about someone else, e.g. Smith. Hence, two'paths'of intentionality have to be recognized: the social and the individual one. As a result of the discrepancy, the strength of intentionality as a semantic, meaning-giving tool is dispersed between 'intending Ortcutt' and 'intending Smith'.48

the world49

the mind Fig. 2 (from Jaszczolt 1997a: 324).

In Fig. 2, the world as a correlate of the mental contains objects that are intersubjective, social, in that they correspond to other people's objects of intentions. Yet they need not correspond to them in a straightforward way of associating names and descriptions with objects, as the dispersion demonstrates. Now, Husserl's idea of the horizon (see Section 3.5) secures 48 49

I owe this idea to a discussion with Barry Smith. This is the common-sense world rather than the physical reality per se. See Smith 1994.

Discourse, beliefs, and intentions

188

Chapter 4

the possibility of acquiring more knowledge about the object and filling the gaps or changing the set of properties assigned to the object. As a result, intentionality that has been weakened in its semantic function by the dispersion, can become stronger and stronger as the distance between the social and the individual paths gradually disappears. The strongest intentionality is the case of referring de re where the individual and the social correspond, so to speak, to the same path. Fig. 2 corresponds to the belief de re about someone else. If the dispersion pertains to the reporter's mental states, then Fig. 2 corresponds to the belief report de dictoj. The weakest case of intentionality is the belief de dicto where intentionality is dispersed to the extent that it does not reach an object, as in Fig. 3.

the world

the mind Fig. 3 (from Jaszczolt 1997a: 326).

This dispersion of intentionality has further consequences for MoPs. It is often said that mental states relate to an object via an aspect, or we can say, a MoP (cf. Searle 1983: 52). But it is not always necessary for the hearer to know the aspect under which the speaker intends an object in his or her expression of belief. On the other hand, the aspect alone may not suffice for individuating the intended object. Unlike Fregean senses, aspects, like MoPs, should not be regarded as bound to individuals because otherwise they cannot account for referential mistakes. In other words, being recognized as Ortcutt has to, somehow or other, lead the reporter to recognizing that the belief can be about Smith when the holder of the belief is referentially mistaken. So, I suggested that when intentionality is the strongest and the mental state is uncontroversially about a certain intersubjectively recognizable object, the MoP need not be recovered by the hearer because this

Section 5

The default de re principle

189

recognition does not contribute anything to the semantics of the sentence. When the intentionality is dispersed between the wrongly intended individual and the individual properly picked out by the term, then the type of MoP, notably the information that there is a referential mistake involved and a 'pointer' at the intended individual is required. De dicto requires a more finely-grained type of MoP or even full MoPs. Intentionality viewed in this way performs another useful function. Frege says that there is no road back from the reference to the sense of an expression. This claim is too strong. In the case of referential mistakes, the mistaken referring term is itself a road back, if not to sense, then at least to the type of MoP. Frege posited objective senses and hence he could not see the road back: free MoPs were beyond his reach. Moreover, since utterances come with corresponding mental states to which the hearer can have some degree of access through the access to various vehicles of thought, then individuating the referent need not mean rejecting the accompanying MoPs just as we are supposed to reject senses, like a ladder, when we reach the referent (in that sense replaces the reference where required). This is psychology. But there is in fact no such rejection in semantic theory either. As the theory of Default Semantics has it, MoPs and types of MoPs are relevant for the semantics just as intentions are. They lead to the appropriate semantic representation for the sentence, the utterance of which is the starting point for the semantic analysis. One important caveat is needed at this point. To approach the puzzles about prepositional attitude contexts by looking at sentences and their putative ambiguities is putting the cart before the horse. There are conversations and utterances in the world, just as there are people, animals and things. The rest is the road from the physical to the mental and should remain as such. Hence, we should not shun the talk about utterances when looking for a successful semantic theory. Utterances have their semantics. Ultimately, the semantic representation of the belief report is assigned thanks to pragmatic information, be it the default intentions or contextual clues pointing to other choices, namely de dictoj or de dicto proper. The sentence uttered by the speaker does not provide sufficient information concerning reference assignment. The semantic representation, at the stage of LF, requires a further determination coming from the speaker's referential intention, embedded in the informative and communicative intentions. Neither does the LF provide any indication concerning the correctness of the name or description used by the speaker. All in all, the truth conditions of the expressed proposition can only be stated had the LF been supplemented with a pragmatic resolution as to whether the referring was performed or not, and if so, whether it was performed correctly. Various degrees of pragmatic contribution to the semantic representation are required for this, depending on the degree of the departure of the interpretation from the default de

190

Discourse, beliefs, and intentions

Chapter 4

re. In other words, intentions interact with the semantic representation, they become a constitutive part of this cognitive representation and allow for the crucial 'disambiguating' (or: completing, fulfilling) information to be read off this complete cognitive, semantic representation. They also secure certain interpretations as unmarked and easily accessible to the hearer. This is, essentially, how intentions 'interweave' with the LF and how utterances matter for the semantics. I have begun this section by repeating the well acknowledged claim that belief states are intentional and are normally directed towards objects in the world. If it is so and if intentionality is a necessary precondition to call a state a belief, then de re beliefs are the default, natural ones. De re reports are also conversationally more salient and unmarked. Other types of beliefs and reports are, so to speak, 'faulty' as far as the intentionality that accompanies them is concerned. De dicto beliefs are not properly about an object; de dictoj reports pertain to the intentionality that is dispersed between the intended object and the intentional object assumed by the hearer, and the report de dicto proper is, again, not about any particular, individuated object. The 'intention of intentionality' is satisfied there in that the state 'sets off' as intentional but for some reason or other which pertains to the background knowledge of the speaker, intentionality becomes dispersed and does not reach any individual. The object was only weakly, 'insufficiently' intended. In other words, intentionality is a matter of degree and although every belief state originates as intentional, not all of them turn out to be that way. De dicto reports are the extreme case of such a failure of intentionality as a semantic tool. So, it can be concluded that the intentionality of mental states provides an indelible argument for the default de re readings of belief reports. Let us call this conclusion the Default De Re principle. The Default De Re principle: The de re reading of sentences ascribing beliefs is the default reading. Other readings constitute degrees of departure from the default, arranged on the scale of the strength of intentionality of the corresponding mental state. The principle can be expanded as follows: (i) The hearer of an expression of belief of the form 'B . The de re and de dicto readings can be easily given different representations now, using lambda-abstraction. Sentence (5) read as de re can be formalized as in (5a) that corresponds to (5b): (5)

John believes that Miss America is bald.

(5a)

Xx [Bel (j, A[B(x)])] (m)

(5b)

The individual denoted by m is such that John believes the proposition that that individual is bald.

(from Dowty et al. 1981: 166). Here m does not stand in the oblique-contextcreating operator 'A'. But this is only a formalization of the distinction. Apart from providing a formal representation in intensional logic, Montague does not discuss these two readings. He assumes that it is an ambiguity and that it belongs to language, and he gives it a semantics.18 As I have argued throughout this book, in view of contemporary dynamic developments of Montague semantics, postulating ambiguity is not an adequate solution. The interpretation of an utterance of an attitude sentence uses information from the context and this context may contribute to the semantics of the sentence. The semantics has to follow the process of arriving at one interpretation by the hearer. And, as I have proposed in the Default Semantics, there is a default reading, the default semantics of attitude reports. In addition, other vehicles of thought help arrive at 'non-default' interpretations. Hence, looking at one vehicle in isolation is unproductive. Dynamic semantics better accounts for this new role of 18

For other accounts see Montague 1970a, 1970b and Montague and Kalish 1959. In the latter the authors propose the so-called 'naming functions', i.e. replacing the occurrences of 'that + noun' with 'that + NO', 'that + N,', etc.

Section 3

Vehicles of thought in attitude

255

context, as is demonstrated in Chapter 7, although by itself it does not eradicate ambiguities. Even there we need help from intentions and intentionality.19 So, it turns out that in semantic theories that make use of possible worlds, the de re/de dicto ambiguity is assumed and is located in language. Attitude reports can have two formal representations. But which representation is chosen by the hearer, and, first of all, rules that govern the hearer in this choice, are left out of the theory. Now, if we incorporate the study of context, including other vehicles of thought and intentions in communication, then postulating ambiguities is no longer needed. The linguistic vehicle contains the possibilities of utterance interpretation that may have to be supplied by contextual information. Since intentions belong to the semantics, they belong with this vehicle. And if they belong to this vehicle, then language does not 'contain' ambiguities but rather leads to one default interpretation of the sentence, guaranteed by the referential intention. This default is the only interpretation that relies solely on the linguistic vehicle. Partee's (1979: 10) suggestion that the psychological representation may be relevant for the semantics of prepositional attitudes appears to be a common necessity in order to go beyond the default, to context-driven interpretations. If there was indeed the de re/de dicto ambiguity of attitude reports, the reporter would have to know whether the holder of the belief would consent to the report. And this amounts to knowing whether the believer possesses the same, or sufficiently similar, information about the object of belief as the reporter. Instead, the hearer does not normally worry about such similarity and reports on the belief as if it was de re, unless he or she is prevented by ensuing contradictions. The interpretation of animals' and children's beliefs gives further evidence of the priority of de re. There have been attempts to go beyond the sentence in possible worlds semantics and ask whether the proposition expressed depends solely on the sentence. Situations proved to be a useful theoretical construct. A situation comprises the time and place of utterance and the identity of the person uttering it. In situation semantics (Barwise and Perry 1983; Perry 1986), situations, rather than worlds, are taken as primitives. Situations are abstracted from and generalized over real situations that are parts of the world (cf. Barwise and Perry 1983: 8). Propositions are true relative to situations. The meaning of a simple declarative sentence is said to be a relation between utterances and situations. The context-dependence allows for logically equivalent propositions to differ semantically. Now, Perry (in Barwise and 19 There are various dynamic-semantic developments of the Montagovian approach that will not be reviewed here, except for DRTused in Chapter 7. An assessment of the present developments in dynamic semantics is not the purpose of this work and the choice of DRT has to be accepted either on trust or as arbitrary.

256

Discourse, beliefs, and intentions

Chapter 6

Perry 1985: 149) says that holding a belief means being in a certain frame of mind. Concepts and ideas are in some way 'anchored' to the world. 'Event-types' are 'conditions on anchors'. A believer realizes that he or she believes a certain 'type of event'. These event-types are composed of ideas of things. But, as in other discussed views, there is a gap here between the linguistic and the mental that is left unexplained. In order to arrive at an interpretation de dictoi or de dicto proper, the hearer may need more than Barwise and Perry's situations with their individuals, properties, relations and location as building blocks. If these were to suffice, how would the hearer draw a link between these objects and the 'degree of referring' intended by the speaker? This degree of referring, the MoPs may have to be recovered from the situation in order to report it correctly as de dicto/ or de dicto proper. We need more finely-grained situations that, in addition, pertain not only to language and the world, but also the speaker's representations, links between utterances in discourse, and the information from non-linguistic activities.20 And this brings us back to Husserl's noema which is an extension of sense beyond the linguistic vehicle, and, also, to dynamic semantics that accounts for the incremental nature of discourse, to be discussed in Chapter 7.

6.3.3. Concepts and the First Vehicle

There are more concepts than corresponding words. Some concepts are not lexicalized and it is easy to observe this fact in everyday speech. Sometimes we have to use a several-word long description to convey a concept. In other words, a person can have a concept without having a corresponding word. There are two explanations of this. Either there is a corresponding 'word' in the language of thought, or the concept is 'carried' by a vehicle other than language. The second possibility seems more plausible since otherwise the language of thought would have to be enormously rich and include 'words' for all possible concepts in all possible cultures - unless we relativize this inner language to the corresponding culture or natural language and, at the same time, make it extend beyond this natural language. But there are no strong theoretical arguments or empirical evidence for this hypothesis. Also, prior to such a relativization, we have to answer the question as to whether concepts are universal, that is whether they can be found in all 20

Situations are rather unfortunately tied up with semantic innocence, the claim that by using referring expressions speakers always refer to individuals rather than to senses or meanings, and hence substitutions salva veritate are admissible (cf. Barwise and Perry 1983: 176). Referring expressions comprise here names, pronouns, and referential and some attributive uses of definite descriptions - the so-called 'outer attributive' uses that do not describe how the believer thinks about the object (cf. ibid.: 201).

Section 3

Vehicles of thought in attitude

257

cultures. Or perhaps there is a subset of universal concepts. For example, there is no single word for camel in Arabic, no single word for snow in Eskimo, and no single word for sand in some Australian aboriginal languages. Instead, there are different words for different kinds of camels, snow and sand respectively. In Russian, there are two words corresponding to 'blue'. The different ways in which languages reflect reality are important for attitude ascription for the following two reasons: (i) the belief is dependent on the linguistic vehicle because it is carried in a natural language with its specific way of 'cutting reality up' into concepts, and (ii) the belief is not fully externalized by language because there are concepts that are not lexicalized. Even if thinking uses a system different from natural language and (i) and (ii) do not apply, there are other problems with cross-linguistic differences in conceptualization. Beliefs are expressed in natural language and natural language is also the vehicle of the second externalization, namely of conveying the belief to another person. Problems in (i) and (ii) will thus apply to the reconstruction of someone's belief, if not to belief formation itself. The postulate in (i) reflects the idea of linguistic relativity of Edward Sapir and Benjamin Lee Whorf according to which languages divide up reality in different ways. Language is, so to speak, a mirror of nature:21 "... no individual is free to describe nature with absolute impartiality but is constrained to certain modes of interpretation even while he thinks himself most free. The person most nearly free in such respects would be a linguist familiar with very many widely different linguistic systems. As yet no linguist is in any such position." Whorf (1956: 214).22 Sapir and Whorf suggested that our perception of reality is shaped by the language we speak. In the strong version of the hypothesis, language determines our thought because we can think only in the categories available in this language. Relativity is not restricted to nouns. Systems of tense and aspect also vary between languages and reflect the perception of the world by their speakers. However, there is an ongoing debate concerning relativity. Concepts may not be wholly arbitrary, some may even be universal. Brent 21

The idea of linguistic relativity was present in the writings of earlier scholars such as Wilhelm von Humboldt and Franz Boas. 22 Cf. also quotation in Chapter 4, fn 21.

258

Discourse, beliefs, and intentions

Chapter 6

Berlin and Paul Kay (1969) largely undermined the relativity of concepts working in the field of colour terms. They demonstrated that conceptualzation is not totally arbitrary. They worked on 98 languages, using standardized colour samples differentiated according to hue, brightness and saturation. Their 329 colour chips represented 40 hues in 8 levels of brightness (320 chips), as well as neutral hue of white, black, and greys (9 chips). Berlin, Kay and their co-workers asked speakers of 20 languages to provide a simple, basic colour term in their language and, next, to point to the chips that corresponded to a particular colour term, including the most focal representatives for a particular colour word. The result was that focal colours were very similar across languages. So, categorization of colours revolves round focal colours whose number but not quality differs from language to language. In the early 1970s, Eleanor Rosch also experimented with focal colours and she found that colours are perceptually more salient than non-focal colours. The experiment was then extended to shapes, where foci or prototypes were also found. Berlin and Kay demonstrated that there is a regularity to these interlingual differences. If the language has two colour terms, like some languages of Papua New Guinea, then they are equivalents of black and white (or light and dark); if it has three, then black, white and red, and so on up to eleven basic colour terms. Such regularity constituted a good argument against relativity in the late sixties. Since then, the appeal of the Sapir and Whorf hypothesis has faded even more, especially as a result of the development of cognitive science. According to cognitivists, the structure and operations of the brain determine the structure of the language, i.e. its grammatical and semantic structure (cf. J. Lyons 1995: 97). Paul Kay (1996) also continued his battle for universalism, claiming that having semantically different expressions for a phenomenon or object in different languages does not necessarily mean different concepts, and that even in one language speakers have access to various ways of conceptualizing things: "In pursuing a deeper understanding of linguistic relativity, we might do well to remain aware, as every translator must be, of the large amount of alternative schematizing of a given event there is available within each language and hence within each speaker of that language." Kay (1996: 111). Now, if the mind determines language, it is only one step from there to claim that not all mental processes need language and that there possibly is a special language (or computational system) that people use for thinking. This system can be Fodor's (1975) language of thought (or Mentalese). One of the arguments for the language of thought comes from the fact that natural languages are ambiguous and vague; quite often, what is said involves not

Section 3

Vehicles of thought in attitude

259

only the meaning of the sentence but also some additional implied information. Yet it seems that this universal computational system can be reconciled with relativity. There is a middle solution in allowing computations to be universal, while also allowing for a different fineness of grain of conceptualizations for different systems. What is lexicalized or grammaticalized in one language, can be accepted as salient in another that does not have this lexeme or grammatical distinction. I shall not go into this proposal here. Chapter 8 shows how it works for belief reports de re and de dicto in Polish and English. The result is a form of universalism, provided we include default meanings on a par with lexicalizations. Universalism is also represented by linguists who base their claims on physical, empirical evidence. Anna Wierzbicka looked at many different languages for support to her thesis that there are conceptual primitives, i.e. primitive, simple concepts that can be found lexicalized in all languages. So, there is one, universal set of basic concepts, semantic primitives, realized in all languages. These primes are innate. Apart from these, there are also many culture-specific concepts and words. They can be defined in terms of the semantic primes, whose number is estimated at about fifty (see Wierzbicka 1996, 1997). These universal conceptual primitives combine to form conceptual structures, also universal. On the whole, we have a language-like innate conceptual system. So, which standpoint is correct, cognitivism with its universal concepts or relativism of concepts? There have been serious objections to Berlin and Kay's early work, both methodological and factual. Russian has twelve basic colour terms, with two words for blue; kinship terms are difficult to place on any universal scale; and, first of all, the method was under attack. Using colour samples may defeat the object of any generalization by imposing a way of classifying colours which is alien to the analysed culture. For example, Hanunoo, a language of the Philippines, has four colour terms translated as white, black, green and red, but they really mean lightness, darkness, wetness, and dryness. Zuni, of the American Southwest, has two terms for yellow - one for becoming yellow through ripening or ageing and the other for things coloured yellow or yellow substances. The first one is a verb, the other an adjective (cf. Lucy 1996: 46). The conclusion seems to be that languages differ in their semantic structure, but there is no evidence that this semantic structure determines thinking, perception, cognition, or memory: "The idea that language could determine (however weakly) the nature of our thinking nowadays carries more than a faint whiff of anachronism; rather it seems to belong to an altogether different age, prior to the serious study of mind as an information processing device." Gumperz and Levinson (1996: 22).

260

Discourse, beliefs, and intentions

Chapter 6

The properties of this device are genetically encoded. According to Fbdor's (1983) influential hypothesis, the mind is composed out of modules, specialized to process visual, auditory and other information. The output of this processing is submitted to the central processing system where thinking takes place. The modules are now understood to be in fact clusters of modules, that is specialized input-output devices (cf. Gumperz and Levinson 1996: 22). On the other hand, recent work by Lucy (1992, 1996) has shown large differences between the ways speakers of different languages classify objects - differences that are said to affect their cognition. Moreover, it has to be asked at this point whether it is language that affects conceptualization or, rather, there is merely a correlation between linguistic and non-linguistic conceptualization. On the basis of their cross-cultural study of spatial conceptualization, Pederson et al. (1998) concluded that the variation in observed languages corresponds to conceptual variation. For example, if the language renders space as north and south, the conceptual, non-linguistic equivalent will also be absolute; if a language uses projections from the human body such as 'in front of me' or 'to my right', the conceptual, non-linguistic equivalent in that culture will also be relative. Linguistic and non-linguistic behaviour match each other, but drawing any conclusions beyond the correlation would be unfounded. The direction of causation has not been established. Differences in conceptualizations signal that various parts of reality are not reflected in languages. As a result, they cannot be easily conveyed in an expression of belief. These missing distinctions can be present in mental images or 'motivations' for actions in the first externalization and, equally, they can be expressed by actions and by making mental images salient in a certain context in the second externalization. Now, since the language of thought is still only a theoretical construct of an unsupported (albeit influential) hypothesis, we can amend it somewhat to make it more amenable to testing. Construing the language of thought narrowly or widely would look as follows. Language, as an abstract object, existing independently of the fact that it is used, can be defined as a mapping of things and sequences of things onto meaning.23 Normally it means a function from sentences to propositions but there are two options for extending it. Firstly, for the language of thought, 'sequences of things'can incorporate images and actions. Let us surmise that the language of thought is in fact a natural language such as English, Polish

23

Cf. Schiffer, e.g. 1993, on a compositional supervenience theory that replaces compositional semantics. Supervenience means a correlation between mental representations and physicalistic states of affairs. I accept both supervenience and compositional semantics. See Section 6.4.

Section3

Vehicles of thought in attitude

261

or Akkadian, and other carriers of thought that act on a par with it. Together they undergo conceptual universalism. There are differences between ways of expressing thoughts when we take vehicles in separation, but all together amount to universal patterns of cognition, thought, and expression of thought. This is the wide construction. In the narrow construction accepted above, the language of thought is a function from inner sentences to propositions, but there are two other vehicles where a function 'from things to meanings' also applies. Language corresponds to the language of thought (see Schiffer 1993 on this correspondence), while actions and mental images are separate vehicles. So, there is a wide and a narrow construal and I have chosen the narrow one. Not much hinges on this theoretical decision. After all, the language of thought, or Mentalese, is a metaphor for a neural state of belief. This neural state has parts and relations between them, just as natural language does. The structure formed by these parts represents a certain proposition: "Talk of a 'language of thought' is simply a useful metaphor for an inevitable view about the nature of the neural states that realize our propositional attitudes, and when glossed in the minimal way just offered is even compatible, I dare say, with connectionist Schiffer (1993: 245). models of mental representation." Propositional attitudes can be realized by various neural states and hence, it seems, the metaphor can be extended beyond the mental equivalent of natural language, notably to mental images and 'motivations' for actions as vehicles of thought. Generally, the existence of a language of thought that would be so different from natural languages is doubtful. Cohen (1986) sees problems with the semantics of such a language, Fodor (1975) argues for computations that resemble sentence processing and, in effect, also natural language syntax and semantics. Wittgenstein goes further. He says that we do not need a language of mental representations because there are no mental representations that would be separable from their expressions. We can have the same thought as somebody else, thoughts are not private sense-data, and the possibility of holding a belief and the possibility of expressing it are the same (cf. Wittgenstein 1988: 11). There are no mental states and no private objects. It is not the case that thinking happens 'in one room' and speaking 'in another'. He adds that if there were private objects or impressions, anything could count as their description (cf. Wittgenstein 1984: 131). And so, as he says (1953, § 318, 329), there are no meanings in one's mind that would accompany one's expressions. Beliefs are not hidden behind belief expressions (cf. also Wittgenstein 1980, §585). And yet this does not preclude the possibility of mistakes in attitude ascription. Once the thought is uttered, there is nothing left for putative other vehicles to externalize. But then, does an expression with

262

Discourse, beliefs, and intentions

Chapter 6

many interpretations indicate that the corresponding thought is vague, 'unfinished'? It seems doubtful. It seems more plausible that mental states may not be exhausted in the verbalization.24 Naturally, Wittgenstein's view is not compatible with the proposal of vehicles of thought presented in this chapter. Since there are referential mistakes and problems of belief ascription that would disappear once the reporter could ask the holder of the belief for further information, other vehicles are justified.25 Finally, Carruthers (1996) offers an argument that natural language is involved in conscious thinking by pointing out that human linguistic and cognitive development normally proceed hand-in-hand and so it is possible that they are interdependent. He supports this claim with evidence from aphasia, in which conceptualization is often affected, as well as from the development of normal and abnormal children. No decisive evidence against Mentalese, a separate language of thought is presented, though. The intuition, however, is plausible: "... when early hominids debated with one another about the best way to hunt an elephant, for example, and one of them said, 'We should attack from both flanks', there would have been some advantage in this statement figuring immediately as an item in practical reasoning, to be considered, manipulated, accepted, or rejected, without having to be translated into a wholly different system of representation (Mentalese).' (ibid.: 232). Hence, using language in thinking is said to have some survival value. Ending with an alternative is not satisfactory, but neither are expeditious decisions. The content of language of thought is an open question. Vehicles of thought can be represented by either metaphor without vitiating the account of their interaction in forming and conveying belief. The choice of the narrow construal is dictated by the naturalness of the correspondence between natural language and the language of thought, and hence also by the conceptual simplicity of this metaphor.

6.3.4. Mental Images: The Second tehicle Apparently contradictory beliefs lead to additional processing effort on the part of the hearer who tries to find an explanation for this apparent 24

Cf. also Dummett 1990 and Peacocke (1983: 215). I shall not review here the philosophical debates on private language that complicate the matter further. It would require a historical perspective of going back to Thomas Aquinas's view that the mental word exists as a result of the person's own understanding (cf. Nuchelmans 1973: 197). 25

Section 3

Vehicles of thought in attitude

263

inconsistency. This explanation can be, for example, that perhaps some beliefs constitute a network, while other beliefs belong to different clusters (cf. Bonevac 1984: 198). The believer sees logical relations between the constituents of this network. As yet we do not know of any principles that would guide people in making connections between attitudes. Some vague hypotheses concerning the degree of importance of the information, distance in time between acquiring the two relevant pieces of information, and other such obvious factors in drawing inferences have been proposed (cf. e.g. Ryle 1979: 127). Now, in Situation B from Chapter 4, taken in the context of Situation A, the reporter is faced with a pair of contradictory expressions of belief, namely that Ralph believes that a Bernard J. Ortcutt is and is not a spy. In A, Ralph sees the person and utters a description. In B, he does not see the person and utters his name. Knowing who Ralph referred to, in the sense of intending to refer to, requires further clues. Perhaps in A Ralph was mistaken because of poor eyesight or heavy fog. Perhaps he thought that Bernard was called Smith. Perhaps he did not know the name of the person at all. In B, perhaps Ralph did not know (or: did not know that he knew) Ortcutt but merely read about him in newspapers. Perhaps he confused Mr Ortcutt with, say, Mr Dortcutt. Perhaps he fell victim to a slip of the tongue. All this can be inferred from the situation. Or, if no relevant information is available, one of the possible scenarios will be assumed by John. Which scenario it is depends on the strength and reliability of the type of identification of the individual. We can surmise that since perceptual individuation is the strongest source of reliable information, or, in Husserl's terms, the most adequate meaning-giving act, then Ralph was talking about Ortcutt in A, without knowing, however, that the man he talks about is called Ortcutt. Belief de re is assumed by the hearer in this scenario. This belief comes with a referential intention that is slightly 'weakened' due to the fact that there is important individuating information, namely the name, that the speaker does not know. This weakening is revealed through the juxtaposition with Situation B. The ignorance of the name is not the case of a referential mistake but it is the case of a belief in which MoP plays a significant role. That it plays this role can be corroborated by the fact that John cannot truthfully report on A using the name 'Ortcutt', without falling into contradiction with B.26 So, it is in general possible to lack some essential information about the object of belief and the belief can still be de re, as long 26

Rundle (1979:156-157) goes further and suggests that the reporter can use indirect speech only if he or she has enough relevant knowledge to judge about the ascription of referents to names by the person uttering the sentence. If the reporter lacks such knowledge, he or she should use either direct speech or some further explanation such as 'someone called...', etc. But such caution does not seem to be common practice. As I have argued here, people seem to be governed by the Default De Re principle.

264

Discourse, beliefs, and intentions

Chapter 6

as the believer can individuate the object deictically. Substitutivity in reports may not hold, but the belief is de re. This scenario provides further evidence for the gradation of referring properties of referring terms. 'The man in the brown hat' is referential but it does not automatically lead to a transparent, de re report.27 I shall not have much to say here about mental images as a vehicle of thought. The topic of carriers of thought is still highly contentious and is a case of theoretical assumptions and sometimes speculations. The aim of this chapter is merely to signal that language may not exhaust the tools for forming and conveying beliefs, and to point out that the process of interpretation of belief reports reaches beyond language to contextual clues that require a theory of intentions and looking at how people behave and act. At the level of mental representations, beliefs must be fully interpretable. If they are not fully interpretable at the level of linguistic expressions, then, assuming that the language of thought is similar to natural language (cf. Fodor 1975), there must be representations in the form of images, or non-verbalized 'motivations' to actions, that conduct the remaining content of thought (but see also Cohen's 1986: 228 rejection of an internal language). To repeat, 'motivations' or 'causes' does not mean dispositions. The idea is that actions may reveal what is going on at the mental level, the level of thought. It is easier to acknowledge mental images and actions as vehicles of thought in the case of languageless thoughts. Since language is excluded by definition as a vehicle, there has to be something else that carries the thought. But even in the case of verbalized thoughts there are reasons to postulate other vehicles. It is clear that language does not externalize thoughts faithfully, unambiguously. So, there must be some other vehicles acting simultaneously, language does not 'take over'as a vehicle. Now, if we accepted the language of thought as a fully disambiguated language of mental processes, then the role of the other vehicles would be considerably diminished. If the language of thought does not exhibit the de re/de dicto ambiguity, there is no need to evoke images and actions as further vehicles for attitude contexts. But we do not know what the language of thought is, and, indeed, neither do we know whether the brain performs operations on symbols. In the times of the revival of connectionism, it is possible to surmise that operations in the brain are not computations on symbols (cf. 27

Cf.: "I may mention someone to you by name, and you may not know who it is. But when a speaker makes an identifying reference to a particular, and his hearer does, on the strength of it, identify the particular referred to, then, I shall say, the speaker not only makes an identifying reference to, but also identifies, that particular. So we have a hearer's sense, and a speaker's sense, of 'identify'." Strawson (1959: 16). In this trivial sense such beliefs are de re.

Section 3

Chicles of thought in attitude

265

e.g. Bechtel and Abrahamsen 1991; Bechtel 1994: 200-210; Maloney 1994: 401-407). Hence, it is safer to propose that in the case of some beliefs, images and actions act as vehicles of thought, but without giving any rule as to when it happens. In principle, the vehicles can act as disambiguating devices in utterance interpretation but they cannot be invoked whenever an uncertainty of referring arises. This is the option of the narrow construal of the language of thought that we chose in the previous section. All in all, language often supplies incomplete meaning, an incomplete proposition, due to, possibly, externalizing only the content formed by this particular vehicle. There are several approaches to mental representations that can be used to trace the contribution of other vehicles to meaning. For example, there is Langacker's (1983, 1991) cognitive grammar, Fauconnier's (1985, 1997) mental spaces, Johnson-Laird's (1983) mental models, or Lakoff's (1987) Idealized Cognitive Models.28 Integrated with mainstream possible worlds semantics, there are, among others, Kamp and Reyle's (1993) DRT and Heim's (1988) File Change Semantics. All of these have further offsprings, the latter in dynamic semantic theories, and the former in the accounts of categorization, concept formation including the role of metaphor (cf. an introduction in Ungerer and Schmid 1996; Lakoff and Johnson 1980; Lakoff and Turner 1989; Johnson 1987; Sweetser 1990). Now, these mental representations are not restricted to language as a vehicle if the language of thought is regarded as structurally similar to natural languages. They make use of images and 'motivations' for action as well. Such mental representations are constructed not only from words. A mental model has to 'fit inside anyone's head' and it comprises the totality of the state of affairs compatible with, e.g., a certain assertion. This model, or mental representation, is a schematic assertion, a sample that can be used in different ways. People build such models in order to communicate, to understand utterances, because words and sentences alone do not suffice for a full understanding of conveyed thoughts (cf. Johnson-Laird 1988: 338-343). After all, prepositional attitudes are not attitudes to sentences. They require an object (proposition) that, so to speak, comprises sentences expressing the same thought.29 The hearer does not 'grasp' such a proposition. The content of thought is formed by means of various possible tools and then recovered from various possible clues.

28

For a detailed review see Jaszczolt 1992, Section 2, Chapter 3. Jackendoff(1985: 27) says that mental representations 'group' real world situations by imposing similarities on them: "These situations are alike for us because we are constructed to be able to construe them as alike". This is also a criticism of Barwise and Perry's (1983) realism of situations that are out there in the world and to which speakers 'attune'. 29

266

Discourse, beliefs, and intentions

Chapter 6

There is one essential difficulty with adopting a version of mental models for the semantics of attitude reports. Although the hearer apparently constructs a model of the speaker's beliefs, he or she is not able to check the validity of this construct prior to forming the report and putting it forward for the believer's potential assent or dissent. The hearer cannot, so to speak, 'remain within the speaker's model'. Instead, the hearer has to construct his/her own model. The access to supporting vehicles of thought is not explained as part of this process of model construction. Mental models do not distinguish between our vehicles of thought. They are compatible with them all and underlie them all. For example, Johnson-Laird says that mental images are views of models; they represent through perception or imagination (1983: 157). Other vehicles distinguished here are also such views of models. Hence, we have to look somewhere else for a solution as to how the vehicles interact. Now, as was pointed out in the discussion of linguistic relativity, if a thought or its aspect is not carried by language, then, depending on how we set the terminology, it may not be conceptualized. Lakoff (1987) and Lakoff and Johnson (1980) allow for such preconceptual structures. They talk about a so-called Experiential Hypothesis. Children have certain preconceptual experiences such as body movements or an ability to move and perceive objects. They also have so-called 'image-schemata' such as things as containers; movements as paths; directions up and down; part and whole relations; location of front and back.30 These preconceptual structures seem to be shortcuts in reasoning. Such structures can be assimilated to the type of vehicle called here mental images. All in all, the problem with reporting on beliefs can be summed up as follows. If states of affairs are thought of in images and actions (or prompts to act, 'motivations' for action) as well as sentences, then we cannot always substitute one expression for another because, while the described situation is the same, the images and resulting actions can differ. Both images and actions have been subsumed here under contextual clues.31

30 Abstract concepts are said to arise from these preconceptual experiences by metaphorical projection, as in (i): (i) The prices are going up. Cf. also Goatly (1997: 41-42). 31 This interaction of vehicles seems to be compatible with Sperber and Wilson's (1986a: 232) and Wilson and Sperber's (1988: 139) idea that the prepositional forms of utterances are interpretations rather than descriptions of thoughts. Utterances may represent thoughts that are merely similar to them in prepositional form. This similarity can be developed into the contribution of other vehicles to the externalization of thought.

Section 3

Vehicles of thought in attitude

267

6.3.5. Actions: The Third Vehicle It is a platitude to say that beliefs are connected with actions. Although a belief need not necessarily lead to action, an action can be evidence of the actor's belief. In other words, actions constitute one of the ways of externalizing thoughts. Beliefs can be stronger or weaker and these different degrees of belief are associated with different behaviour.32 Armstrong (1973) compares what he calls a 'full belief to a map we use to steer, a 'mere thought' to an unused map, and a 'partial belief to a map used only in certain circumstances. The degree of belief, however, cannot be easily compared to the number and type of circumstances in which the belief is activated and triggers an action. This behaviourist simplification will not do. In simpler terms, grasping someone's belief does not equal predicting that person's behaviour. But observing behaviour may shed more light on the interpretation of that person's utterance or, in general, may help in reconstructing that person's belief. The connection between beliefs and actions depends, to some extent, on what is believed. Some beliefs bear directly on actions, others do not. When a person believes, say, that there are wolves in the woods, he or she will normally either avoid going there or be careful and well equipped against them - naturally, if the person desires not to fall victim to them. If Ralph believes Ortcutt to be a spy, he will probably report him to the police, tell someone that he is a spy, or try to catch him spying. The adverbs that qualify the connection in the above sentences signal that there is no necessary link there but merely inference from conventions and life experience. On the other hand, believing that Frege wrote Uber Sinn und Bedeutung may not bear directly on actions. In general, people tend to act in a certain way that agrees with their belief. But this agreement does not mean that acting is the belief, as Ryle (1949) has it, or that a belief is a disposition to act as Stalnaker (1984) claims.33 People take it for granted

32

Armstrong (1973:108-109) says that belief ranges from a perfect certainty that a proposition is true to a perfect certainty that it is false. In this account, belief comprises knowledge on one end, and disbelief on the other end of the spectrum. 33 The behaviourist thesis that is relevant here says that to believe that/7 is to hold a belief that p and to be disposed to act as if p was true. Stalnaker (1984: 15) is not far from this tenet: "To believe that P is to be disposed to act in ways that would tend to satisfy one's desires, whatever they are, in a world in which P (together with one's other beliefs) were true." Also, "We believe that P just because we are in a state that, under optimal conditions, we are in only if P, and under optimal conditions, we are in that state because P, or because of something that entails P." (ibid.: 18). Nota bene, belief states explain the apparent contradictions between beliefs: beliefs can belong to different belief states.

268

Discourse, beliefs, and intentions

Chapter 6

in their activities that the world is like the world of their beliefs. Since activities and beliefs are empirically connected,34 it is possible to think up an experiment to check whether, and to what extent, knowing the believer's behaviour would contribute to belief ascription. But since this connection is well tested in everyday life, a common-sensical observation will do.35 Now, once the observer knows an intention with which a person performed an action, he or she knows that person's reason for this action. Davidson (1963: 91) says that, in analysing actions, apart from recognizing a reason for an action, one also has to evaluate the beliefs and other attitudes that lead to a certain action. A person may have several reasons for an action and be unaware which one led to the actual performance. Davidson (ibid.: 93) gives the following example. When you poison someone and think that your reason was to save him pain, you may be mistaken as to your reasons. The other reason may have been to get him out of the way. While the first reason was pleasing, the other was not. The first was erroneously selected as a salient motive that, you think, made you poison the man. So, the connection between reasons and actions may be less revealing than the ones among the vehicles of thought. Thoughts that are externalized in actions seem to distinguish a class of events that are performed with a pro-attitude to them. This is a class of intentional activities. Again, the criterion for distinguishing this class is required. Unless we accept that the information coming from all the vehicles of thought is accumulated during the process of belief ascription, there is a danger of circularity: intentional activities are distinguished as the ones with underlying beliefs and intentions, while thoughts are recognized on the basis of activities. When the observer, or the hearer of a belief expression, searches for information to ascribe a belief, activities are assumed to be intentional. If a person moves his hand towards a painting by Vermeer and says that Rembrandt was an astonishing painter, the movement of the hand is very likely to be taken as an act of ostension, and the referring expression is very likely to be taken as used to refer to Vermeer even if it was an unintentional gesture.

34

Stalnaker would also add desires. Stich (1982:169-170) claims that the correspondence between belief and action can be put into rules: "In identifying a belief as the belief that p, we are picking out the belief by producing an example of the effect it typically has among us, where'typically' is construed not statistically, but by appeal to a typical causal pattern and a special role within that pattern." However, this statement is based on the conviction that action alone identifies a belief and, moreover, it identifies it not as the genuine mental state but rather as what the observer takes to be one's belief, which is, in Stich's view, the only reality. 35

Section 3

Vehicles of thought in attitude

269

There are two questions to be addressed concerning the role of actions. Firstly, how and to what extent actions help in belief ascription, and secondly, how and to what extent they help with the interpretation of belief reports. A formal description of human actions is a difficult task, though. The interpretation is largely dependent on the specific situation, the 'there and then', while the rest relies on conventions, customs and habits. Actions are rarely decisive as a disambiguating device in referential puzzles. Even if Ralph was waiting for someone round the street corner with a gun in his hand or with a policeman accompanying him, and the person approaching was Ortcutt, the observer still would not be in a position to conclude with certainty that by The man in the brown hat is a spy' Ralph referred to Ortcutt, unless he knew that the speaker waited for Ortcutt. And this requires knowing the intentions and beliefs. In practice, such an inference would have normally been performed, unless another situation ensued in which Ralph explicitly denied that Ortcutt was a spy. But this caveat shows that the inference is not always valid. Similarly, if a man is walking with an umbrella open although it is not raining, the observer may infer that he believes that it is raining. But such an inference is not always valid. It is based on common knowledge that people normally open their umbrellas when it is raining. In this instance, however, the man may be trying to protect himself against the sun or against a neighbour watering plants on the terrace. Similarly, seeing a boy praying in the chapel, the observer may feel justified in his inference that the boy believes that God exists. But what if the boy is only trying to please his parents by kneeling there and saying his prayers? It seems that knowing the beliefs and intentions is indispensable. Actions are a useful source of information provided that they are not the only source of information. They can hardly ever, or perhaps never, be conclusive, although they are frequently decisive in practice due to people's reliance on conventions and, so to speak, jumping to conclusions. Coming back to the role of actions in belief ascription and in the interpretation of ascriptions, we can safely surmise that it is equally important in the first and in the second, when the linguistic vehicle leaves speaker's reference unresolved. The lack of activity is also relevant. It is as informative as the action itself. Not doing something, resisting or postponing something, also provides useful information. The order of application of these vehicles is an approximation rather than a steadfast rule. If, say, Ralph points at Smith and says 'Ortcutt is a spy', the action of pointing does not 'cancel' the linguistic vehicle. It adds further clues which can only result in the lack of certainty who Ralph intended to refer to. It is possible that neither vehicle takes over. Ralph may have had blurred vision, but equally he might have been mistaken as to Smith's name. One has to account for people's absent-mindedness and 'slips of action' such as forgetting that a certain action has already been performed and performing it again, forgetting the aim or the order of the activities, confusing two

270

Discourse, beliefs, and intentions

Chapter 6

different activities or switching back, unconsciously, to old habits. All this often happens when the processes become automatic (cf. Cohen 1989:18-19). Moreover, people's memory for events or for other people may be distorted. It is possible not to be able to distinguish between reality (perception) and imagination, especially when the event is not worth remembering or when emotions, including stress, prevent one from remembering correctly. It is possible to forget or change the picture of a past event when a new event or piece of information affects the memory. The same concerns recognizing people's faces or names. It is possible to mistake one person for another, forget who someone is, remember only certain facts about him or her but not others, be unsure about the identity, or fail to identify the person at all. Considering all these weaknesses, actions turn out to be only partial evidence for attitude ascription. Delineating the role of actions would be easier if behaviourism was a correct theory. In other words, if mental states were always translatable into dispositions to certain behaviour, or were this behaviour, then observing this behaviour would be more reliable evidence for belief ascription.36 But mental states coexist and influence one another, and intentional action ensues from this totality of mental states rather than being another way of talking about a particular state. Moreover, there are unintentional activities that have to be, in some way or other, clearly distinguished from actions. Finally, in integrating information from various vehicles, the reporter may realize that the speaker's expression was an echoic utterance or an instance of sentence non-literality. This interpretation can be arrived at through the observation of the speaker's non-verbal behaviour such as the blinking of an eye, smiling in a revealing way and other facial expressions and gestures. These expressions may prevent an inference from utterance to belief from taking place. So, there is a high degree of situation dependence in the interaction of the vehicles but some general patterns can be discovered. The utterance comes first, it is processed in the context of discourse that reveals other thoughts as well as the thought corresponding to the utterance, and in the context of non-verbal actions. But, in spite of the possibility of ordering, there is a significant difference between actions and utterances. Utterances have default meanings that correspond to sentence meaning, enriched, where necessary, to speaker meaning. Actions do not always have such conventional significance. Physical movements may not even be intentional actions by default. They can be interpreted as such or not, depending on the circumstances.

36 According to Ryle (1949: 57),"... the styles and procedures of people's activities are the way their minds work" rather than being their imperfect equivalents.

Section 4

Vehicles of thought in attitude

271

All in all, despite the differences, multiple possibilities of interpretation are present both in language and in actions. They may have to be combined as sources of information about the belief in order to be reliable externalizations of a belief. In isolation, all the sources of information for attitude ascription have limited power and may not be sufficient to allow for an adequate report. Linguistic and non-linguistic actions are all subject to different kinds of errors. Therefore, any ascription of attitude must be regarded as adequate to a higher or lesser extent, at least as long as there can be some hidden factors not accounted for in formulating the report. The same goes for the interpretation of belief reports. Language, as analysed e.g. in Fauconnier's mental spaces, allows for possibilities of interpretation but does not specify when these possibilities occur, i.e. how to choose between them. Actions indicate possible motives for them and possible associated beliefs but are not a reliable guide to them, to evoke only the example of opening an umbrella when it is not raining. Thinking in images may also add to the construction of the belief but this vehicle is the least accessible one during the reconstruction of the belief. Moreover, it has to be remembered that vehicles are evidence of a thought only when they occur against a background of other beliefs, utterances and situations. The influence of these is incorporated in the DRT account in Chapter 7. Normally, though, the utterance is a trustworthy form of externalizing a belief because there are default interpretations and the multiple possibilities of reading are only theoretical possibilities. In practice, default intentions and conventions of use lead the hearer to one interpretation. Additional clues are only resorted to when required. So, the question of vehicles of thought does not frequently arise in practice. Enough said. Some important cross-linguistic differences are considered in Chapter 8.

6.4. Default Interpretations The intentionality of mental states is prior in the order of explanation to the intentionality of the vehicles. It also guarantees it, as the intentionality of the totality guarantees the intentionality of its parts - or the intentionality of the phenomenon guarantees the intentionality of its instantiations. Intentional utterances, intentional actions and intentional mental images are similar in that they all convey meaning. They all undergo the PI and DI principles and so they have stronger and weaker interpretations as far as the reference is concerned, and the strongest interpretation is the default. A mental image of, say, a frog in danger of being squashed by my foot includes the default matching with the real, small, green and slimy creature on the path in front of me. An action of lifting the frog to move it to a safe place off the path

272

Discourse, beliefs, and intentions

Chapter 6

is an action directed at this very animal. It is in this naive sense that default intentionality works for vehicles number two and three. Language, the vehicle number one, requires default intentionality more urgently than these. Defaults are less obvious there to a researcher who lists the possible interpretations. These defaults are only obvious when one realizes how little has to be said for the hearer to pick out the referent correctly. Beliefs are directed at objects by force of their intentionality. Similarly, utterances are about things. The referential intention is the strongest when this aboutness amounts to a clear correlation between physical objects (where applicable) and relations between them on one side, and the sentence on the other. This is a common-sensical, weak version of supervenience, used here for a different purpose from that in Schiffer's supervenience theory proposed throughout his work. While Schiffer uses it as a replacement, so to speak, for compositional semantics, our supervenience strengthens the compositionality of semantics by allowing for this supervenience to be weaker or stronger. Supervening on physical objects and relations between them can be direct or obfuscated, correct or incorrect. This obfuscation, in the form of sensitivity to structure, is taken up in Chapter 8. Gradation is caused by the variable suitability of meaning-giving acts to result in direct reference. Among these acts, perception is on the strong end, and Husserl's'shallow thinking'on the weak one.

6.5. The Core of Meaning There is no common content or object shared by attitudes. There is no proposition towards which beliefs, doubts, or fears are attitudes. There cannot be one because MoPs, 'personalized' ways of thinking about the object, are context-dependent. Geach's famous example (here: 6) demonstrates that the reporter has to take these MoPs into consideration in attempting any substitutivity of coreferentials. It also demonstrates that this attention to MoPs is assumed by the hearer to have taken place. (6)

Hob thinks a witch has blighted Bob's mare, and Nob wonders whether she (the same witch) killed Cob's sow.

(after Geach 1967: 147). The binding of the pronoun by the antecedent expression results in either transparent or opaque readings: Nob may or may not be aware that it is the same witch in both cases. On one hand, binding by an antecedent should create a transparent reading, but on the other, binding by an indefinite noun phrase does not. No reference to a specific witch is made. And yet, a hearer would not have any difficulty in inferring that sameness of the witch with that of Hob's is part of Nob's MoP of the

Section 5

Vehicles of thought in attitude

273

character. This reading seems to be the default interpretation and is arrived at, presumably, through the assumption of the interaction of various sources of information in context. Mental states are individuated by intentionality. As a result, it seems, knowing the degree of intentionality disambiguates utterances. In fact, what remains to be disambiguated is merely insignificant remnants in comparison with the enormous task that the grammar of English has already done for the speakers. Nota bene, this role of grammar should not be ignored in the somewhat expeditious attempts to deny compositional semantics to natural languages. Most referring expressions are perfectly clear and come with a standard referring function which stems out of the hierarchy of referring properties of these expressions which is at the same time the hierarchy of givenness/accessibility of the referent (discussed in Chapter 5). Language is the first vehicle but other vehicles can act alongside it and interact with it.37 Instead of Procrustean efforts to make natural language match sequences of things thought about, one should let in other vehicles of thought. Now, utterances convey messages through the semantic representation of the sentences and the inferences they lead to. Some of these inferences result in enriching the LF, others add new messages. There is a steady core of meaning on which speaker meaning is founded. This steady meaning is unmarked. For the linguist, this meaning can be incomplete until he or she considers default intentions. For the interlocutors, the contribution of intentions is part and parcel of the process that also includes the recognition of the LF. This process was discussed in the proposal of Default Semantics in Chapter 2. There are no stages involved, no transition from the underspecified LF to the fully truth-conditional prepositional representation. This leads to the conclusion that, although language does not suffice as a vehicle of thought, it frequently suffices in practice because some interpretations are unmarked, default, and context-independent.

37 If the interaction of vehicles were to be reconciled with Fodor's modularity of mind, the interaction would have to take place in central cognitive systems that are not modular, i.e. not domain-specific, informationally encapsulated, etc. (see Fodor 1983:101-103). Generally, these boil down to thought or problem-solving. See also Kasher (1991: 389-391) on central, pragmatic knowledge based on general principles such as the principles governing rational intentional action. Kasher also reports on some evidence for and against lateralization of grammar and pragmatics - in the left and right hemisphere respectively.

274

Discourse, beliefs, and intentions

Chapter 6

6.6. Summary Let us take stock. Language does not reflect intentional experience faithfully. There is a distance between language and experience that has to be filled in by other vehicles. This 'filling in' works on two levels that I called the first and the second externalization. The first externalization means forming a thought, or, in a more theory-laden, phenomenological picture, it means bringing the thought from the 'third realm' to consciousness. The second externalization means conveying one's own thought to the observer. Language, mental images and actions play a role in both externalizations. They are equally important in the interpretation of the resulting expression, forming a belief report and the interpretation of this report by the hearer. There is one item that spoils the symmetry, though. In the second externalization, actions allow the observer to judge about someone's belief. In the first externalization, there are mental or neural 'motives' for actions instead. These are changes in the brain that may make the thinking person perform a physical action - naturally, not to be confused with behaviourists' dispositions. In other words, it was suggested in this chapter that reporting on beliefs may sometimes require a consideration of all the means by which the belief is formed (or brought before the believer's mind) and by which it becomes known to the reporter. The interpretation of a belief report may also require this information about the vehicles of thought. These vehicles include a linguistic expression, mental image, and non-verbal action. I have suggested a paradigm of their ordering and interaction that is based on Husserl's idea of meaning-giving acts. The meaning of a belief expression or belief report is also determined by other beliefs that belong to a network. The impact of other beliefs and other meaning-intending acts is summed up in Husserl's notion of a horizon. The priority of perception among meaning-intending acts confirms the idea of the unmarkedness of the referential reading of belief expressions and the de re reading of belief reports. Vehicles of thought also account for the pragmatic contribution to the semantic representation of prepositional attitude constructions in that they allow for MoP to include non-linguistic forms of presentation of the object. The linguistic vehicle does not exhaust the truth-conditionally relevant aspects of meaning and the semantics of propositional attitude expressions requires this systematic treatment of non-linguistic context. Meaning-giving acts and the horizon of the act provide an easily utilizable and largely ignored framework for the semantics and pragmatics. After all, a good semantic and pragmatic theory has to explain not only how the hearer interprets an utterance, but also why he or she does it this way. Or, in Dummett's (1975: 4-5) words, the theory of meaning has to perform at least the following three tasks: (i) to explain the meaning of expressions of the language; (ii) to explain

Section 6

Vehicles of thought in attitude

275

what it is to have the concepts which can be expressed by the analysed language; and (iii) to associate concepts with words. And concepts, whether private or public, belong to thoughts.

This page intentionally left blank

CHAPTER 7

Discourse Representation Theory and Propositional Attitudes Contents 7.1. 7.2. 7.3. 7.4. 7.5.

Semantics and Pragmatics Revisited Context Belief Expressions and Belief Reports in DRT DRT and Intentions Concluding Remarks

DISCOURSE, BELIEFS, AND INTENTIONS Current Research in the Semantics/Pragmatics Interface, Vol. 2 K.M. Jaszczolt © 1999 Elsevier Science Ltd. All rights reserved 277

278 280 284 296 298

278

Discourse, beliefs, and intentions

Chapter 7

7.1. Semantics and Pragmatics Revisited I argued in the previous chapters that all referring expressions have to have intentional content in order to be used to refer to objects because intentionality secures reference. The shift of interest from the fact that Ralph made reference to somebody to how it happened that he attempted to make reference to somebody leads to an answer through a theory of intentionality. The next step is to give a formal DR-theoretic account of the ways in which intentionality secures discourse referents.1 Reports on propositional attitudes, of which belief is the most prominent representative, have had the unfortunate fate of being the object of intra-theoretical disputes of philosophers and logicians, and only indirectly permeating to linguistic pragmatics. The move towards pragmatic explanation or, indeed, blurring the boundary between semantics and pragmatics happened through the disillusionment of some scholars with semantic ambiguities, to mention only Davidson (1968-69, 1970, 1974) or Donnellan (1966, 1970). Semantic underspecification followed suit as a symptom of the tendency to shift more and more responsibility for puzzles with belief attribution from semantics to pragmatics. In other words, the ongoing trend was to see the study of language use as the legitimate participant in the study of sentence meaning. Traces of this tendency are already visible in continental phenomenology from the turn of the century, and notably in Husserl's Investigations (1900-1901) where language is discussed as a social phenomenon, as well as in Frege's insistence on an intersubjective component in semantics in the form of objective sense, grasped by language users (see Dummett 1986: 251). It was brought back to the fore by Strawson's (1950) emphasis on speaker meaning and by other authors who either (1) revive the importance of MoP and stress its contribution to the proposition, although not necessarily to its content or, (2) immersed in the linguistic theoretical debate instead, allow pragmatic information in general to play a role in the semantic content of an utterance. The first trend was reviewed and discussed in Chapters 4 and 5. The latter trend, discussed in Chapters 1 and 2, has evolved into several streams such as the neo-Gricean approaches, relevance theory and its kin the theory of impliciture, as well as dynamic-semantic approaches such as DRT. These tendencies helped redefine the role of pragmatics from a magic wand excluded even from the study of intensional contexts (Quine 1956 is a classic example) to a component of a semantic theory. In this chapter I demonstrate that DRT provides an adequate framework for representing sentences that express reports on beliefs but it requires one essential addition. All the possible readings of a belief report that discriminate

1

The main tenets and terms of DRT were introduced in Section 2.7 above.

Section 1

Discourse representation theory and prepositional attitudes

279

between different ways in which the individual is known to the believer are treated equally, which is not satisfactory. As I have argued in the preceding chapters, some of these readings are more salient to the hearer than others and the systematic gradation of this salience is parallel to the degrees of referential intention involved in uttering the belief report. So, this gradation should be reflected in the discourse representation structures. I demonstrate how an account of intentions supplements DRT and disposes of the need to postulate ambiguities or underdeterminacy. I distinguished two stages of this project. Firstly, one has to estimate the extent to which the semantic representation of an utterance guides the hearer towards the recognition of the communicated content, and secondly, one requires a theory of what supplements semantic representation before it becomes a truth-conditional propositional representation. The first is provided by any theory of implicature and the need for the distinction between what is said and what is implicated does not arouse particular controversies. The second problem is the subject of the ongoing debate concerning the content of what is said, to which my solution was presented in Chapter 2. I shall now use this solution, of the interaction of intentions with the semantic representation, in the DR-theoretic analysis of belief expressions and belief reports. Dynamic-semantic approaches are particularly prone to such amendments because they have a means of incorporating the changes of context into the representation. And, as I argue in Section 7.2, the context includes intentions. Hence, instead of an epistemologically redundant category of a vague, underspecified representation, there is an interaction of the output of grammar (the logical form) and the intentions that accompany the utterance in producing an adequate, single propositional representation. The 'pragmatic intrusionism'of DRT easily allows for this move. Intentions, allowing for default values, also reflect the role of conventions in people's inferences. These conventions and default interpretations are emphasized in neo-Gricean pragmatics and provide another shortcut for semantic theory. To use Fodor's (1994: 45) slogan, "Mostpeople do not marry their mothers; that, surely, should define the norm". Or, in other words, "The world (...) sustains a harmony between the (extrinsic/ historical) properties in virtue of which a Mentalese sentence is a mode of presentation of the proposition that P and the (intrinsic/ syntactic) properties in virtue of which a Mentalese sentence is reliably a cause of the sorts of behavioral proclivities that the laws of psychology say that P-believers share." (ibid.: 53-54). Conventions and intentions are the devices I propose for improving the account DRT gives of belief constructions.

280

Discourse, beliefs, and intentions

Chapter 7

7.2. Context The basic advantage of DRT is that, in a discourse, a sentence is interpreted in the context of preceding sentences. The interpretation of a new sentence relies on the syntactic structure of the sentence and the structure that represents the context, the earlier sentences (cf. Kamp and Reyle 1993: 59). In other words, the context is updated with the information coming from this new sentence (see van Eijck and Kamp 1997: 181). The content of this new sentence draws on the content of previously interpreted sentences, which has important implications for the semantics of the whole language and is perhaps particularly salient in the analysis of anaphoric expressions. Discourse referents enable the representation of coreferentiality, they capture the fact that sentences can be linked by having the same reference. Concerning the role of context in semantics, it has also been suggested that the focus/background structure of a sentence has semantic effects and helps determine the structure of the discourse (see Asher 1995 and Kubon 1995, the latter with respect to its role in negation).2 The novelty of this perspective is its dynamism. While traditionally meaning was tied to truth with respect to models, now it is tied to truth defined with respect to conditions of context-change. This is also how Montague semantics became dynamic and has allowed for the inclusion of pragmatic processes. DRT copes easily with anaphoric links that were problematic for static approaches, such as in (l)-(6):

(1)

A man entered. He smiled.

(2)

Every man who meets a nice woman smiles at her.

(3)

If a man enters, he smiles.

(4)

Hob believes a witch blighted his mare. Nob believes she killed his sow.

(5)

John entered the room. He switched on the light.

(6)

Whenever John entered the room, he switched on the light.

2

Pace de Swart (1995: 199). Concerning anaphora and the topic/focus distinction, she says: "The reasoning is that a topic is interpreted in the top box [of a DRS], and a referent in the top box is available as the antecedent of a discourse anaphor. Being a topic is thus a sufficient, but not a necessary condition to license discourse anaphora."

Section 2

Discourse representation theory and prepositional attitudes

281

(from van Eijck and Kamp 1997: 182-183). Since the interpretation of the first sentence or clause then becomes a context for the interpretation of the other, the anaphoric dependencies concerning objects (in 1-6) and tenses (in 5 and 6) can be captured. For example, in (1), the anaphoric link between the indefinite NP and the pronoun is captured by the DRS condition y = x as in Fig. 1:

x y

man (x) entered (x)

y=x smiled (y)

Fig. 1 (adapted from van Eijck and Kamp 1997: 185).3

Now, the notion of context is quite far-ranging in DRT. Van Eijck and Kamp (1997: 189-190) say that in sentences of the type (7), the definite description 'the street' may not have an antecedent in the previous discourse or text but instead it can be salient to the interlocutors due to sharing common assumptions called 'common ground': (7)

A man was walking down the street.

(from ibid.: 189). This shared background makes the discourse referent for 'the street' anchored to an object, a particular, known street. DRT represents this salience of a referent as a type of an anaphoric link. Van Eijck and Kamp introduce a discourse referent for the street in question and next, by means of the equational condition, link the referent of the expression 'the street' from sentence (7) with that referent. This seems a rather roundabout route, though. For DRS-conditions see Kamp and Reyle (1993: 63). For formal details of DRT and the translation to FoL see e.g. van Eijck and Kamp 1997, Section 4.

282

Discourse, beliefs, and intentions

Chapter 7

Instead, perhaps the scale of salience of referring expressions, such as the accessibility or givenness hierarchy (see Section 5.6) would better handle the shared availability of the referent.4 Although extending anaphora thus far may be somewhat contentious, the intuition behind it is correct. Context is not merely the preceding sentences but also shared knowledge, beliefs, assumptions, and all the situational and cultural background. Kamp (1990: 53) explicitly allows for non-verbal information to enter context. In terms of the approach developed here, it means allowing information coming from vehicles of thought other than language and, necessarily, from intentionality. Kamp (1984: 1) advocates the merger of meaning as truth conditions with meaning as hearer's understanding. In other words, there are conventions of language use and they are brought into the semantics. This perspective has proven to be a natural locus for intentions which, after all, both lead to the hearer's understanding of an utterance and explain how a particular interpretation of the sentence affects truth conditions. So, representations in the form of DRSs depend on a variety of information. We should also include in this information the very specific properties of the involved concepts (cf. Kamp and Rossdeutscher 1994). The very general properties of concepts must also contribute to DRSs. And it can be said without doubt that intentionality is such a universal and omnipresent property of the concept of belief. It follows from this summary of sources of information that intentions that accompany sentences, as well as intentionality that is part and parcel of thoughts and beliefs, are also part of context of interpretation and may enter DRSs. This inclusion of intentions in DRSs will become particularly important in the analysis of belief expressions and belief reports in the following section. Although it seems to be a natural consequence of Kamp's approach to draw on intentions, they have not been used as yet to discriminate among possible readings of belief contexts. I demonstrate in Section 7.3 how it can be achieved. Now, once all this contextual information is used for the analysis of the content of the current sentence, is such a semantics still compositional? Van Eijck and Kamp demonstrate that it is but add that sentences with the same truth conditions may differ in their semantics because they have a different ability to affect the context, as in (8) and (9): (8)

Someone didn't come to the party. He was ill.

Cf.:

"When the context available to the hearer does not contain a representation of the referent of a definite description, he may accommodate this context so that it now does contain such a representation, and then proceed as if the representation had been there all along. However, under what conditions precisely accommodation is possible is still a largely unresolved problem." van Eijck and Kamp (1997: 190).

Section 2

(9)

Discourse representation theory and prepositional attitudes

283

Not everyone came to the party. *He was ill.

Dynamic semantics can capture these differences in the context-change potential through (i) its sequential composition ('sequential merge') of representations ('proto-DRSs'), strategies for which are provided in van Eijck and Kamp 1997, or through (ii) analysing every next sentence 'within' the DRS already present as in Kamp 1984 or Kamp and Reyle 1993. In this chapter I adopt the latter method. Discourse referents can be shared between such representations. In other words, the reading of the quantified expression 'someone' contributes to the semantics. Although in my account these sentences also have different truth conditions, this discrepancy is not significant here. What matters is that pragmatic processes and conventions of language use are let in. 'Someone' obtains a discourse referent and can act as an antecedent for the pronoun 'he'. The dynamic perspective of DRT does not in itself suffice to solve the problem of readings of belief reports and belief expressions. It only provides a linguistic semantics and vague principles for supplanting it with the outcome of browsing through the context. The theory of how reference works in such contexts would strengthen these vague principles.Van Eijck and Kamp (1997: 221-222) make similar claims concerning anaphora resolution: "It would be unreasonable to demand of a theory of linguistic semantics - and it is that which DRT originally aimed at - that it incorporate a detailed account of anaphora resolution, which would have to rely on a host of pragmatic principles as well as on an indefinite amount of world knowledge. It seems not unreasonable, however, to demand of such a theory that it offer a suitable interface to other (pragmatic and/or extralinguistic) components of a comprehensive theory of meaning which are designed to deal with anaphora resolution (...) and to allow these other components to come into action at those points when the information needed for anaphora resolution has become available and the resolution is necessary for interpretation to proceed." Only after the pragmatic processes have been employed does the assignment of anaphoric indexing proceed. A similar strategy seems to apply to expressions of propositional attitude and attitude reports. Once the information from intentions (including the default value of the referential intention) has been utilized, one can provide the semantics, and complete the DRS, of the sentence in question. In other words, the information from

284

Discourse, beliefs, and intentions

Chapter 7

the utterance of the sentence contributes to its semantics. Intentions and conventions, being part of context, determine the content of this semantic representation.

7.3. Belief Expressions and Belief Reports in DRT The readings of belief reports that we have to represent in DRT, and place in the order of accessibility, can be distinguished by the following LFs in (10) to (12). Representations in (10) and (12) are alphabetic variants of each other and are contrasted as descriptions of different situations, x / y. (10)

(3x) (Ralph believes that x is a spy)

(11)

Ralph believes that (3x) (x is a spy)

(12)

(3y) (Ralph believes that y is a spy)

In either (10) or (12), Ralph is referentially mistaken. Let us say that he is mistaken in (12). (11) is a representation of a proper de dicto report. Although the distinction between de re and de dicto cannot be adequately captured by the difference in the scope of the traditional existential quantifier, this is a good enough approximation for now. Its role is merely to start off the building of DRSs. The correct representation will have to reflect the fact that interpretations de re and de dicto depend on the speaker's intentions and that there are degrees to which MoPs matter for the semantics. The distinction appears to be pragmatic, while, at the same time, accountable in semantic theory. The intrusion of intentions to DRSs accounts for the degrees of salience of these readings and, as a part of the same job, wipes off ambiguities between de re, de dicto proper, and de dicto/.5 The treatment of belief reports in DRT relies on the notion of eventuality, or state and event. Events originate with Davidson (1967, 1969), at least in modern philosophy of language. They give rise to discourse referents and are themselves referred to. To repeat, a discourse referent stands for an individual, including state or event, or for a group of individuals. It is a way of representing the individual formally. Davidson (1970: 58) incorporates time and place indexicals into truth-conditional semantics because the truth value of a sentence may vary depending on the time of utterance, the speaker, and the audience. Events are like 'concrete individuals', to name only births, movements, or explosions. The differences between eventualities represent the differences of reading of belief reports, as is presented in Figures 5-8. See Chapter 4 for this tri-partite distinction.

Section 3

Discourse representation theory and propositional attitudes

285

Dealing with attitude reports involves an assignment of a theoretical status to the embedded sentence. One can regard it either as a sentence (utterance) of the holder of the belief or as a proposition. Davidson (1968-69), similar to Hintikka (1962), regards the complementizer that as a demonstrative. Although it is a rather radical theoretical commitment, perhaps some traces of this demonstrative meaning are useful to retain. Then we can take what follows 'that' to be an utterance of the reporter that is in itself an interpretation of the original belief expression. This utterance may already include corrections of the referring expressions that were wrongly used, disambiguation and enrichment of what was said, to the extent to which the reporter was able to add them. These pragmatic processes are at work prior to the formulation of the belief report and hence contribute to the semantics, as was observed already by Davidson in 1968-69. Davidson (1974: 144) is sceptical, though, about the use of beliefs and intentions for interpreting belief reports because they are all so tightly interrelated. However, as was discussed in Chapter 4, the theories of both use MoPs and the main difference is that whereas the theory of beliefs must specify the MoP of the individual, the theory of ascription resorts to coarser-grained distinctions. It requires only those MoPs and types of MoPs that make a difference to truth conditions of sentences. And these pertain to the tri-partite distinction between de re, de dicto proper, and de dictoj. Discourse interpretation is incremental in the sense that the meaning recovered from the utterance relies on the foundations prepared by the understanding of previous utterances, non-linguistic context, and the interaction of the recovery of the speaker's assumptions by the hearer with the creation by the hearer of new assumptions. This last tenet I have called a doubly-dynamic approach (in Section 2.9 and Jaszczolt 1996b), as opposed to singly-dynamic approaches where context is established after its relevance for the interpretation has been assessed. DRT well reflects this dynamism. Rules, called a construction algorithm, represent a discourse as a DRS with its truth conditions. As was mentioned in Chapter 2, discourse representations are mental representations formed by the speakers and hearers as a result of the utterances to which they are exposed (cf. Kamp 1984: 5). A DRS consists of two components: a set of discourse referents (the universe of the DRS) and a set of DRS-conditions (cf. Kamp and Reyle 1993: 63). A DRS is true iff there are individuals in its universe that correspond to the discourse referents and these individuals satisfy the DRS-conditions. The different readings of belief reports are represented by the selection and placement of discourse referents that stand for individuals, here in the subject position of the embedded sentence. I shall now demonstrate that the construction of the context, that tells us which readings are more and which less salient, is tied with the role of intentions in semantics as presented in the theory of Default Semantics in Chapter 2.

286

Discourse, beliefs, and intentions

Chapter 7

Kamp (1990: 41) argues as follows. We cannot specify the truth-conditional content of individual beliefs, so we have to look for the intentionality of a complex attitudinal state whose complexity can be grasped by constructing representations of sharing intentional elements in the DRS. In other words, people are in certain cognitive states which have the structure of a DRS (cf. Kamp 1984,1990, in progress) and truth conditions can be specified only for these states as a whole rather than for particular attitudes. But these states comprise attitudes that are connected according to certain rules, and notably by sharing discourse referents. Discourse referents can act as links between conditions that come from different sentences (or propositions - as contents of particular constituents of attitudinal states).6 Now, the discourse analyst normally focuses on the interpretation of a single belief report and we ultimately want to know the truth conditions of such reports on single attitudes. Conventions of language use seem to fill the missing gap. In using language, the interlocutors operate largely on the principle of making predictions from past linguistic experience and on the conversational principles which specify how much and how little we ought to say, such as Levinson's (1987) I, Q, and M principles, relevance theory, or other derivatives of Grice's Co-operative Principle. Therefore, the reporter, on hearing (13), is very likely to be able to recover the speaker's intentions because he is guided by some such principles of co-operation. (13)

The best Italian painter painted this picture.

A DRS of a report on (13), such as in (14), has two properties that make it doubly-dynamic: (i) it gathers all that there is to be recovered from the conversation up to utterance (14), and (ii) it gives options as to the placement of the discourse referent in the DRS. (14)

Mary believes that the best Italian painter painted this picture.

The options of placement at subordinate levels were introduced for other types of contexts in Kamp and Reyle (1993: 298). These options are marked in Fig. 2 by a broken line that delineates the range of possibilities of placement of the discourse referent in the universes of various DRSs. The three DRSs in which y can be placed correspond to the readings de re, de dicto] and de dicto proper. The distinction is made by means of the depth of embedding of y. 6

Strictly speaking, beliefs and other attitudes are not relations between the holder of the attitude and a DRS. Instead, people are in cognitive states that can be represented as DRSs (cf. Kamp 1990: 42).

Section 3

Discourse representation theory and propositional attitudes

287

m x s

best Italian painter (y) Mary (m) picture (x)

m believes e

Fig. 27

's' and 'e' stand for a state and an event respectively. This is a partial representation, the temporal reference is not represented in the DRS as it is not relevant for the present purpose (see Kamp and Reyle 1993, Chapter 5). The depth of embedding of the discourse referent y is inversely proportional to the degree of referential intention that accompanies the use of the referring expression standing for y. In other words, the deeper y is in the DRSs, the less referential intention accompanies the use of the corresponding expression. The degree of their intensionality is tied to the degree of their intentionality. And, as we know from Default Semantics and the Default De Re principle, the least depth of embedding marks the default, de re interpretation. I retrace the steps that led to Fig. 2 below. It has to be demonstrated next to what representation these degrees of intentions apply. We have to present representations of the readings de re, de dicto proper and de dictoi and then show what makes them ordered on the salience scale for the process of utterance interpretation. I shall abandon for this purpose Kamp's complex attitudinal states and instead of looking at many pairs, I shall look at one pair of a belief indicator (Bel) and a DRS representing its content. In fact, I shall look just at the content rather 7

This DRT account of belief reports originated in Jaszczolt 1998d.

288

Discourse, beliefs, and intentions

Chapter 7

than a pair because the object of analysis is linguistic expressions. This analysis of belief expressions is founded on the analysis of beliefs for the following reason: "What is needed (...) is a theory of belief which does justice both to (i) the intentionality of beliefs and (ii) to the fact that their identity conditions are stricter than those of the corresponding sets of worlds." Kamp (1990: 30). This should be founded on a theory of practical reasoning, including beliefs, assumptions, and resulting intentions. Kamp is saying here that while sets of worlds are traditional constructs for the semantics of sentences, they are too coarsely-grained for mental states. More finely-grained devices are needed. But, as was discussed in Chapter 4, expressions and reports on beliefs also need a more finely-grained device. While beliefs always have to be analysed in terms of MoPs, belief expressions and reports sometimes do, i.e. when the default reading is not the case. So, they may need MoPs or types of MoPs. We seem to have a scale of fineness here. In other words, to represent the meaning of belief sentences and reports, we need the same devices as for representing beliefs, only slightly coarser-grained: not all the MoPs have to be relevant for the semantics. Now, beliefs that underlie belief expressions are recovered with the help of default intentions. Intentions indicate how the discourse referents are connected with the belief. In other words, they indicate how to choose from within the box marked by the broken line in Fig. 2. As a result, we obtain a DRS that is fully determined as to the use of the referring expression. Let us examine the possibilities of referring in more detail, beginning with expressions of belief. As ever, I concentrate on definite descriptions and proper names. When the NP denotes uniquely, the proposition is normally called singular and the DRS and the referent are 'externally anchored', they are directly about the object in the world (cf. Kamp 1990: 51 and in progress). As was discussed in Chapter 5, proper names are quite often singular terms, pronouns normally are, and definite descriptions denote uniquely when they are used referentially. In (13), if 'the best Italian painter' refers uniquely, then it acts as a directly referential term similar to an indexical or a demonstrative NP (cf. 'this picture' represented in Fig. 2). Naturally, the description may require expansion to, e.g., 'the best Italian painter according to the speaker', 'the best Italian Renaissance painter', etc. These pragmatic processes are ignored here. This reading is represented in Fig. 3. The discourse referents for 'the best Italian painter' and 'this picture' are anchored to the real entities - the person a and the object b.

Section 3

Discourse representation theory and prepositional attitudes

289

x y e

best Italian painter (x) picture (y)

Fig. 3

External anchoring, however, is not always the case and it can only be regarded as a norm because the referential reading of definite descriptions is the default. Other readings also have to be represented. If the description 'the best Italian painter' is used referentially but the speaker is referentially mistaken, then there are no external anchors. The only available information specifies how discourse referents are taken by the speaker, i.e. delineate the MoP. At best, this is done by means of the so-called internal anchors (formal conditions), as in Fig. 4 (cf. Kamp 1990: 55 and in progress). Cn(x) stands for an n number of descriptive conditions C that describe the speaker's knowledge about the referent, such as 'the man I met at an exhibition', 'the person on this photograph', etc. The person the speaker is thinking of may not be a painter at all. x y e

best Italian painter (x) picture (y) Anch [x]

Fig. 4

290

Discourse, beliefs, and intentions

Chapter 7

Perhaps this case of a referential mistake is more plausible in instances of perceptual mistakes such as in uttering 'the man in the brown hat', who is in fact Ortcutt, and thinking of Smith. This situation, discussed as A and B in Chapter 4, is followed up later on. I am modifying Kamp's external and formal anchors somewhat in order to account for the tri-partite distinction in referring properties of definite descriptions. Kamp (e.g. 1990: 75) uses the device of sharing the discourse referents in order to represent belief ascription. Sharing reflects the hearer's assumption that the speaker and the hearer think about the same individual. This device is too strong and unnecessarily complicated, though. Firstly, it seems to assume the identity of descriptive conditions that lead to reference assignment both by the reporter and the believer. Secondly, it does not capture correctly the spirit of de re beliefs and expressions being directly about an individual and de re reports being directly about the shared individual.8 Now, definite descriptions can be used referentially and correctly, as in Fig. 3, referentially and incorrectly, as in Fig. 4, or they can be used attributively and not anchored. By default, they are used as in Fig. 3. The distinction into singular and general reference adopted in Kamp 1990 is not very useful now because we have degrees of referential properties instead and three readings to place on the scale of salience. On these terms, Fig. 3 represents the 'strongly' referential reading, whereas Fig. 4 depicts the 'weakly' referential one, the one allowing for a referential mistake. The attributive reading is not represented here, the description simply remains unanchored for the reasons to be explained shortly. Proper names can also be intended to pick out a person or object and the intention can be either strong or dispersed due to a referential mistake. But they can also be intended as constituents of a proposition which cannot be further divided but instead is to be taken as an eventuality rather than a predication about an individual (see Cresswell 1985 and Chapter 8 below). By this reorganization of the playground I have adhered to the distinctions postulated in Chapter 4 and brought them into DRT. To sum up, the referring function of descriptions and names is taken to be pragmatic and has to be captured by drawing links between DRSs and making choices between the possible anchorings. This placement, as well as linking DRSs that share discourse referents, is guided by the recognition of the speaker's intentions in communication. It is not a semantic distinction recoverable from the LF but it belongs to the DRS. Now, although Kamp and Reyle observe that who is meant by a proper name depends on the context, including the speaker's intentions, there are no rules as to how to incorporate this information into semantics. Ambiguities of scope are said

This weakness was attended to in Kamp (In progress: 1) but also left open there.

Section 3

Discourse representation theory and propositional attitudes

291

not to arise because proper names refer directly. But, as was argued in Chapter 5, proper names can be used with a referential mistake or without a referential intention, similar to definite descriptions, and this use affects the semantics of the sentences. Direct referring is merely the default. Hence we need possibilities of placement in multiple-embedded DRSs and intentions to discriminate between these possibilities. Intentions also explain the tendency of definite descriptions and names to trigger the de re interpretation. In Fig. 2, I translated the types of anchoring into levels of embedding of the discourse referent in the DRSs. Let us analyse this translation in more detail. Kamp (1990: 60) says that, analogously to formal and external anchors, there are formal and true de re beliefs. This is incontestable: a belief can be taken to be de re but still pose referential puzzles, as was instantiated in Situations A and B in Chapter 4. We are interested in tying this distinction with the process of interpretation of belief expressions and reports of the form A believes that B cps'. We want to know how the hearer decides that the speaker holds a belief de re, de re about someone else (here: formally anchored), or de dicto. Now, a belief report reports on , where Bel is the attitude and K is the DRS. In the case of strongly referentially used expressions, K is externally anchored9 and the DRS of the celebrated report on Ralph's belief looks as in Fig. 5. The proper name 'Ortcutt' is strongly referential here and the discourse referent 'o' is placed in the main DRS.

r o s Ralph (r) Ortcutt (o)

r believes s'

Fig. 5 (adapted from Jaszczolt 1998d: 14). 9

Externally and formally, because 'No external anchor without a formal anchor' (Kamp 1990: 61). In other words, there are always MOPs that lead to the individuation of the referent. But, as I argue below, the formal conditions do not contribute to the semantics on the default reading.

292

Discourse, beliefs, and intentions

Chapter 7

When the proper name is used referentially but mistakenly, the DRS is only formally anchored. I have translated this anchoring of belief into the second level of embedding in the belief report, as in Fig. 6. The discourse referent 'o' is introduced in the middle box that represents the world of Ralph's belief rather than the objective reality.

r s

Ralph (r)

s1 o r believes s' Ortcutt (o) s

o is a spy

Fig. 6 (adapted from Jaszczolt 1998d: 14).

Fig. 6 represents a DRS for a belief report with a de dictoj reading, corresponding to a belief de re about someone else, formally anchored. Similar to Fig. 2, Figures 3-6 are only partial DRSs as they do not represent the temporality of the eventualities. Now, although according to DRT optimally referring expressions normally establish their discourse referents in the universe of the main DRS (Kamp and Reyle 1993; Reyle 1993), this default may not take place. Instead, the expression can be used without the backing of the necessary individuating information but with mistaken information or no sufficient information at all. This means a departure from direct reference and adopting instead direct reference as a default reading which can, nevertheless, not happen on a particular occasion of use. External and formal anchors in beliefs account for the readings de re and de dictoj of belief reports. De dicto proper requires a different treatment. On this interpretation, the believer holds an attitude towards the proposition as a whole, e.g. that 'red giants become white dwarfs' or that 'Botticelli painted Flora', whatever the entities referred to are. As was indicated above, here Kamp (1990) proposes the idea of sharing discourse referents between various

Section 3

Discourse representation theory and propositional attitudes

293

holders of attitudes. For instance, a belief may be inherited from someone else, together with the assumed reference of the potentially referring expressions. But it seems that sharing, albeit psychologically correct, will not suffice in pragmatic theory. It can be taken for granted that someone else's utterance or other behaviour acted as an instigator in forming this semi-propositional10 belief but this fact is not interesting. The source of the belief may no longer be accessible to the speaker or the hearer. The referents, to all intents and purposes, remain oblique and unidentified. The belief is about the eventuality as a whole. Naturally, if information about sharing is available, the DRS of the source's belief will contain this information. But what really matters is what the particular belief state in question, inherited or not, is like. It is de dicto and can be represented as in Fig. 7:

r s

Ralph (r) s1 r b>elieves s'

s'

o Ortcutt (o) o is a spy

Fig. 7 (adapted from Jaszczolt 1998d: 16).

So, the anchoring that accounts for various types of beliefs can be disposed of in belief reports because it has been translated into the levels of embedding of the discourse referent. Anchors function in belief interpretation. Beliefs have MoPs that are important for their content. For linguistic expressions, a more coarsely-grained device of levels of embedding will suffice. I have suggested this device, in the form of degrees of fineness of types of MoPs, in Chapter 4, as a development of Schiffer's types of MoPs. The depth of embedding represents the degree of opacity and, more directly, the degree of referential intention with which the referring expression was used - or taken by the In Sperber's sense. See Chapter 2.9 above.

294

Discourse, beliefs, and intentions

Chapter 7

hearer to have been used. It can be seen that Kamp (1990, in progress) provides an account of beliefs that is immensely valuable for my account of belief expressions and belief reports. My levels of embedding are a direct consequence of types of conditions (anchors), with the small proviso that direct reference in its standard form has to be rejected in favour of degrees of intentions. The gist of my proposal is as follows. In our Default Semantics, there are degrees of 'being de re'. The three different cases of anchoring (including no anchoring) do not reflect these degrees very well. Instead, they seem to indicate three totally separate options of interpretation. To remedy this, we should think about them this way. External anchors are the ultimate, objective type of link between an entity and a representation. As Kamp says, they require the co-presence of formal anchors. One has to have sufficiently solid and effective knowledge of the referent to individuate it correctly. In other words, one has to have enough sufficiently detailed MoPs to hold a belief de re, use a term referentially, or utter a de re report. But in the de re/referential case, these MoPs, or descriptive conditions acting as formal anchors, can be ignored for the semantic purposes. This is the case of the default direct reference and the external anchor 'overrides', so to speak, the need for any descriptive conditions. The referential mistake also relies on some formal anchors, but these do not lead to external anchors. They, or rather MoPs or types of MoPs, are in fact relevant to the semantics of the expression in question. They explain the problem of the lack of substitutivity of coreferential expressions in such cases. Finally, thinking or talking about whoever undergoes the expression relies on a particular MoP that is present in the sentence itself. There are no anchors other than the very sentence. Hence, although the MoP matters very much to the semantics in this case, I call the discourse referent unanchored. This is a departure from Kamp's more extensive use of formal anchors but it seems to render the differences between the readings more accurately. In DRT, sentences in a discourse can be linked through their referents, as the notion of sharing indicates. Attitudes are linked analogously: "Distinct attitudes may have contents that are connected in much the same way as the DRSs for the different sentences of a cohesive text." Kamp (in progress: 6). Similarly, discourse is linked with attitudinal states through discourse referents. As a result, if beliefs are connected with intentions, then so are belief expressions and belief reports. Intentions are even more suited as a device for the interpretation of linguistic expressions because they belong, so to speak, 'one level higher' than expressions, whereas with beliefs they simply enter into complex attitudinal states. Moreover, only the

Section 3

Discourse representation theory and prepositional attitudes

intention to refer has to be considered, all the other intentions for granted, assumed to accompany a speech act. To sum up, of interpretation of the report on Ralph's belief are analogous in Fig. 2 and are presented in Fig. 8. Proper names behave in analogously to definite descriptions.

r s

295

being taken the options to the ones this respect

o

Ortcutt (o) Ralph (r)

s1 s

r believes s1 s1 |o is a spy

Fig. 8 (adapted from Jaszczolt 1998d: 16).

The instantiation of the options of placement of the discourse referent in Fig. 8 in the outer, middle or inner box depends on the referential intention that accompanies the utterance. This referential intention guides the hearer in utterance interpretation. The hearer is also guided by the conventional, default status of the de re interpretation that corresponds to the placement of the discourse referent 'o' in the outer box. This default reading was proposed in Chapter 4 as the Default De Re principle and the arguments for its acceptance will not be repeated here. The objective of this analysis was to demonstrate that the intention-based semantics is compatible with DRT and, in fact, improves the latter theory in that the possibilities of reading of referring expressions are not treated as a case of ambiguity or indeterminacy but instead they are classified on a scale of salience. The sole enumeration of the readings, without the intention-based classification, would be uninteresting from the point of view of discourse interpretation. The device of the box marked with a broken line was initially introduced by Reyle (1993: 129) to mark the underspecified meaning of NPs. Reyle attempts to represent scope ambiguities by adopting underspecified structures. For

296

Discourse, beliefs, and intentions

Chapter 7

example, he says that the meaning of an indefinite NP is underspecified as to the positioning in the multiply-embedded DRSs. But it seems that these options need not necessarily mean semantic underspecification. After incorporating intentions into the context of interpretation it becomes clear that, although his device of representing possibilities of reading is very useful, underspecification does not follow suit. As I have demonstrated, in intensional contexts of belief reports, the broken line is a result of superimposing various possibilities of reading. The relative salience of these readings and their semantics are a separate story. In other words, there is more to the semantics of belief contexts than Figures 2 and 8 demonstrate. These figures are theoretical constructs. 'Real' semantics is presented in the remaining figures and is intention-based. This account also reflects the conclusion we reached independently in Chapter 4 that a unitary formal account of attitudes and reports cannot be provided. The account of reports is much less finely-grained, MoPs are considered only when they are required for explaining why the default interpretation is not the case. Also, sometimes only some elements of full MoPs can be relevant, i.e. the types of MoPs. On the contrary, an account of beliefs should incorporate the MoPs of the objects. Who is meant by 'Ortcutt' or 'the man in the brown hat' depends on the context (cf. Kamp and Reyle 1993: 62, 253). But this context may not be relevant to the semantics of the belief expression or the belief report either because it is the case of the default reading or, simply, because this context is not known to the hearer. In the DRSs of reports, the position of the discourse referent for the referring expression reflects the conditions for finding the individual to which the name applies and these conditions contribute to the truth conditions of the represented proposition. Only to the extent of selecting the appropriate conditions does the MoP matter to the semantics of these expressions. Similarly, in (15), these conditions will select the reading where 'the professor who supervises his dissertation' has, or does not have, the scope ranging over 'every student' (with 'his' as the so-called 'dangling pronoun'): (15)

Every student fears the professor who supervises his dissertation,

(from Kamp and Reyle 1993: 297).

7.4. DRT and Intentions This merger of the theory of intentions with DRT is an extension of Kamp's (1990: 30) suggestion that a theory of belief should incorporate the intentionality of beliefs, as well as the fact that the sets of worlds

Section 4

Discourse representation theory and propositional attitudes

297

corresponding to these beliefs do not suffice for providing their identity conditions. Since the theory of belief attribution is merely a coarser-grained version of the theory of beliefs, both of them founded on grades of MoPs, intentionality of beliefs figures in both. So, a pragmatic account of belief expressions and belief reports has to presuppose the intentionality of beliefs and use information coming from intentions in communication. However, although Kamp clearly states that a theory of propositional attitudes must account for their intentionality, he allocates intentions a rather insignificant role (see ibid.: 85-87). He adopts complex attitudinal states as the object of analysis beyond which we cannot reach, or rather we do not yet know how to reach. To repeat, the states combine beliefs, desires, intentions and other modes and there is no way of distinguishing particular components for a truth-conditional interpretation, although some attempts at representing the interrelations among them have been made (see Kamp in progress). These states, coming with the lack of discrimination of intentions, do not suffice for the analysis of attitude expressions. Intentions clarify the belief, knowing the speaker's intentions helps ascribe beliefs to the speaker. So, a complex attitudinal state cannot really be a primitive, the interpretation of expressions of the type 'A believes that B (ps' relies on parts of these complex states. It relies on the underlying beliefs and on the intentions that accompany the utterance. It may be fair to say that we do not have a satisfactory insight into people's beliefs and intentions but then 'complex attitudinal states' are the admission of our ignorance rather than states of any epistemological utility. Kamp and Rossdeutscher (1994: 167) admit that text interpretation and inference are notions of darkness as they are based on complex interrelations of linguistic and extralinguistic information. Intentions and conventions provide the missing explanation. The speaker communicates certain propositions with the intention that they are understood by the hearer in a particular, intended way. Even when a vague or ambiguous message is communicated, it is an intended vagueness or ambiguity. Meaning grasped should thus equal meaning intended and it is this grasped intended meaning that enters the DRS. This contribution of intentions is captured by the POL principle. What it means for the analysis in terms of DRT is that we do not have semantic representations of the type as in Figures 2 and 8. Instead, there are propositional representations of belief reports as in Figures 5-7, and there are representations of belief expressions as in Figures 3 and 4, as well as the unanchored one. Figures 3 and 5 present the respective defaults. The DRSs confirm the conclusion reached before that intentions come in various strengths and they are markers of the different readings of the sentences. The default status of one of these readings also comes from DRT. It is stated there as a directive that definite descriptions normally establish their discourse referents in the main DRS, similar to the expressions traditionally grouped as directly referential.

298

Discourse, beliefs, and intentions

Chapter 7

Now, when Kamp and Reyle (1993: 292) talk about the processing of indefinite NPs that it must allow for "a considerable spectrum of alternatives", it looks like these are alternatives to the default. This default representation is that indefinite NPs establish a discourse referent in the current DRS. Short and relatively uninformative indefinite NPs as in (16) normally acquire a nonspecific interpretation and narrow scope: (16)

Every boy in Mary's class fancies a girl.

(from Kamp and Reyle 1993: 293). These constructions were discussed in more detail in Section 2.7 above (cf. example 20 there). Since there is no formal rule as to when narrow and wide scope ensue (also called 'short' and 'long' indefinites), then all we have is a disjunction of processing options for the hearer to choose from. This is not satisfactory. Processing options will not do, neither will they do for referring expressions in attitude reports. The default reading is always present and has to be accounted for, and so must the non-default readings. This is why intentions (and the lack of intentions) are crucial.

7.5. Concluding Remarks

Conversation requires sharing of meanings, which entails sharing of discourse referents. Conversation also presupposes the concept of other person with his/her intentionality. Since the other person is necessary for communication, then so is that person's intentionality which becomes shared intentionality (cf. Searle 1990b: 415). To summarize the role of sharing would be nothing else but evoke Grice's Co-operative Principle. The hearer makes assumptions about the speaker which serve as premises in inference. However, intentionality has to be assumed. It has been smuggled into DRT just as it has been smuggled into Default Semantics. Intentionality, as a property of mental states and in particular mental acts, is prior to the meaning of linguistic expressions in the order of explanation and hence fits the role very well. As the property of aboutness, it accounts for the major property of speech acts, that is referring to objects. Since it seems to be gradable, it allows for the classification of DRSs of a sentence on a scale of this aboutness. All in all, since it fits the purpose so well, and as a property of mental acts is almost a platitude, there seems to be no harm in assuming it. Demonstratives provide some evidence that intentionality of mental states does not differ principally from the intentionality of linguistic expressions. Demonstratives refer because their meaning is a combination of the meaning of the demonstrating word and the intentional content of the act of perception. And perception, like thought, is one of Husserl's meaning-giving acts. This

Section 5

Discourse representation theory and propositional attitudes

299

identity of genus of intentionality of linguistic expressions and intentionality of perceptual acts, adopted since by Searle 1983 (see also de Mulder 1994: 6-38), signals that many different mental states have it, and hence so does language, as a vehicle of thought. To sum up, I have demonstrated in this chapter that the pragmatic intrusionism to which DRT subscribes includes intentions in communication and that the referential intention contributes to the DRSs. As a result, it enables the ordering of readings of sentences with referring expressions on the scale of their salience in utterance interpretation. DRSs take the best of the two worlds and put it together: model-theoretic semantics and the hearer's utterance processing. In this chapter I have concentrated on improving the view of the latter and its impact on the first. Intentions explain why neither ambiguities nor underspecification have to arise. I have departed from Kamp's theoretical commitments at some places, while retaining most of his DRT apparatus. For example, he maintains that the discourse referent for a proper name is always externally anchored (cf. van Eijck and Kamp 1997: 191; Kamp and Reyle 1993: 247-8). This is a common practice for direct-referentialists. However, since we have taken speaker's referring as the starting assumption in building our semantics, the referent can be salient to a certain degree to the speaker and this is reflected in the placements of the discourse referent in the multiple embedded DRSs for belief reports, or in the types of anchoring (including none) for extensional contexts. The device of degrees of embedding replaced sharing discourse referents which was found to be overly finely-grained to render the three readings of belief reports. Also, the role of formal anchors in semantics was reduced here to the case of de dictoj and other expressions with referential mistakes. As was argued in Section 7.2, the inclusion of the speaker's perspective on referring is justified in DRT through the inclusion of intentions, including the referential intention, in the construal of context. Finally, let me place this intention-based DRT in a wider perspective. There are two traditions in the philosophy of language called contextualism and anti-contextualism (see Recanati 1994: 156-157). Contextualism allows for the contribution of contextual information to the propositional form. For example, in this account, the meaning of the natural language connective and differs from the meaning of truth-functional conjunction in first-order logic. Anti-contextualism treats contextual information as implicatures that are separate from the propositional form. It merely allows for reference assignment to indexical expressions. According to anti-contextualism, there are context-independent, 'eternal' sentences, i.e. sentences whose truth conditions can be formed without taking the context of utterance into consideration. DRT subscribes to contextualism. The truth-conditional representation requires the resolution of reference that can be provided by

300

Discourse, beliefs, and intentions

Chapter 7

the context.11 It may also require the standard indexical resolution, lexical and syntactic disambiguation, as well as completion and expansion of the proposition expressed (discussed in Chapters 1 and 2, see also Bach 1987a; Levinson 1988; Recanati 1981, 1989b). All these processes contribute to the truth-conditional representation and therefore their effects are shown in the DRSs. DRT, a dynamic approach, shows how to incorporate information from other utterances and from conventions of language use. I have suggested here how to add information from intentions and beliefs. I have not made use of the full possibilities of DRT in capturing links between utterances in discourse. This aspect has been well researched and is continued to be so for example in the studies of anaphora. Instead, I concentrated on the links between utterances and attitudinal states. I have extracted beliefs and intentions from the latter and proposed an analysis of belief expressions and belief reports in terms of (some of) the distinctions Kamp proposes for them, such as external and formal anchors. These anchors proved to be easily translatable into levels of embedding of the discourse referent in DRSs for belief reports.

11 Although it may not require knowing the MoPs. To repeat, this would be an overly finely-grained account for semantic purposes.

CHAPTER 8

Belief Reports in a Contrastive Perspective "How can I stick to a particular language (or to an angle from which the world is seen, or to a rule of interpretation of the whole of experience) and not endow it with a privileged cognitive prowess? And if I pretend to have at my disposal a higher, or even an absolute, language, then it either would be suitable only to talk about other languages but not about the reality they refer to, or it would be a standard language of which the other[s] are incomplete dialects. If it is the latter, it is really a divine absolute tongue, embracing all conceivable points of view. Such a tongue is impossible (...). If it is the former, my particular language (of the first degree, the language of things), may not indeed make claims to any privileged position, but I would not be capable of explaining in it the absence of this privileged position: to do that I should abandon my language and turn to a super (or meta) language in which my particular position is inexpressible." Kotakowski (1988: 102).

Contents 8.1. 8.2. 8.3. 8.4.

Belief, Culture, and Translation Contrastive Semantics and Pragmatics Context and Markedness Complementizer That in Contrast 8.4.1. English and Polish: A Surface Overview 8.4.2. Degree of Clausiness 8.4.3. A Lambda-Categorial Language and Sensitivity to Structure 8.4.4. The Application: Defaults and Clausiness 8.5. Final Remarks: What Languages Reveal and Hide

DISCOURSE, BELIEFS, AND INTENTIONS Current Research in the Semantics/Pragmatics Interface, Vol. 2 K.M. Jaszczolt © 1999 Elsevier Science Ltd. All rights reserved 301

302 306 310 311 311 311 320 324 332

302

Discourse, beliefs, and intentions

Chapter 8

Translation is important for the semantics of belief reports for one fundamental reason. Whereas in English attitude reports yield a choice of interpretation between de re and (varieties of) de dicto, this need not be the case in other languages. If de re and de dicto are overtly, lexically or grammatically, differentiated, then this distinction may shed some light on the semantics of belief reports in general. In this chapter I take up the claim that the de re and de dicto readings of belief reports exhibit different sensitivity to the structure of the embedded clause. In other words, a varying amount of syntactic information is relevant for their semantics. This claim is supported by the analysis of the Polish equivalents of these constructions where the different meanings are grammaticalized and lexicalized. The de re reading, which exhibits the greatest sensitivity, will again turn out to be the default one. The structure of this chapter is as follows. First, I introduce some common problems with the criteria for adequate translation, including the notion of unit of comparison or tertium comparationis. Next, I discuss the translation of the prepositional and implicated content of utterances by distinguishing semantic and pragmatic equivalence in translation. In Section 8.3 the question of markedness and non-bi-unique translation is taken up, followed by a contrastive study of belief reports in English and in Polish in Section 8.4. Polish has been selected because it exhibits a lexical/grammatical marking of the de dicto reading of attitude reports in negative sentences. This study confirms the default status of the de re reading, which strengthens the ideas of the intention-based Default Semantics. Section 8.5 assesses the role of this contrastive study for the semantics of belief reports and sums up the findings.

8.1. Belief, Culture, and Translation It has been well acknowledged that faithful, adequate translation includes translating the intentions behind the production of the sentence rather than merely translating the structure and style (cf. Gentzler 1993: 58; de Beaugrande 1980: 291; and, originally, Nida 1964). The equivalence of the beliefs and intentions that underlie a sentence and its translation is essential. But before we use evidence from translation for the semantics of belief reports, we have to accept a theoretical commitment. This is a tenet that propositions can be recognized cross-culturally, or, in other words, we have to accept cultural rationalism and realism based on the tenet that the criteria of rationality do not differ from culture to culture.1 According to the opposite standpoint, that of cultural relativism, people of different cultures live in different worlds, a translation of a sentence would have to consider the 1

This discussion originated in Jaszczolt 1998c.

Section 1

Belief reports in a contrastive perspective

303

differences between these cultural worlds and, in effect, would be impossible. On the other hand, once we have accepted cultural rationalism, we have allowed for language-independent propositions and for pragmatic equivalence based on universal rules of inference captured by the Gricean or post-Gricean theories of implicature. This assumption of rationalism and universalism was defended in Section 2.9 above in the discussion of Sperber's (1985a, b, 1996,1997) idea of the 'epidemiology of representations'. To repeat, Sperber argues that some ideas are 'contagious' and spread more easily than others. Humans are naturally disposed to form certain concepts that subsequently spread across the community. These 'basic concepts' are similar across cultures, they are formed by means of combining ostension with some innate schemata. They are easy to translate. Ultimately they lead to the skill of interpreting utterances through forming hypotheses about meaning (cf. Sperber 1996: 69; 1985b: 81-82). On one end of the spectrum, representations can use these concepts and be transparent to their holder, while on the other they can be symbolic and stored for future understanding. Although the original idea is frequently transformed during the process of transmission due to the differences in memory and assumptions among the interlocutors, the essential point is that it can spread, be understood, and that the same mechanism is followed in all cultures. Generally, people have representations of states of affairs in the form of memory, beliefs, or intentions. Representations that have become public are repeated more often than others, they have become shared beliefs, norms, myths, techniques, or classifications. They can be communicated by a representation that is either a description or an interpretation of this shared belief. Interpretation occurs when, for example, the communicator has to expand and clarify the original belief in order to make it understandable to the hearer.2 So, interpretations of other people's beliefs are possible, they are often speculative when conducted cross-culturally, but they adhere to the principle that people are rational and have rational beliefs. Translation is here the formation of a belief similar to the original one. In other words, "The anthropologist must (...) go beyond mere translation: only then can she hope to understand what she hears, and thus be genuinely able to translate it. She must speculate, synthesize, reconceptualize." Sperber (1996: 39). This 'going beyond' means either generalization to similar phenomena, finding common patterns, making a comparison with a familiar phenomenon which is similar in function, or explaining why this particular idea became 2

This is based on Sperber and Wilson's (1986a: 232) idea of mental representation as an interpretation of a representation or a description of a state of affairs. See their diagram in ibid.

304

Discourse, beliefs, and intentions

Chapter 8

'contagious' (cf. ibid.: 50). Semantic and pragmatic equivalence, discussed in the following section, is founded on such universal schemas for the interpretation of beliefs. Now, since humans can represent not only experience but also other people's representations or mental states, they can represent attitudes towards representations or the lack of understanding of someone's representation. As was discussed in Section 2.9, this metarepresenting accounts for apparently irrational beliefs that spread through the community in the form of mysteries. But it also accounts for the lack of a full understanding of the sentence due to not knowing the identity of the referent. This is very important for de dicto proper belief reports. If this lack of individuating information is lexicalized or grammaticalized, then the potential ambiguity of belief reports becomes language-specific due to its absence in some languages, notably the ones in which such overt marking of the type of reading occurs. The interpretative ambiguity is no longer semantic if semantics is based on universal principles. This languagedependence raises hopes for finding some marking of the readings in languages such as English where the marking of the de re/de dicto distinction is neither lexical nor grammatical. And perhaps it strengthens the thesis of Default De Re. But this is the task for Section 8.4. To sum up, people can hold attitudes towards propositions that they do not fully understand. As objects of attitudes, these propositions allow for differing fineness of detail. In other words, a person may hold a belief about a state of affairs that is expressed in a proposition whose structure is either relevant or irrelevant for the belief. In the case of de dicto beliefs, propositions can be taken as primitive, further undifferentiated objects. De re beliefs are about an entity and its having certain properties. This proposal of the 'sensitivity to structure', based on Cresswell (1985), is applied in Section 8.4. All in all, reporting on beliefs, as well as translating belief reports, may require knowing how the belief was acquired and how it was communicated. The reporter and the translator have to look at the speaker's behaviour in the context of his or her anthropological culture. This is essential in the case of representations of ideas whose sense is not fully understood, as well as representations that are de dicto due to the lack of individuating knowledge about the referent. So, in order to arrive at what is said and what is implicated, world knowledge and the assumption of cultural rationalism are essential. What is said and what is implicated are in principle available cross-culturally. There is no strict boundary between the two due to the pragmatic processes of expansion and completion of the propositional content (cf. Bach 1994a: 270). This fuzziness is not a hindrance. After all, in reporting a belief and translating the report, the speaker tries to preserve what is said. If the hearer can be presumed to have problems with understanding the minimal proposition or the propositional radical that underlies the uttered sentence, then an expansion or completion, respectively, have to be explicitly stated.

Section 1

Belief reports in a contrastive perspective

305

Now, this generalization would be adequate were it not for implicatures. They may have to be communicated explicitly in the report or the translation. For example, in sentence (41) of Chapter 1 repeated here as (1), attaining pragmatic equivalence requires preserving obvious implicatures such as 'there is no need to worry', 'the wound will heal', 'it is not serious', etc. To repeat, the sentence is uttered by a mother to a boy who cut his finger: (1)

You're not going to die, Peter.

While recovering these implicatures is not a problem for reporting on her sentence, it may become a problem for translation to preserve the salience of these implicatures. Here semantic equivalence may have to be compromised in order to preserve the pragmatic one. For example, an equivalent of (2) may have to be uttered: (2)

There is nothing to worry about, Peter.

So, in translation, these two levels of equivalence have to be weighed against each other. The importance of pragmatic equivalence has been widely recognized since at least the 1960s. Nida (1964) and Nida and Taber (1969: 24) talk about dynamic equivalence, measured by the similarity of responses that a sentence and its translation engender in their respective cultures; Gutt (1991: 21-22) talks about translation as communication rather than encoding and decoding messages.3 Now, Nida and Taber allow only linguistically implicit information to become explicit. They exclude cultural background from contributing to translation. But as Sperber's account of cultural rationalism demonstrates, expansions and completions may be required to report on a belief in a different language. After all, the pragmatic processes that are responsible for such embellishments are universal and, through being pragmatic, they are also linguistic. Hence, the earlier constraint has to be abandoned. If beliefs and intentions are to be rendered faithfully, artificial constraints are not helpful. Translation as communication has respectable roots in the philosophy of language, notably in the work of Rudolf Carnap that dates back to the 1940s. Carnap (1942,1947) approached semantic meaning as follows. Intralingually, two sentences are equivalent if both are true or both are not true, and the individual expressions are equivalent, stand for the same thing. Equivalence can be of three types. It can be material, of objects and facts; logical, where the truth of the sentence can be deduced from the semantic rules; and we 3 Krzeszowski (1989) distinguished seven types of equivalence. Pragmatic equivalence, applied to texts, means there evoking 'maximally similar cognitive reactions' (ibid.: 65).

306

Discourse, beliefs, and intentions

Chapter 8

also require the equivalence of intentions, of the way of thinking about objects and eventualities. For this stronger equivalence one requires also the equivalence of structure, i.e. intensional isomorphism. For example, 'three' and 'the square root of nine' are not equivalent in the strong sense because they do not have the same intensional structure. Now, Carnap went wrong in defining belief in that the belief that p was for him a disposition to assent to sentences intensionally isomorphic with p. But although this behaviourist approach failed, intensions survived into contemporary possible worlds semantics and the notion of semantic equivalence. A complete adequacy in translation may not always be attainable. And even when it is, there may not be steadfast rules that govern this practice. The question as to whether to follow semantic, pragmatic, syntactic or stylistic equivalence may not perhaps have an answer; the preferences may change even within one text. But if translation aims at successful communication, semantic and pragmatic equivalence are both relevant. In the same way, in the earlier chapters I took on the claim that semantics and pragmatics are intertwined and a large part of pragmatic processes during utterance interpretation belongs to semantics. In other words, these processes contribute to the semantic representation of the sentence. It is also essential to consider conventions and defaults. Some expansions and completions may proceed through default interpretations, guaranteed by socio-cultural and linguistic conventions, as was discussed in the example of Levinson's (1995) heuristics. This problematic boundary between the two in contrastive studies is the topic to which I now turn.

8.2. Contrastive Semantics and Pragmatics I begin with some terminological remarks. The term 'contrastive analysis' stands for applied contrastive studies that subsume a theoretical linguistic component. This theoretical component gave rise to theoretical contrastive studies (TCS). TCS can be performed on any level of linguistic analysis such as phonology, lexicon, semantics, pragmatics, or text linguistics. Theoretical studies have as the unit of analysis a universal category and as their objective the investigation of how this category is realized in the contrasted languages (see Fisiak et al. 1978; Jaszczolt 1995b: 2). This ensures the equal treatment of both analysed languages. According to TCS, the differences between contrasted languages LI and L2 can be generally allocated to the following three categories: (i) structural contrast (e.g. allowing for or disallowing subjectless sentences); (ii) categorial contrast (e.g. adjective in L! but adverb in L2 is used); or (iii) functional contrast (e.g. subject in L! becomes an object in L2). Languages can also differ in word order. English is SVO, while Polish

Section 2

Belief reports in a contrastive perspective

307

has free order and a rich inflection system. TCS concentrate on the syntactic, semantic and pragmatic contrast. These differences are also exemplified in attitude constructions. In TCS, languages have to be contrasted by means of using a stable tool, a platform of reference for comparison, frequently called tertium comparationis (Krzeszowski 1984,1990:15; Jaszczolt 1995b: 2,1995c: 562; Fisiak 1990). The tertium comparationis is determined by the purpose of contrast. Syntax, semantics, pragmatics, and every other level of analysis have their own criteria of comparison and two expressions may appear similar or different depending on the assumed level. In pragmatics, illocutionary force is frequently contrasted. For example, while in English a compliment normally requires a speech act of thanking as a response, in Polish it often requires self-denigrating. The resulting sentences may be very different semantically and syntactically, and yet create an interesting pair for a comparison because they both constitute equivalent responses to the same type of speech act. Normally equivalence on one level of analysis is investigated, although claims have been made in the literature that these 'equivalences'of different levels may 'add up', so to speak, to an overall equivalence, or, rather, the latter is the function of the first (see Kalisz 1981). In this section I concentrate on the semantic and pragmatic equivalence in translation and the overlap between the two. The first attempt at a contrastive pragmatics may look as follows. The effects an utterance produces in a context should be kept constant between the situations in the two languages, and we should look for differences and similarities in the ways the two contrasted languages achieve these effects. But we would have to know how to pinpoint these effects. Illocutionary effect will not do because a particular speech act may require different types of speech acts as responses in the two contrasted languages, as the example of complimenting demonstrates. Also, if we compare the ways of realizing a particular speech act (cf. Oleksy 1984), then indirect speech acts would be difficult to classify. Moreover, an illocution in one culture can be a perlocution in another (cf. Wierzbicka 1991). Finally, as Sperber and Wilson (1986a: 245) observe, some speech acts can be successful without being classified by the hearer as belonging to a particular type. For example, (3) need not be recognized as a prediction: (3)

The weather will be warmer tomorrow.

(from ibid). Naturally, speech acts need not be cut so finely. Calling it an assertion may suffice instead. But the indeterminacy surrounding the classification of speech acts certainly does not help found the equivalence on their types. Moreover, as a result of the collapse of the performative hypothesis, there is no reliable connection between the sentence and the

308

Discourse, beliefs, and intentions

Chapter 8

speech act type. The literal force hypothesis (see Levinson 1983) also had to be abandoned: the performative verb is no clear indicator of the type of the speech act performed with its help. Instead, pragmatic equivalence will have to be founded on the semantic equivalence. It will have to rely on the propositional form of the utterance, entertained by the speaker with a particular type of attitude so that it resulted in an assertion, a question, a request, an advice, and so on.4 So, it is perhaps best to conclude that utterances are pragmatically equivalent if they communicate the same content. In looking at belief reports, we shall not be interested in metaphors and other non-literal use, and hence for our purpose pragmatic equivalence will presuppose semantic equivalence. The overlap between semantic and pragmatic equivalence can be safely predicted to follow that of semantics and pragmatics. For example, generalized conversational implicatures as from 'three' to 'exactly three' can be regarded either as semantic, as in DRT, or as pragmatic, as in relevance theory. This overlap is an expected and satisfactory outcome. Semantic equivalence is the equivalence of the propositional representation, often equated with what is said, and the propositional representation often requires pragmatic processes, including expansion and completion. Pragmatic equivalence remains as the equivalence of what is implicitly communicated. Although pragmatic processes involved in the production and recovery of implicatures are the same as those contributing to what is said, implicatures are easy to keep apart because they have their own logical forms. Since what is said has been subject to extensive studies in recent years (see Chapter 1 and in particular Turner, ed. 1999), we can simply draw on this research in the semantics/pragmatics interface. It is important to take sides in the semanticization/pragmaticization controversy. Similar to DRT, in Default Semantics the pragmatic processes that contribute to the propositional form are semanticized. And this is the tenet I shall accept in contrastive studies. Now, as is well known, some philosophers of language claim that objective translation is impossible. For Quine (1960) translation is dependent on a theory or 'translation scheme' and since there are many such translation strategies, we are left with the indeterminacy of radical translation.5 So, in translating a sentence, we have to follow one of many strategies, hoping that we have selected the one that is common in society. 4

See Sperber and Wilson's diagram in 1986a: 232. Cf.: "... manuals for translating one language into another can be set up in divergent ways, all compatible with the totality of speech dispositions, yet incompatible with one another." Quine (1960: 27). Nota bene, for Quine this was a theoretical problem and his theory is applicable to the analysis of knowledge. But since Quine does not say that we need a special language for expressing theories. it is not against Quine's intention to utilize his claim for the analysis of discourse. 5

Section 2

Belief reports in a contrastive perspective

309

Then we regard it as the universal strategy, ignoring all the rest. But this would not lead us very far. It seems more adequate to accept a level of language-independent meanings. Quine thinks of beliefs as of attitudes towards a sentence. In other words, they are relative to the particular language of the holder of the belief. Yet language-independent units of meaning, that is propositions, have been widely accepted. Being language-independent, propositions constitute the required platform of comparison, or tertia comparationes, a universal semantic category that is instantiated as sentences in particular languages. Also, Lakoff's preconceptual experiences discussed in Section 6.3.4 suggest that language uses some general, cognitive categorization mechanisms.6 Putnam's stereotypes can also be regarded as constraints on translation. They make translation more reliable and less dependent on the selection of a strategy. Since the stereotypical meaning of a word must be known in order to use the word correctly, there are therefore some limits on the divergence of possible translations. Similarly, there are conventional and default meanings that are, so to speak, extralinguistic and that constrain the translation. I shall return to this point below in comparing English and Polish belief reports. The existence of the independent, underlying level of propositions as objects of attitudes rules out the complications of language dependence. All the translator has to do is to find a sentence in his or her language that expresses the same proposition as the reporter's sentence. But it is not an easy task. Sentences have a logical form that may require further expansion or completion into a propositional form. This prepositional form is the unit of semantic equivalence in translation. In other words, this semantic equivalence cannot be derived from syntactic constructions alone. It has to be text, or even context-bound rather than systematic, where 'systematic' means the comparison of grammatical and lexical systems (cf. Krzeszowski 1989; Jaszczolt 1995b: 2). This conclusion is an instantiation of a general claim that semantics has to incorporate the outcome of some pragmatic processing. All in all, instead of talking about translation, it is more appropriate to use the TCS-term 'contrasting' languages. After all, we investigate how a universal category of the de re/de dicto distinction is instantiated in the two contrasted languages, giving them equal importance and attention.

6

The existence of universal concepts is directly proportional to the extent to which languages are shaped by human nature rather than by culture. See Wierzbicka 1992, 1996, 1997.

310

Discourse, beliefs, and intentions

Chapter 8

8.3. Context and Markedness Quine divides sentences into 'standing' (e.g. 'Snow is white') and 'occasion' (e.g. 'It is snowing'). Occasion sentences seem to demand less effort from the hearer as they are immersed in the current situation that is shared by the interlocutors. If the speaker and the hearer both perceive the object of the speaker's belief, then this object is usually referred to by an identifying description or an indexical expression, sometimes accompanied by an act of ostension. Using a proper name would be redundant, overly informative for this point on the scale of referent accessibility. An act of ostension accompanied by a proper name seems to carry an implicature to the effect that, in addition to predicating something about its bearer, the speaker wants to ascertain the correctness of calling the person by this name. In effect, in the case of a referential mistake, ostension and the name clash. Two vehicles of thought, action and language, point to different persons. This situation can be called double referring. Context, in the form of the act of ostension, makes the use of a demonstrative an unmarked case and the proper name marked, carrying additional implicatures. The two utterances are exemplified in (4) and (5): (4)

That man [act of pointing] is a spy!

(5)

Ortcutt [act of pointing] is a spy!

They can be thought of as modifications to Situation A from Chapter 4. The observation of markedness also agrees with Levinson's (1987) minimization discussed in Section 1.4.1, as well as the relevance-theoretic account of the balance between processing effort and cognitive effect. Now, while pronouns, demonstratives, demonstrative NPs and perceptionbound definite descriptions do not normally engender choices of interpretation, proper names and other definite descriptions do as they can be used, respectively, purely denotatively or attributively. In belief contexts, they may engender (proper) de dicto readings. Here another recourse to markedness is required. Context, in the sense of conventions and intentions, makes the referential Ide re reading unmarked. Translation has to preserve this default status. If it does, then it corroborates it. Polish has two ways of expressing disbelief in the form of negative sentences of belief reports. If one of them is proven to be unmarked, the default, then the reading of the English translation to which it corresponds, i.e. de re or de dicto, is also very likely to be the default.

Section 4

Belief reports in a contrastive perspective

311

8.4. Complementizer That in Contrast 8.4.1. English and Polish: A Surface Overview The main difficulty with analysing the meaning of sentences in reported speech is that oratio obliqua lacks the context the original communication was placed in. The sentence reports on a state or an event, not being a part of it itself. A report is more objective in that it adds the reporter's perspective on the situation, but it also provides less information than direct speech which reflects the feelings, personality and mood of the speaker. Reported speech is usually regarded as depicting the reporter's view on the event and uses expressions from the vocabulary the reporter normally uses. So, for example, if the report is de dicto, then it is the reporter who lacks the individuating information rather than the holder of the belief.7 Moreover, languages can vary as to the degree of discrepancy between direct and indirect speech. One of the factors here can be the number of different complementizers used in attitude contexts where they are not exact synonyms. English has one, Polish has several, as is discussed below. Principles for reporting on attitudes differ from language to language. I shall now discuss the ways they are rendered in Polish and contrast the structures of Polish examples with their English equivalents. Situation B' is the Polish version of Situation B from Chapter 4. Situation B'

'Bernard J. Ortcutt zostatdzisiaj aresztowanyX pod zarzutem szpiegostwa. J

/Alez Ortcutt nie jest szpiegiem \Qn jest uczciwym obywatelem

Ralf wierzy, ze Ortcutt nie jest szpiegiem.

Jan

7

Albeit it is not a steadfast rule. See Chapter 4. These characteristics of reported speech in Polish are discussed in Bak (1977: 465) and Doroszewski and Wieczorkiewicz, eds (1959: 290-291).

312

Discourse, beliefs, and intentions

Chapter 8

The utterances are repeated as (6)-(8) below8: (6)

(7)

(8)

Bernard J. Ortcutt

zosta*

dzisiaj aresztowany

Bernard J. Ortcutt

become-35g Past

today

pod

zarzutem

szpiegostwa.

under

accusation-//wfr

espionage-Gen

Alez Ortcutt nie

jest

But

be-3Sg Pres spy-Instr

Ortcutt not

uczciwym

obywatelem!

honest-Instr

citizen-Instr

Ralf wierzy,

ze

szpiegiem! On jest

Ortcutt nie jest

Ralph believe-3Sg Pres that

Ortcutt

arrest-Paw PPle

He be-JSg Pres

szpiegiem.

not be-3Sg Pres

spy-Instr

When we compare the use of (8) with that of its English equivalent in (9), it turns out that (8) is more restricted in its application. Namely, if Jan had taken Ralf's belief to be de dicto, or, indeed, did not possess individuating information concerning Ortcutt, he would have uttered (10) rather than (8):

(9)

(10)

Ralph believes that Ortcutt is not a spy.

Ralf utrzymuje,

jakoby

Ortcutt nie by*

Ralph maintain-3Sg Pres

as+Subj

Ortcutt

not be-3Sg Past

szpiegiem. spy-Instr

Through its use of subjunctive mood, sentence (10) conveys the attitude of the speaker's distancing himself from the statement. Subjunctive constructions with the verb 'wierzyc' (believe) are introduced below (cf. example 36). As will be proved by the syntactic analysis, the use of the subjunctive in Polish is strong evidence that the attitude is not held about the person referred to in the subject position of the embedded sentence, but rather about the proposition as a whole. The embedded sentence does not have strong links with the main clause. On the contrary, the use of a ze-clause in negative Only the relevant grammatical features have been marked in the translations.

Section 4

Belief reports in a contrastive perspective

313

constructions often indicates the 'acquaintance' with the referent and the de re attitude.9 By uttering (10), the speaker communicates that either (i) he does not share Ralf's belief, or (ii) Ralf does not know who he is talking about, or (iii) the speaker does not know who Ortcutt is, or (iv) the speaker suspects that Ralf is referentially mistaken. In other words, in spite of the newspaper report, Ralf maintains that somebody called Ortcutt is not a spy. This may be a different person than the one meant in the newspaper report, or somebody called Ortcutt and otherwise unknown to Ralf or Jan. In other words, the possibility of either of the de dicto readings (fife dicto proper or de dictoi) is very strong. Negative constructions with the subjunctive, as opposed to those in the indicative mood, tend to express this distancing. In English, this distancing would have to be communicated descriptively, in a more elaborate way as, for example, in (11): (11)

Ralph believes that the person whom he believes to be Ortcutt is not a spy.

There is a strong reason to suppose then that the construction in the indicative mood in (8) normally corresponds to English (9), whereas (10) cannot be faithfully rendered by (9) without a further descriptive comment concerning the distancing from the claim as in (11). Nota bene, (11) has a slightly stronger de dicto flavour than (10). Now, if Jan used a demonstrative or a personal pronoun as in (12), then the subjunctive mood would indicate Jan's detachment from the truth of Ralf's statement rather than the lack of identifying information about the referent. Let us assume that the anaphoric link to the previous text or situation is clear. (12)

Ralf utrzymuje, jakoby on byl szpiegiem. he

Naturally, both (8) and (9) can be used in the de re and de dicto sense. However, the existence of the subjunctive mood signalling detachment and the de dicto sense as in (10) leaves (8) as the natural rendering of 9

To compare, the subjunctive in German, and also in English where it is realized as past tense constructions such as in (i), also seems to signal a lack of individuating information about the referent and the 'nonreferential' use of the VP, thus serving the purpose of creating a distance between the speaker and the uttered statement (see Rundle 1979:166). However, the NP is usually still used referentially. (i) Ralph believed that Ortcutt had seen him before. The lack of information or distancing can be marked on theVP without being marked on the NP. Rundle (ibid.: 185-186) calls it an asymmetry between noun and verb. Where the type of the NP is standardly used referentially (e.g. in the case of personal pronouns and proper names), detachment from this default has to be signalled descriptively.

314

Discourse, beliefs, and intentions

Chapter 8

the de re sense. If (9) is indeed the English equivalent of (8), then its de re sense is the natural, prior, unmarked, or default reading. This suggestion will be strengthened below by the analysis of constructions where the attitude verb is negated. By 'unmarked' I mean two things. Firstly, that constructions of the type A believes that B cps' are normally de re and only in clearly indicated circumstances de dicto (in other words, de re is their unmarked interpretation), and secondly, that they are lexically unmarked as opposed to de dicto that require a descriptive comment. Analogously, Polish constructions in the indicative mood are de re by default and they are grammatically unmarked as opposed to de dicto that requires the subjunctive mood. As this comparison demonstrates, reports on attitudes differ from language to language. Polish is by no means the only language that uses the subjunctive for distancing. The phenomenon has been widely discussed in French (see Fauconnier 1985; Sweetser and Fauconnier 1996: 19). About Spanish, Mejias-Bikandi (1996: 173) says that ".. .when the indicative mood is used instead of the subjunctive mood in a particular context, the speaker assumes some responsibility over, or feels closer to, the content of the proposition expressed by the complement clause." He exemplifies it in (13) and (14): (13)

(14)

Juan no cree que tu hijo esta enfermo. John doesn' t believe that your son \s-Ind sick.

Juan no cree que tu hijo este enfermo. John doesn't believe that your son is-Subj sick.

(from ibid). Mejias-Bikandi explains this phenomenon in terms of Fauconnier's mental spaces as follows. The negation no sets up a counterfactual space M, different from the space of the speaker's reality R. The indicative mood signals the accessibility of information from M to R. The use of the subjunctive, on the other hand, signals that there is no such accessibility. This explanation captures the intuition correctly but does not go far beyond the intuition. The choice of the grammatical mood produces an effect which is said to be ".. .very difficult to state explicitly, but which can be informally characterized as the speaker feeling closer to the propositional content of the clause in the indicative mood." (ibid.: 175).

Section 4

Belief reports in a contrastive perspective

315

In order to go further beyond intuitions in the analogous Polish examples, I analyse their structure in terms of a lambda- categorial language and discuss them in terms of the well established de re/de dicto distinction. I shall now use the distinction made in Polish negative constructions and compare their structure with the structure of the English belief report in order to look for a possible unmarked interpretation and, in practice, further test the Default De Re principle. This contrasting exercise has to be so designed as to allow for the information from vehicles of thought other than language to be included in utterance interpretation. Moreover, languages differ as to the degree of discrepancy between direct and indirect speech. In other words, the boundary between the extensional and intensional contexts is not sharp. In Polish, the complementizers are not all exact synonyms. Some of them convey additional information as compared with the basic complementizer that. Whereas Polish makes use of a few variants (ze, zeby, aby, azeby, iz, by, jakoby...), there is only one lexical equivalent 'that' in English and the additional information has to be added descriptively. On the other hand, to some extent, the differences between the Polish complementizers are regional variations or historical variations. For instance, iz is a more archaic version of ze, intersubstitutable with it in nearly all contexts. All these forms are still used in contemporary Polish. The complementizers aby, zeby, jakoby, by contain the subjunctive mood marker 'by'. I shall say more below about the types of clauses they introduce. It is also worth noticing that certain English sentences of this type do not seem to allow for the omission of 'that'. Notably, it happens when the main clause and the subordinate clause contradict each other, i.e. one seems to deny what the other affirms. While (15) is acceptable, (16) and (17) are not: (15)

I believe he'll do it.

(16)

*I disbelieve he'll do it.

(17)

??It's /wconceivable he meant it.

(from Bolinger 1972: 40, my emphasis). Bolinger observes that negative sentences of this type require the presence of the complementizer 'that'. This phenomenon, if, indeed, the intuitions are correct, seems comparable with using the subjunctive in other languages in order to signal detachment such as Polish 'ze + by' complementizer that includes the mood marker 'by' and is used only in negative constructions, as is discussed below. However,

316

Discourse, beliefs, and intentions

Chapter 8

incorporated negation in English does not have as strong an effect as the subjunctive in Polish. Also, to some native speakers, (16) is unacceptable even with 'that' inserted, while (17) seems correct.10 Belief constructions in English and in Polish exhibit a number of structural differences. Some of these differences signal that the links between the embedded sentence and the main clause are weaker in Polish than in English. According to Kalisz (1981: 110), Polish ze-clauses have a more independent clause status and a looser clause union with the main clause. Let us compare and contrast some of the grammatical characteristics. Firstly, Polish ze- clauses occur more often than their English counterparts because there are no synonymous constructions. Sentences (18), (19), and possibly also (20) have only a f/za/-clause equivalent in Polish as in (21) since there is no ACI construction in Polish and since (22), the literal translation of (21), sounds awkward: (18)

I believe Ortcutt to be a spy.

(19)

I believe that Ortcutt is a spy.

(20)

I believe in Ortcutt's being a spy.

(21)

Wierze,

ze Ortcutt

jest

Believe-7Sg Pres

that Ortcutt

be-3Sg Pres

?Wierze

w

Ortcutta bycie szpiegiem.

Believe-/5g Pres

in

Ortcutt-Gen being spy-Instr

(22)

szpiegiem. spy-Instr

In other words, the content of the that-c\ause cannot be rearranged in the sentence. Next, in Polish, there is a 'zeby' complementizer. It belongs to the category of complementizers that introduce the subjunctive mood. It is composed of 'ze', a clause complementizer, and 'by', the mood marker. It is followed by an infinitive only in purpose clauses and adjective phrase complements. In belief constructions, where 'zeby' occurs only in negative sentences, it is followed by a clause with the subjunctive verb form as in (23). The fact 10

Pointed out to me by Billy Clark (personal communication). In fact, the rules are not strict here. Cf.: "Other factors interfere. Frequency might well disqualify disbelieve even if the sense of the word did not." Bolinger (1972: 41). Moreover, if the main verb means astonishment and the thai-clause expresses a high degree of astonishment,'that'can be omitted as in (i). Weak astonishment does not allow for this omission, as exemplified in (ii). (i) It's inconceivable he even looked at her. (ii) *It's inconceivable he saw the picture. (from ibid.: 41).

Section 4

Belief reports in a contrastive perspective

317

that the subjunctive ranges over both the complementizer and the verb of the embedded clause makes the that-clause a closely integrated unit. It is an optional form of (24): (23)

Nie wierzg, zeby Not believe-/^ Pres that+Subj

Ortcutt byf Ortcutt be-3Sg Past

(24)

Nie wierzg, ze Ortcutt jest Not believe-lSg Pres that Ortcutt be-3Sg Pres

szpiegiem. spy-Instr

szpiegiem. spy-Instr

There is also a difference in topicalizing the complement. There is no passive of 'believe' in Polish and consequently no equivalent of (25) as attempted in (26). At best, a subjectless reflexive construction is used, as in (27): (25) (26)

Ortcutt is believed to be a spy (by Ralph). *Ortcuttjest Ortcutt

(27)

wierzony

bye szpiegiem przez Ralfa.

be-3Sg Pres believe-Pas? PPle be-Inf spy-Instr

by

Ralph-Ace

?Wierzy sie, ze Ortcutt jest szpiegiem. Believe-JSg Pres Reft that Ortcutt be-3Sg Pres spy-Instr It is believed that Ortcutt is a spy.

The complement NP can be topicalized in Polish by moving its subject to the initial (not subject) position of the main clause as in (28) and (29): (28)

O Ortcuccie Ralf myslal, About Ortcutt-Loc Ralph think-3Sg Past About Ortcutt Ralph thought that he is a spy.

ze jest szpiegiem. that be-3Sg Pres spy-Instr

(29)

To o Ortcuccie Ralf myslal, ze jest szpiegiem. Dem It is about Ortcutt that Ralph thought that he is a spy.

Sentence (29) implies contrast. It implies that it was Ortcutt, and not, for instance, Smith who was taken to be a spy. These sentences use the verb 'myslec' (to think) rather than 'wierzyc' (to believe). In the case of 'wierzyc' the topicalization would look much more awkward. Moreover, English (30) has two possible translations into Polish as in (31) and (32): (30)

John believes that Ortcutt is not Ortcutt.

(31)

Jan wierzy, ze Ortcutt John believe-JSg Pres that Ortcutt

(32)

Jan

wierzy,

John

believe-3Sg Pres that Ortcutt

nie jest Ortcuttem. not be-3Sg Pres Ortcutt-Instr

ze Ortcutt to

nie Ortcutt.

Dem not Ortcutt

318

Discourse, beliefs, and intentions

Chapter 8

Sentence (31) is more adequate than (32) since (32) is more likely to have the contradictory reading. It can be argued that the requirement of the inflected form also makes the clause more closely-knit. Finally, in English the complementizer 'that' can nearly always be omitted, except for the contentious constructions as in (16) and (17) and when it occurs at the beginning of the sentence introduced by 'the fact (that)...'. In Polish, on the other hand, it can be omitted only before the 'by' particle as in (33) where 'by' remains to function as the full complementizer. Nota bene, 'jakoby' does not undergo a similar division. Also, the parenthetic construction in (34) is very uncommon: (33)

Nie wierze, fzejby Ortcutt by! szpiegiem.

(34)

jest szpiegiem. Ortcutt, wierze, Ortcutt believe-7Sg Pres be-3Sg Pres spy-Instr

This last observation indicates more clearly than the others that the Polish ze-clause is not as strongly connected with the main clause as the English that- clause.11 The use of a comma before the complementizer in Polish and the lack of the sequence of tenses seem to point to the same conclusion. While these observations allow one only to speculate informally about the strength of the clausal links, the analysis of the structure in Section 8.4.4 corroborates it more formally. This is where the role of the subjunctive fits in. The subjunctive in the Polish zeby-clause makes it less permeable than the ze-clause from the position of the subject of the main clause. This means that the attitude is not held about an individual and its having certain properties but instead it is likely to be de dicto. On the other hand, belief reports in the indicative mood express a belief de re, about a particular individual. In other words, the constituents of the embedded sentence in (35) are relevant for the content of the attitude, while in the case of the subjunctive clause as in (36), the speaker, or the holder of the belief, does not have to know who Ortcutt is.12 The clause is taken as the content of the belief without being broken down into constituents. This is an instance of the de dicto proper: (35)

Ralf nie wierzy, ze Ortcutt jest szpiegiem.

(36)

Ralf nie wierzy, zeby Ortcutt byt szpiegiem.

Sentences (35) and (36) will now be analysed as typical examples of reports in Polish, with their single English equivalent in (37): 11

For more arguments see Kalisz 1981. 'The speaker or the holder of the belief, because the report can render the lack of individuating information on the part of either, as was discussed in Chapter 4. 12

Section 4

(37)

Belief reports in a contrastive perspective

319

Ralph does not believe that Ortcutt is a spy.

Since (36) does not have an affirmative counterpart in the subjunctive mood, we shall confine the discussion to negative sentences. Sentences (35) and (8) will be treated as interchangeable, expressing the same meaning, although (8) may be felt by some native speakers to be somewhat stronger as an assertion and more clearly de re. Now, the presence of 'zeby' and its other subjunctive equivalents in Polish clearly demonstrates why compositionality of belief reports is such a problem. There are two embedded clauses, as in (38) and (39), and the choice is meaningful: (38)

ze Ortcutt jest szpiegiem

(39)

zeby Ortcutt byl szpiegiem

Both forms are widely used and both allow for choices of interpretation. However, there is a strong tendency for (38) to be used in de re reports, and (39) in de dicto. Since there is no steadfast rule for their use and no bi-unique match of form and meaning, the problem with compositionality remains and is only brought to the surface by the fact that both (38) and (39) correspond to a single construction (40) in English: (40)

that Ortcutt is a spy

Let us now try to list some of the possible translations of the subjunctive construction (36). In English, the de dicto reading is normally rendered descriptively as in (41)-(44): (41)

Ralph does not believe that somebody called Ortcutt is a spy.

(42)

Ralph does not believe that a man called Ortcutt is a spy.

(43)

Ralph does not believe that it is Ortcutt who is a spy.

(44)

Ralph does not believe that Ortcutt could be a spy.

While (41) and (42) are strongly de dicto, (43) signals a referential mistake or at least that Ralph believes that someone else is a spy. (44), the closest in grammatical form to Polish (36), is, however, neutral as to the degree of referring, signalling distancing from the proposition for whatever reason. In English, it is not the complementizer that triggers the differences. Sentences (41) to (44) are certainly not as similar to one another in form and content as are Polish (35) and (36).

320

Discourse, beliefs, and intentions

Chapter 8

8.4.2. Degree of Clausiness To recapitulate, Kalisz (1981: 22) suggests that the relation between the main and the subordinate clauses is stronger in English than in Polish. The complementizer can be more easily deleted in English, the sentence can be passivized, there is a sequence of tenses, and there is only one main way of expressing belief reports in Polish, i.e. in the form A believes that B cps'. Kalisz (1981: 110) describes the characteristics of Polish ze-clauses as (i) and (ii): (i) higher degree of clausiness, i.e. more independent clause status, and thus (ii) looser degree of integrity of the sentence, i.e. looser clause union with the main clause. This strong clausiness is even more apparent in zeby-clauses, producing a closed entity hardly undergoing the de re reading. Although ultimately the readings are achieved with the help of context including intentions and the output of other vehicles of thought, there are strong tendencies towards one or the other reading inherent in the two Polish grammatical constructions (35) and (36). Needless to say, the fact that 'believe' is a non-factive verb makes these degrees of certainty (i) grammaticalized and lexicalized in Polish, and (ii) accounted for descriptively in English. That is not to say that e.g. a factive verb 'know' does not allow for choices of interpretation, but these choices have little in common with the semantics of the relevant expressions. The exercise of comparison is simple and need not be discussed here.

8.4.3. A Lambda-Categorial Language and Sensitivity to Structure In order to capture the possible interpretations of belief reports, I shall use categorial grammar and in particular a /a wfo/a-categorial language presented in Cresswell 1985. The choice of the framework is dictated by the fact that categorial grammar clearly marks the derivational history of its categories and hence is best suited for discussing compositionality of the analysed constructions. The grammar used here is relatively unsophisticated in comparison with more recent developments (cf. e.g. Moortgat 1988 on the Lambek calculus) but it suffices for our purpose. Sentences, a syntactic category, are equal in this approach with'well-formed formulae' (wffs). The corresponding semantic category to a sentence/wff is a proposition which is marked by the number '0'. So, '0' is used for sentences since propositions are their intensions. Building the categorial language proceeds as follows.'Not' is a functor that makes a sentence (0) out of another sentence (0) and thus its category is 0/0; 'and' makes a sentence out of two

Section 4

Belief reports in a contrastive perspective

321

sentences and its category is 0/00. A name is "an expression whose semantic function will be to do no more than denote something" (Cresswell 1985: 96), it denotes things and is of category 1. In Cresswell's semantics, senses or MoPs are not allowed. This view of names can only be made compatible with the place we assigned earlier to MoPs thanks to Cresswell's idea of the 'sensitivity to structure'. Although an expression is directly referential, it need not act as such on the particular reading of the sentence. Some of the structural distinctions can be 'ignored', so to speak, and opacity may ensue. Categories 0 and 1 are basic; other categories are derived. For example, one-place predicates (e.g. 'sleep') are assigned 0/1 since they make a sentence out of a name. In order to make complex predicates out of sentences, the lambdaabstraction mechanism is needed. Cresswell (1985: 97) gives the following example. The formula in (45) is an abstract of 'is an x such that x does not whistle', where x is an individual variable: (45)

< 1, x, <not, <whistles,

By forming such complex predicates Cresswell tries to get rid of the difficulties with intensional contexts.13 His categorial language falls in the tradition of Montague's intensional logic, although the language of the semantic representation is simpler here and his sense/reference distinction is idiosyncratic. He says that the whole semantics is referential, there are no senses in semantics. Reference of a linguistic expression is an intension, and senses are complex structures of these intensions. The semantics is compositional, taking these references into account. As a result, all prepositional attitude reports are structurally ambiguous. He achieved this by postulating that the complementizer 'that' belongs to different syntactic categories. These categories differ in the amount of the structure of the embedded proposition considered in a given sentence. Postulating different 'thats' stands in contrast to Russell's view that it is the predicate 'believe' that is ambiguous. Cresswell founds this proposal on the observations that (i) an attitude (belief) is a relation that includes the meaning, i.e. interpretation, of the sentence to which the believer stands in this relation; and (ii) meaning 13

Cf. also: "... a X- categorial language is to test a semantic theory, in the sense that if the theory is to be plausible, it must be formalizable in such a language." Cresswell (1985: 98), and "In using a /I-categorial syntax we are not claiming that the syntax of a natural language works in just this way...". Cresswell and von Stechow (1982: 516).

322

Discourse, beliefs, and intentions

Chapter 8

depends on the structure of the sentence.14 In other words, attitude reports are structurally ambiguous and this ambiguity is engendered by the complement clause rather than by the attitude verb.15 Using CresswelPs categorial language does not mean that we shall follow all of his theoretical assumptions. Instead, I call these readings semantic possibilities of interpretation since there is nothing overt in the structure of expressions of this type that would warrant the strong claim of structural ambiguity. Moreover, there is no way of representing de dicto/, the reading with the referential mistake, as a separate structure, different from de re and de dicto proper. Hence, the degrees of intentions do not strictly match the degrees of the sensitivity to structure. Translating degrees of reference into degrees of sensitivity has its limitations. I believe that this difference is not significant for the present analysis, although it is not merely terminological either. By allowing in MoPs and information from other vehicles, I preclude the possibility of calling the differences of interpretation structural, as well as do without postulating ambiguities. All in all, the choices of interpretation can be called 'structural ambiguity' only in a broad understanding of the terms. There is no essential difference in the structure between the de re and de dicto readings, as is shown in the tree diagrams in Section 8.4.4. The only difference is caused merely by 'ignoring', so to speak, the relations between the constituents of the embedded sentence in the de dicto case, and taking them into consideration in de re. In other words, the relations between the constituents differ only with respect to the 'sensitivity' of the complementizer to the structure of the following clause. Nota bene, the structure as such remains the same in both readings. Whether this makes the alleged ambiguity syntactic is rather contentious and, certainly, depends on theoretical commitments. Whether this makes it an ambiguity is a separate story, answered negatively throughout Chapters 1-7. Diagrams in the following section demonstrate how this sensitivity to structure works in practice and how it supports the Default Semantics of belief reports. Now, that-clauses are assigned category 1, they are regarded as names. The intuition behind this move can be captured in sentences (46) to (48): (46)

Mary believes what John told her.

(47)

Mary believes what she is expected to believe.

14

See also Cresswell 1980 and Cresswell and von Stechow 1982 on the topic-focus distinction. Cresswell (1985: 105) says that V(says), where 'V is a meaning assignment, is a function from DI x DI into DO, where 'D' is a domain. In other words, it is a two-place operation. The first argument is an entity capable of saying something and the other is 'anything capable of being the meaning of a thai-clause" (ibid.: 106). The meaning of say can operate on the intension of the complement sentence, but that can also be sensitive to the structure of the complement sentence (cf. ibid.: 114). 15

Section 4

(48)

Belief reports in a contrastive perspective

323

Mary believes something.

(cf. Cresswell 1985: 29). The complementizer that turns the embedded sentence into a NP, i.e. into an expression of the same syntactic category as 'something', 'what John told her', etc. In other words, that is a 'name-forming operator'. In the simplest case, where the structure of the embedded proposition is not relevant, 'that' is of category 1/0. It is marked 'that0' because it operates on the sentence as a whole. So, ''are used for senses of expressions. For example, the structure is the sense of the expression '5 + 7', while 12 is the reference. Cf. Cresswell (1985: 26). To repeat, Cresswell's terminology differs from Frege's sense/reference distinction. Reference is an intension, and senses are structures of intensions. In Cresswell (ibid.: 86-87), the intension of 'is a spy' is marked cois a spy; the intension of 'not' is (On0t. The most sensitive 'that' takes the intension