New Waves in Philosophy of Language

New Waves in Philosophy of Language

New Waves in Philosophy Series Editors: Vincent F. Hendricks and Duncan Pritchard Titles include: Otavio Bueno and Oystein Linnebo (editors) NEW WAVES IN PHILOSOPHY OF MATHEMATICS Boudewijn DeBruin and Christopher F. Zurn (editors) NEW WAVES IN POLITICAL PHILOSOPHY Vincent F. Hendricks and Duncan Pritchard (editors) NEW WAVES IN EPISTEMOLOGY Yujin Nagasawa and Erik J. Wielenberg (editors) NEW WAVES IN PHILOSOPHY OF RELIGION Jan Kyrre Berg Olsen, Evan Selinger and Søren Riis (editors) NEW WAVES IN PHILOSOPHY OF TECHNOLOGY P.D. Magnus and Jacob Busch (editors) NEW WAVES IN PHILOSOPHY OF SCIENCE Thomas S. Petersen, Jesper Ryberg and Clark Wolf (editors) NEW WAVES IN APPLIED ETHICS

Sarah Sawyer (editor) NEW WAVES IN PHILOSOPHY OF LANGUAGE Kathleen Stock and Katherine Thomson-Jones (editors) NEW WAVES IN AESTHETICS Forthcoming: Jesús Aguilar, Andrei A. Buckareff and Keith Frankish (editors) NEW WAVES IN PHILOSOPHY OF ACTION

Thom Brooks (editor) NEW WAVES IN ETHICS Allan Hazlett (editor) NEW WAVES IN METAPHYSICS Nikolaj Pedersen and Cory Wright (editors) NEW WAVES IN TRUTH Future Volumes New Waves in Philosophy of Mind New Waves in Meta-Ethics New Waves in Formal Philosophy New Waves in Philosophy of Law

New Waves in Philosophy Series Standing Order ISBN 978–0–230–53797–2 (hardcover) Series Standing Order ISBN 978–0–230–53798–9 (paperback) (outside North America only) You can receive future titles in this series as they are published by placing a standing order. Please contact your bookseller or, in case of difficulty, write to us at the address below with your name and address, the title of the series and the ISBN quoted above. Customer Services Department, Macmillan Distribution Ltd, Houndmills, Basingstoke, Hampshire RG21 6XS, England

New Waves in Philosophy of Language Edited by

Sarah Sawyer University of Sussex, UK

Selection and editorial matter © Sarah Sawyer 2010 Chapters © their individual authors 2010 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No portion of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, Saffron House, 6-10 Kirby Street, London EC1N 8TS. Any person who does any unauthorized act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The authors have asserted their rights to be identified as the authors of this work in accordance with the Copyright, Designs and Patents Act 1988. First published 2010 by PALGRAVE MACMILLAN Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited, registered in England, company number 785998, of Houndmills, Basingstoke, Hampshire RG21 6XS. Palgrave Macmillan in the US is a division of St Martin’s Press LLC, 175 Fifth Avenue, New York, NY 10010. Palgrave Macmillan is the global academic imprint of the above companies and has companies and representatives throughout the world. Palgrave® and Macmillan® are registered trademarks in the United States, the United Kingdom, Europe and other countries. ISBN: 978–0–230–22436–0 hardback ISBN: 978–0–230–22437–7 paperback This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental regulations of the country of origin. A catalogue record for this book is available from the British Library. A catalog record for this book is available from the Library of Congress. 10 9 8 7 6 5 4 3 2 1 19 18 17 16 15 14 13 12 11 10 Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne

For my son, Marcus, and his love of language.

This page intentionally left blank

Contents Series Editors’ Preface

viii

Notes on the Contributors

ix

Introduction Sarah Sawyer

1

1 Vagueness and Non-Indexical Contextualism Jonas Åkerman and Patrick Greenough 2 Semantics and the Place of Psychological Evidence Emma Borg 3

Naturalism in the Philosophy of Language; or Why There Is No Such Thing as Language John Collins

8 24

41

4 Referring When Push-Comes-to-Shove Kevan Edwards

60

5

87

Semantic Normativity in Context Anandi Hattiangadi

6 Literal Force: A Defence of Conventional Assertion Max Kölbel

108

7

A Plea for Understanding Guy Longworth

138

8

From the Expressive to the Derogatory: On the Semantic Role for Non-Truth-Conditional Meaning Stefano Predelli

9

Analyticity in Externalist Languages Gillian Russell

164 186

10 The Modified Predicate Theory of Proper Names Sarah Sawyer

206

11

226

Introduction, Transmission, and the Foundations of Meaning Jeff Speaks

12 Linguistic Puzzles and Semantic Pretence James A. Woodbridge and Bradley Armour-Garb

250

Index

285 vii

Series Editors’ Preface New Waves in Philosophy Series The aim of this series is to gather the young and up-and-coming scholars in philosophy to give their view of the subject now and in the years to come, and to serve a documentary purpose, i.e., ‘this is what they said then, and this is what happened’. It will also provide a snap-shot of cutting-edge research that will be of vital interest to researchers and students working in all subject areas of philosophy. The goal of the series is to have a New Waves volume in every one of the main areas of philosophy. We would like to thank Palgrave Macmillan for taking on this project in particular, and the entire New Waves in Philosophy series in general. Vincent F. Hendricks and Duncan Pritchard Editors

viii

Notes on the Contributors Jonas Åkerman is a PhD student in Theoretical Philosophy in the Department of Philosophy at Stockholm University. His main research interests are in the philosophy of language. He has a particular interest in vagueness, context sensitivity, relativism, indexical reference, and speaker intentions. Brad Armour-Garb is an Associate Professor of Philosophy at the University at Albany-SUNY and a Fellow of Wolfson College, Oxford. He has written extensively in the philosophy of language and logic, with special attention to theories of truth and to the liar, and other semantic paradoxes. Emma Borg is a Professor of Philosophy at Reading and has published widely on issues at the semantics/pragmatics interface, including a monograph, Minimal Semantics (Oxford University Press, 2004), on the debate between minimalism and contextualism. In 2006 she was awarded a Philip Leverhulme Prize for ‘outstanding young researchers’ and is currently working on a second book, provisionally entitled Minimalism and Minds, which explores the relationship between minimalism in semantics and claims about the mind and its place in the world. John Collins is a Senior Lecturer in Philosophy at the University of East Anglia. His principal research is in the philosophy of language and mind, especially the results and methods of generative linguistics, the concepts of truth and representation, and the status of naturalisation programs. He is currently working on a monograph on the ‘unity problem’. He is the author of Chomsky: A Guide for the Perplexed (Continuum, 2008). Kevan Edwards is an Assistant Professor at Syracuse University. He works on issues in philosophy of mind and philosophy of language, in particular areas where the two fields overlap. His present philosophical obsession is exploring the prospects for a reference-based approach to concept individuation, which he hopes to ground in a naturalistic theory of representation, and connect to language via a use-theoretic approach to linguistic communication. Patrick Greenough is a Senior Lecturer in the Department of Philosophy at the University of St. Andrews and an Associate Fellow of Arché. His main research interests are in the philosophy of logic, the philosophy of language and epistemology. He has a particular interest in vagueness, the liar paradox, ix

x

Notes on the Contributors

indeterminacy, contextualism, relativism, minimalism, realism, truth, antiluck epistemology, self-knowledge, basic knowledge, discrimination, assertion, belief, representation, and the nature of time. Anandi Hattiangadi is a Lecturer in Philosophy at Oxford University and Fellow of St Hilda’s College. She has been a visiting Fellow at the Swedish Collegium for Advanced Studies, and a Research Fellow at Trinity College, Cambridge. She is the author of Oughts and Thoughts: Rule-Following and the Normativity of Content (Clarendon Press, 2007), along with various articles in philosophy of language and mind. Max Kölbel is ICREA Research Professor in the Department of Logic, History and Philosophy of Science at the University of Barcelona. He works primarily in the philosophy of language, epistemoloy, and metaethics and has published several articles in these fields as well as the monograph Truth Without Objectivity (Routledge, 2002). Guy Longworth is a Senior Lecturer in Philosophy at the University of Warwick. He worked previously at Birkbeck College and at University College, London. He works mainly on issues in the Philosophy of Language and the Philosophy of Linguistics, often at their intersections with Epistemology and the Philosophy of Mind. He is co-editor, with Jennifer Hornsby, of Reading Philosophy of Language (Blackwell, 2005). Stefano Predelli is a Professor of Philosophy at the University of Nottingham. He works in philosophical semantics, with particular interest in indexicals, proper names, quotation, and non truth-conditional meaning. He is the author of Contexts (Oxford University Press, 2008). Gillian Russell is an Assistant Professor of Philosophy at Washington University in St Louis. She received her PhD from Princeton University in 2004 and was a Killam Postdoctoral Fellow at the University of Alberta (Canada) in 2005, a visiting fellow at Berkeley in the spring of 2009, and will be taking up a three month fellowship at the Tilburg Center for Logic and the Philosophy of Science in the Netherlands in the fall of 2009. She is also a frequent visitor to Australia, especially the University of Melbourne. Her main research interests are in the philosophy of language and logic, and epistemology. She is the author of the book Truth in Virtue of Meaning: A Defence of the Analytic/Synthetic Distinction (Oxford University Press, 2008). Sarah Sawyer is a Senior Lecturer in Philosophy at the University of Sussex. She is interested in topics that connect philosophy of language, philosophy of mind, epistemology and metaphysics, especially anti-individualism, de re thought and the semantics of fiction.

Notes on the Contributors xi

Jeff Speaks is an Assistant Professor of Philosophy at the University of Notre Dame. In the philosophy of language, he is interested in the foundations of semantics, the semantics/pragmatics distinction, and the metaphysics of propositions, as well as a variety of issues in natural language semantics. He has also worked in the philosophy of perception, and is interested in topics at the intersection of the philosophy of mind and the philosophy of language. James A. Woodbridge is an Assistant Professor of Philosophy at the University of Nevada Las Vegas. He has published several papers in philosophy of language and philosophical logic, in particular on deflationary theories of truth, semantic pretense, dialetheism, and semantic pathology.

This page intentionally left blank

Introduction Sarah Sawyer

The recent path and influence of philosophy of language can be traced back to its origins in logical positivism and the subsequent demise of logical positivism at the hands of Quine.1 Logical positivism aimed to make philosophy scientific, and tried to do so by appeal to an empiricist theory of knowledge grounded in a verificationist theory of meaning. Although the theory of meaning was inherently problematic, the rigour, clarity and respect for argument that the logical positivists advocated had a profound and longlasting effect on philosophy as a discipline. The positivist movement, influenced by Frege through Russell, Carnap, and Wittgenstein, propagated the view that linguistic meaning (rather than first principles or the nature of concepts) was the proper starting point of philosophy, and there was a sharp break from philosophy’s historical legacy. By the 1950s the so-called linguistic turn was firmly entrenched.2 The focus on linguistic meaning was taken up in two distinct ways. The first, derived from Frege, modelled linguistic and philosophical investigation after science, logic and mathematics, and as such championed theory over intuition. The second, derived from Moore, looked instead to ordinary practice as the source of linguistic and philosophical insight, and eschewed theory in favour of everyday examples. But it was only when philosophers began to draw on both traditions in the 1960s and 1970s that philosophy of language came into its own. Thus theory was brought to bear on everyday examples in such a way that philosophical conclusions could be derived from the examples which had hitherto been unanchored, and aspects of ordinary language such as vagueness, ambiguity, indexicality, singular reference, implicature, and intensionality, which had hitherto been ignored as irrelevant to science, could be analysed within a rigorous theoretical framework. As such, issues associated with logical form, meaning and reference came to prominence, and work on these issues then paved the way for a wealth of more recent developments in the philosophy of language. Some of the most notable contributions include Strawson’s work on referring and presupposition,3 Quine’s work on language, logical form and ontological 1

2

Sarah Sawyer

commitment,4 Davidson’s work on truth and meaning and more generally his approach to language through logical form, inherited from Quine,5 Dummett’s work on meaning and understanding,6 Grice’s work on meaning and communicative intentions and on conversational implicature;7 and the development of theories of reference that can be traced from Russell’s theory of definite descriptions, through the progressively more radical suggestions of Searle and Strawson, Donnellan, and Kripke and Putnam to the claim that reference is not determined by the psychological states of speakers but depends in part on causal connections between speakers and the things to which they refer.8 Direct reference theories and semantic externalism have been the result.9 But the history of the philosophy of language is also the history of analytic philosophy and in all these issues there is a clear debt to Frege, who is now widely recognized as the father of analytic philosophy.10 This volume is a collection of new research articles by some of the most innovative young philosophers currently working in philosophy of language. Given the rich legacy in the philosophy of language, it is not surprising that the views of the contributors are shaped by it and located within it. Indeed, the issues mentioned before all appear in one form of another throughout the book. And yet the variety of issues covered and the originality of the individual contributions demonstrates the healthy state of philosophy of language now, enriched by developments in psychology and linguistics, as well as by developments in other areas of philosophy. I will provide here an overview of the contributions to the volume, but beyond that I will let the authors speak for themselves. The chapters in the volume are ordered alphabetically by author, and that order is followed here. The volume opens with a chapter by Jonas Åkerman and Patrick Greenough who examine the currently popular contextualist response to vagueness. They begin by distinguishing content context-sensitivity from truth context-sensitivity, which distinction yields a taxonomy of contextualist responses to vagueness: the claim that the proposition expressed by a sentence containing a vague predicate depends on the context of utterance yields forms of indexical contextualism concerning vagueness; whereas the claim that the truth of a sentence containing a vague predicate depends on the context of utterance yields non-indexical contextualism concerning vagueness. Åkerman and Greenough go on to argue that the most promising contextualist solution to the problem of vagueness is non-indexical contextualism, as this fares better in all relevant respects than any form of indexical contextualism about vagueness. In her contribution to the volume, Emma Borg explores the relationship between semantics and psychology. Her particular interest lies with minimal semantics which, like Gricean semantics, assumes that semantic facts depend in some way on psychological facts. This metaphysical dependence of semantics on psychology implies a corresponding epistemic dependence, according to which psychological evidence is relevant to semantic

Introduction 3

theorizing. However, minimal semantics, again like Gricean semantics, seems not to accord with the relevant psychological evidence. In particular, minimal propositions seem not to be available to subjects and hence seem to be unsuited to playing the role of semantic content. Borg aims to defend minimalism against this charge by construing it as a theory of semantic content that underlies the ability to grasp speaker meaning rather than a theory of conscious content. Thus, according to Borg, minimalism does not specify the content a competent language user in fact recovers when grasping speaker meaning; rather, it specifies the content a competent language user can recover if psychological resources are not diverted from processing literal meaning; but the psychological processes realizing the semantic theory are, she claims, sometimes stopped, for one reason or another, before they deliver sentence-meaning, and this accounts for the psychological evidence that purportedly tells against minimalism. Minimalism, then, is consistent with the psychological evidence. John Collins, in his contribution to the volume, advocates a naturalistic approach to language. He begins by distinguishing a metaphysical construal of naturalism from a purely methodological construal of naturalism. The metaphysical construal, he argues, is to be rejected, while the methodological construal is to be endorsed. The principal claim is that the various naturalist programs that concern mental phenomena, linguistic phenomena, epistemic phenomena, and so on require neither a conception of a fixed naturalizing base nor a fixed conception of what is to be naturalized. Rather, proper scientific inquiry will settle both issues. Finally, Collins argues, on the basis of his understanding of methodological naturalism, that naturalistic inquiry into language in fact has no need for the notion of a language. The idea of an idiolect and the idea of a shared public language may be coherent, but they are both explanatorily redundant: the notion of a language as a non-cognitive entity plays no role in linguistic explanation. Kevan Edwards is concerned in his contribution to defend direct reference theories against the so-called threat of collapse. Direct reference theories are motivated by Kripkean intuitions concerning the modal and epistemic profile of certain expressions, but have to appeal to something like pragmatically imparted descriptive information or guises in order to accommodate broadly Fregean intuitions concerning apparent differences in the semantic content of co-referential expressions. The threat of collapse arises in the realization that any attempt on behalf of a direct reference theory to accommodate Fregean intuitions is fated to conflict with the Kripkean intuitions that motivates it. Edwards argues that direct reference can survive the threat if embedded in a linguistic pragmatist framework. Within such a framework Fregean intuitions can be understood as stemming from facts about what speakers intend to express or communicate by their use of the relevant expressions, and Kripkean intuitions can be understood as stemming from the literal meaning of the expressions in question. Crucial to the strategy is

4

Sarah Sawyer

the claim that literal meaning is what is communicated ‘when push comes to shove’, and that this is directly referential. When speakers perceive a threat to communicative success they are disposed to retreat to uses of language that are increasingly literal, uses that demonstrate an increased sensitivity to an expression’s epistemic and modal profile. A pragmatist conception of language, then, has the resources to account both for Fregean intuitions and for Kripkean intuitions, and the threat of collapse is disarmed. The normativity of meaning is the subject of Annandi Hattiangadi’s contribution to the volume. On a widely held view of the normativity of meaning, the semantic correctness of an utterance depends on whether the sentence uttered is true, given its literal, linguistic meaning. Hattiangadi draws on context-shifting, incompleteness, and inappropriateness arguments from the literature on semantic contextualism to argue that this account of semantic normativity fails to cohere with our intuitions about the correctness and incorrectness of utterances in specific cases. She considers three alternative accounts of semantic normativity: one according to which the correctness of an utterance depends on the truth of the sentence uttered; a second according to which the correctness of an utterance depends on the content of the utterance; and a third according to which the correctness of an utterance depends on whether the sentence uttered expresses the proposition that the speaker intended to express in making the utterance. These principles, she argues, support at most the claim that semantic correctness is instrumental, which falls short of the claim that meaning is normative in any interesting sense. Max Kölbel, in his contribution, aims to motivate and defend a conventional approach to assertion and other speech acts. He distinguishes a conventional notion of assertion from an intentional notion of assertion. According to the former, going through a procedure that conventionally counts as asserting that p is sufficient for asserting that p. According to the latter, asserting that p involves certain characteristic communicative intentions, so no conventional procedure could be sufficient for asserting that p. Kölbel argues that the conventional notion is explanatorily useful, and defends it against two primary objections. Taking up Dummett’s claim that a theory of meaning must engage with a theory of understanding, Guy Longworth, in his contribution, notes that little attention has been paid historically to a theory of understanding, and urges that this be rectified. He distinguishes four kinds of understanding: (i) dispositional understanding of standing meaning; (ii) occurrent understanding of standing meaning; (iii) dispositional understanding of what is said in particular utterances; and (iv) occurrent understanding of what is said, and focuses primarily on the last of these. Longworth argues that an account of understanding must provide an account of intake understanding – of the episodes by which one takes in what is said – and an account of uptake understanding – of the form of openness to what one thereby takes in that

Introduction 5

enables one to exploit facts about what is said in one’s theoretical and practical reasoning. The chapter provides an engaging exploration of the possibilities for and demands on an account of understanding, and paves the way for further contributions on the topic. Stefano Predelli, in his contribution, provides an analysis of the meaning of expressive prefixes, such as ‘alas’ and ‘hooray’. Predelli argues persuasively that the meaning of such expressions cannot be understood in terms of truth-conditional contributions to sentences, or character, but should instead be understood in terms of expressive constraints formally definable as restrictions on classes of contexts. The semantic treatment invokes the notion of non-defective uses of expressions, and yields a corresponding notion of expressive validity, defined in terms of non-defectiveness preservation. The analysis is then extended to expressives that modify subsentential expressions and, tentatively, to derogatory terms, whose meanings often involve both truth-conditional and expressive elements, thus involving both truth-conditional and non-defectiveness constraints. Gillian Russell, in her contribution to the volume, proposes a new understanding of analytic truth. The chapter is centred on four recent developments in the philosophy of language – directly referential names; semantic externalism; minimalism about semantic competence; and indexicality – each one of which generates a puzzle for analyticity. The puzzles arise because these aspects of language are inconsistent with the folk picture of meaning according to which the meaning of an expression plays three distinct roles: it is (i) what a speaker must know in order to understand an expression, (ii) what determines the extension of an expression, and (iii) what an expression contributes to what is said by a sentence containing it. However, Russell claims that three distinct aspects of meaning need to be distinguished: character, reference determiner and content. Once these have been distinguished, she argues, the notion of a reference determiner can be employed to yield a definition of analytic truth. My own contribution to the volume focuses on proper names. I defend the view, initially proposed by Tyler Burge, that a name is a predicate in its own right as opposed to either a manufactured or abbreviated predicate or an individual constant. The view has three virtues: it provides a unified treatment of singular uses of proper names (as in ‘Oliver is three’) and non-singular uses of proper names (as in ‘There are three Olivers at the nursery’); it assigns the correct truth-conditions to sentences containing plural occurrences of names by allowing quantifiers to range over all individuals who share a name rather than all individuals who are known by the speaker to share a name; and it allows for a fully formalized truth-theory for a speaker at a time. I respond to a series of objections that have been thought to tell against the view and go on to argue that the view has two additional benefits: first, it allows for a unified theory of singular thought by treating singular uses of proper names and demonstrative expressions

6

Sarah Sawyer

as semantically alike; and second, it yields a straightforward solution to Kripke’s puzzle about belief. Jeff Speaks, in his chapter, is concerned with mentalist theories of meaning. The mentalist strategy takes two principal forms, one of which analyzes linguistic meaning in terms of communicative intentions, and the other of which traces linguistic meaning to belief. Speaks provides a novel and general argument against each of these forms of mentalism. He argues against the former, Gricean analysis on the grounds that it suffers from a version of Moore’s paradox; and he argues against the latter version on the grounds that facts about meaning are more fine-grained than facts about beliefs. He goes on to offer an account of term introduction and meaning transmission which grounds linguistic meaning in the contents of perceptual representation and grounds the contents of propositional attitudes in a combination of perception and linguistic meaning. On the one hand this account vindicates the mentalist claim that linguistic representation is derived from a kind of mental representation – perceptual representation; but on the other hand it tells against the mentalist claim that thought is prior to linguistic meaning. In the final contribution to the volume, James A. Woodbridge and Bradley Armour-Garb set out a novel approach to resolving a number of the familiar dilemmas that provide philosophy of language with much of its subject matter. The approach postulates semantic pretence at work where these puzzles arise. After detailing the puzzles and explaining the approach, they go on to look at some further dilemmas that arise more generally in metaphysics and epistemology and point to promising new directions in contemporary analytic philosophy. The primary issues in the philosophy of language have remained the same over recent years, and yet progress continues to be made. This collection serves as a testament to some of that progress.

Notes 1. For a collection of articles representative of the logical positivist movement see Ayer (1959). For the classic arguments against see Quine (1951). 2. For a collection of articles representative of the linguistic turn see Rorty (1967). 3. Strawson (1950). 4. Quine (1948 and 1960). 5. Davidson (1984). 6. Dummett (1975 and 1976). 7. Grice (1989). 8. See Russell (1912); Searle (1958); Strawson (1950 and 1971); Donnellan (1966); Kripke (1972); and Putnam (1970). 9. See, e.g., Salmon (1986) and Soames (2002). 10. See in particular Frege (1884 and 1952). I am indebted in this introduction to Burge (1992).

Introduction 7

Bibliography Ayer, A.J. (ed.) (1959) Logical Positivism (New York: Free Press). Burge, T. (1992) ‘Philosophy of Language and Mind: 1950–1990’, Philosophical Review, 101, 3–51. Davidson, D. (1984) Inquiries into Truth and Interpretation (Oxford: Oxford University Press). Donnellan, K. (1966) ‘Reference and Definite Descriptions’, Philosophical Review, 75, 281–304. Dummett, M. (1975) ‘What Is a Theory of Meaning?’, in S. Guttenplan (ed.) Mind and Language (Oxford: Oxford University Press/Clarendon). —— (1976) ‘What Is a Theory of Meaning (II)?’, in G. Evans and J. McDowell (eds) Truth and Meaning (Oxford: Oxford University Press/Clarendon). Frege, G. (1884) The Foundations of Arithmetic, trans. J.L. Austin (Evanston, Ill.: Northwestern University Press). —— (1952) Translations of the Philosophical Writings of Gottlob Frege, P. Geach and M. Black (eds) (Oxford: Basil Blackwell). Grice, P. (1968) Studies in the Way of Words (Cambridge, Mass.: Harvard University Press). Kripke, S. (1972) Naming and Necessity (Cambridge, Mass.: Harvard University Press). Putnam, H. (1970) ‘Is Semantics Possible?’, in his Philosophical Papers, vol. 2 (Cambridge: Cambridge University Press). Quine, W.V.O. (1948) ‘On What There Is’, Review of Metaphysics, 2, 21–38. —— (1951) ‘Two Dogmas of Empiricism’, Philosophical Review, 60, 20–43. —— (1960) Word and Object (Cambridge, Mass.: MIT Press). Rorty, R. (ed.) (1967) The Linguistic Turn (Chicago: Chicago University Press). Russell, B. (1912) The Problems of Philosophy (Indianapolis, Ind.: Hackett). Salmon, N. (1986) Frege’s Puzzle (Cambridge, Mass.: MIT Press). Searle, J. (1958) ‘Proper Names’, Mind, 67, 166–73. Soames, S. (2002) Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (Oxford: Oxford University Press). Strawson, P.F. (1950) ‘On Referring’, Mind, 59, 320–44. —— (1971) Logico-Linguistic Papers (London: Methuen).

1 Vagueness and Non-Indexical Contextualism Jonas Åkerman and Patrick Greenough

1

Introduction

Contextualism concerning vagueness (hereafter ‘CV’) is a popular response to the puzzle of vagueness.1 The purpose of this chapter is to highlight some of the most basic components of CV, and to show that it is crucial to distinguish between two types of context-sensitivity in order to evaluate CV properly. In Section 2, we sketch a generic form of CV. In Section 3, we distinguish between Indexical and Non-indexical context-sensitivity and two corresponding versions of CV.2 In Section 4, we discuss the extent to which various forms of ‘blindness’ are problematic for these two versions of CV. Non-indexical CV is found to fare better than Indexical CV in this respect. In Section 5 we address a challenge posed by Keefe (2007) to the effect that CV entails that any speech report of what has been said by a particular vague utterance, where the context of utterance and the reporting context are relevantly different, will almost always be inaccurate. While this challenge is prima facie effective against Indexical CV it proves to be less effective against Non-Indexical CV.3

2

Generic contextualism concerning cagueness

Generic contextualism concerning vagueness (hereafter ‘GCV’) is characterized via four key theses. First, we have: (GCV1): Vagueness consists in a certain kind of context-sensitivity. GCV1 is not simply the claim that all vague expressions are context-sensitive since it is perfectly consistent to follow Travis (1985 and 2008) or Searle (1978 and 1980) and accept ‘Radical Contextualism’ – the thesis that all expressions (and so all vague expressions) are context-sensitive – and yet deny that vagueness is a species of context-sensitivity. What is needed is the claim that vagueness is constituted by a particular species of context-sensitivity.4 8

Vagueness and Non-Indexical Contextualism

9

Say that a sentence type S is generically context-sensitive with respect to contextual factors c1, ... ,cn if and only if there is a context of utterance C and a context C’ differing only with respect to c1, ... ,cn such that either S is true relative to C but false relative to C’ or the proposition expressed by S in C differs from the proposition expressed by S in C’ (relative to a fixed subject-matter for S).5 So, while a shift in truth-value is sufficient for (sentential) contextsensitivity, it is not necessary.6 The sentence ‘I am here’ cannot take different truth-values in different contexts of utterance. When uttered, this sentence is always true.7 Yet it can be used to express different propositions in different contexts, which is sufficient (though not necessary) for context-sensitivity. Which features of the context are relevant to vagueness? Any sensible form of CV ought to represent many of the context-sensitive features of vague expressions as having nothing per se to do with vagueness. The expression ‘here’ is vague, but its vagueness need have nothing to do with the fact that its reference can shift depending on the place of use.8 Equally, the application of the predicate ‘is tall’ can vary as a function of the operative comparison class and/or what is taken to be typically tall. But such shiftiness in the extension of ‘is tall’ need have nothing as such to do with vagueness.9 But then just what kind of contextual parameters are responsible for the shiftiness in truth-value or proposition expressed which is constitutive of vagueness? We will not try to answer this question in what follows. Rather, we will use the term v-standards as a neutral placeholder for whatever contextual parameters are taken to be responsible for the shifts. However, the v-standards cannot be just any kind of contextual parameters, so the following two qualifications are needed. First, shifts in comparison class can effect shifts in standards. To borrow an example from Fara (2000), the sentence ‘John is rich’ might be used to express the proposition that John is rich for a philosopher on one occasion and used on another to express the proposition that John is rich for an executive at Microsoft. Here what counts as rich depends upon which comparison class is being invoked. But not all shifts in standards are due to such shifts. When the comparison class is made explicit in the syntax of the predicate then the standards for when the predicate counts as applying can typically still shift – at least if vagueness remains in play. For example, take the predicate ‘is tall for a Ugandan pygmy’. In the context of writing an academic paper, this predicate may fail to apply to Ugandan pygmies of a certain height, yet in the context of a discussion which takes place in the pub, the predicate may apply. Indeed, the application of many predicates does not involve an implicit comparison class at all, but rather a relativisation to a (contextually fixed) paradigm case.10 Given this, the v-standards must remain in play even after the relevant comparison class and/or paradigm case has been fixed. Second, the v-standards cannot quite be Lewis’s standards of precision (see Lewis 1979). For many vague expressions it does not even make sense to talk about a standard of precision. Paradigm examples of vague predicates

10 Jonas Åkerman and Patrick Greenough

such as ‘is bald’, ‘is a heap’, or ‘is red’ are not such that their correct application depends on standards of precision. So it seems better to think of the v-standards as something like standards of application rather than standards of precision. The idea would be that in some contexts a greater amount of the relevant property is required for an object to fall in the extension. For instance, in some contexts a yellowish-orange surface might count as yellow, but in other contexts where the standards of application are higher, it might not count as yellow. However, raising the standards for one predicate may entail that we lower it for another predicate, if those predicates are contraries like ‘is red’ and ‘is yellow’, or contradictories like ‘is red’ and ‘is not red’. For instance, if I go through a sorites series of patches ranging from yellow to red, and keep judging increasingly more reddish patches as yellow, I will simultaneously lower the standards of application for yellow, and raise the standards of application for red. Thus it makes no sense to say that the v-standards, in the case in hand, have become more or less strict simpliciter. Rather they have become more liberal for ‘is red’ and less liberal for ‘is not red’.11 With these qualifications in place, and given the definition of generic context-sensitivity, we can now state the second key principle of GCV thus: (GCV2): A sentence type S is vague if and only if S is generically contextsensitive with respect to the v-standards.12 The context-sensitivity which is constitutive of vagueness is not unconstrained. In particular: (GCV3): The context-sensitivity which is due to vagueness must respect polar case constraints. For instance, when we have a sorites series like the one just described, positive polar case constraints mandate that ‘is red’ express a property that determines (together with the relevant non-linguistic facts) that positive polar cases, like the first couple of clearly red patches in the series, belong to its extension. The same holds, mutatis mutandis, for negative polar case constraints. In general, across a typical sorites series for F-ness there is a zone of variation (‘the borderline area’) in the middle of the series, whereby whether ‘is F’ applies to an object in this zone depends on the v-factors. Outside the zone of variation, polar case constraints mandate that the predicate applies (or does not apply) regardless of the v-factors. A further key thesis (with respect to predicate vagueness) is: (GCV4) If x and x’ are adjacent objects in the sorites series running from F to not-F, it is not the case that there is a context C such that x and x’ are considered together as a pair in C and ‘is F’ (as used in C) is true of x and ‘is F’ (as used in C) is false of x’.13

Vagueness and Non-Indexical Contextualism

11

GCV4 is classically equivalent to: (GCV4)* If x and x’ are adjacent objects in the sorites series running from F to not-F, then for all contexts C, if x and x’ are considered together as a pair in C then if ‘is F’ (as used in C) is true of x then ‘is F’ (as used in C) is true of x’. Roughly, GCV4 says that, when considered pair-wise, adjacent members of the series are never category different. GCV4 is a principle of weak tolerance since it permits that (a) there is a context C and a context C’ such that ‘is F’ (as used in C) is true of x and ‘is F’ (as used in C’) is false of x’, and that (b) there can be a sharp boundary within C if x and x’ are not considered together as a pair in C. One symptom of vagueness is that vague predicates draw no known boundary across their dimension of comparison.14 GCV4 provides us with a way to explain how this symptom of vagueness arises: as we inspect each pair of adjacent items in the series, GCV4 ensures that the members of each adjacent pair cannot be category different. Given the factivity of knowledge, there is no context C such that there are two adjacent items x and x’, which are considered together in C, such that a subject knows that ‘x is F and x’ is not-F’ (as used in C) is true. Roughly, no (context in which there is a) boundary between saliently similar objects in the series entails no (context in which there is a) known boundary between those objects. Moreover, we can also draw on GCV4 to explain the appeal of the induction step of the sorites paradox: It is plausible because we confuse the weak and valid principle of tolerance GCV4* with the following strong and invalid principle of tolerance: (ST) If x and x’ are adjacent objects in the sorites series running from F to not-F, then for all contexts C, if ‘is F’ (as used in C) is true of x then ‘is F’ (as used in C) is true of x’. Roughly, we easily confuse the (true and plausible) claim that there is never a boundary between any adjacent items considered together as a pair with the stronger claim of ST that there is no boundary. Such a ready confusion confers plausibility onto the stronger claim – explaining why we believe ST despite its falsity. However, while GCV4* is valid, ST is invalid since it can be used as the major premise in a standard sorites paradox.15

3 Indexical CV and non-indexical CV We have seen that a sentence type is (generically) context-sensitive just in case in different contexts tokens of this sentence type can take different truth-values or express different propositions or both (relative to a fixed

12

Jonas Åkerman and Patrick Greenough

subject-matter). Given this disjunctive characterization, we can make the following distinction: Content Context-Sensitivity: A sentence type S is content context-sensitive with respect to contextual factors c1, ... ,cn if and only if there is a context of utterance C and a context C’ differing only with respect to c1, ... ,cn such that the proposition expressed by a use of S in C differs from the proposition expressed by a use of S in C’. Truth Context-Sensitivity: A sentence type S is truth context-sensitive with respect to contextual factors c1, ... ,cn if and only if there is a context of utterance C and a context C’ differing only with respect to c1, ... ,cn such that S as used in C is true while S as used in C’ is false (where the subject matter for S does not vary between C and C’). Restricting ourselves to sentences which are ‘utterance-contingent’ (i.e., not always true/false when uttered) then Content Context-Sensitivity entails Truth Context-Sensitivity, though not vice versa.16 Given this distinction, we can draw a further distinction between two forms of CV as follows:17 Content CV: A sentence type S is vague if and only if S is content contextsensitive with respect to the v-standards. Truth CV: A sentence type S is vague if and only if S is truth contextsensitive with respect to the v-standards.18 There are two important forms of Content CV:19 Surprise Indexical CV: the context-sensitivity which is constitutive of the vagueness of a term is indexical context-sensitivity. On such a view, vague terms would be indexical in much the same way that ‘I’, ‘now’, and so on, are indexical. The only difference between the indexicality which is constitutive of vagueness and the familiar indexicality of such terms is that vagueness-related indexicality is not obvious indexicality – it is what Cappelen and LePore (2005, p.12) call ‘surprise indexicality’. Soames seems to be the only extant contextualist who sponsors this view: To say that vague predicates are context-sensitive is to say that they are indexical. While the semantic content of an indexical varies from one context of utterance to another, its meaning does not (2002, p. 445).20 On a broadly Kaplanian model, this means that vague terms have two levels of meaning: a character (a function from contexts to content) and a content

Vagueness and Non-Indexical Contextualism

13

(a function from circumstances of evaluation to extensions/truth-values) (see Kaplan 1989). Another form of Content CV is as follows: Hidden Indexical CV: Vague terms contain a typographically (and phonetically) unrealized argument place at the level of logical form – a so-called hidden variable.21 For example, a vague sentence ‘Tube 50 contains red paint’ has something like the logical form Tube 50 contains red paint relative to standards X, where context fills in the relevant parameters for X. Roughly, on both forms of Content CV, the proposition expressed by an utterance, made in paint shop A, of the sentence ‘Tube 50 contains red paint’ is Tube 50 contains red paint by the standards of paint shop A. This proposition is true in the world of A and so the utterance is true simpliciter. If this sentence is uttered in paint shop B then it expresses the proposition Tube 50 contains red paint by the standards of paint shop B. This proposition is false in the world of B and so the utterance is false simpliciter. So, as the proposition expressed shifts as a function of shifting standards, so can the truth-value of the sentence. Though truth context-sensitivity may come in a variety of forms, the most developed exemplar of the view is what MacFarlane (2007 and 2009) calls Non-Indexical Contextualism. MacFarlane develops a non-indexical model within a Kaplan-style semantics, whereby the following schema fixes the conditions under which a sentence, relativized to a context of use, is true: (T) A sentence type S, as used in context C, is true simpliciter just in case the proposition expressed by S in C is true at the circumstances of evaluation determined by C.22 Here the context of utterance plays two roles with respect to the determination of sentential truth: first, the context of use (plus the conventional linguistic meaning of S) determines which proposition is expressed; second, the context of use fixes the circumstance of evaluation against which this expressed proposition is evaluated for truth. MacFarlane (2009) usefully calls these the ‘content-determinative role’ and the ‘circumstance-determinative role’, respectively. Content context-sensitivity involves the former role, while truth context-sensitivity involves the latter. Given this, we thus have: Non-Indexical Context-sensitivity: A sentence S is non-indexically context-sensitive with respect to contextual factors c1, ... ,cn if and only if there is a context of utterance C and a context C’ differing only with respect to c1, ... ,cn, and a proposition P, such that a use of the sentence S in C expresses P and P is true relative to the circumstances of evaluation as determined in C, while a use of the sentence S in C’ expresses P and

14

Jonas Åkerman and Patrick Greenough

P is false relative to the circumstances of evaluation as determined in C’ (where the subject matter for S does not vary between C and C’). All those who accept the (standard) model theoretic notion of a proposition which can be true relative to some worlds and false relative to others are committed to a minimal form of Non-Indexical Contextualism. Temporalism – the view that a proposition can be true at one time and false at another – is also a form of Non-Indexical Contextualism whereby the circumstances of evaluation contain a time parameter (in addition to a world parameter). Given the definition provided, the form of Non-Indexical Contextualism we are going to be concerned with here can be defined as follows: Non-Indexical CV: A sentence type S is vague if and only if S is nonindexically context-sensitive with respect to the v-standards. Whatever the context, the proposition expressed by a use of the vague sentence ‘Tube 50 contains red paint’ is always the (v-standards-neutral) proposition Tube 50 contains red paint. This content may be true relative to the v-standards of C and false relative to the v-standards of C’. Given that ‘Tube 50 contains red paint’, as uttered in C, is true simpliciter if and only if the proposition expressed by this sentence in C is true relative to the circumstances of evaluation determined by C, then we can explain why the sentence, relative to C, is true simpliciter and, relative to the context of use in C’, is false simpliciter. So, while propositional truth is v-standards relative, utterance truth, i.e., truth for sentence tokens (sentence-context pairs) is absolute.23 In what follows, we will only be concerned with Surprise Indexical CV (henceforth ‘Indexical CV’) and Non-indexical CV, and we will argue that the latter is a better alternative, at least with respect to the issues we will be concerned with. Could extant proponents of CV accept Non-indexical CV or are they committed to other forms of context-sensitivity? Well, in fact, many of them are somewhat unclear about what kind of context-sensitivity they are committed to. Soames is an exception; as shown in the quote, he explicitly endorses Indexical CV. However, since his account of vagueness is only concerned with extension shifts, and does not rely on Indexical CV in any crucial way, he too could reject Indexical CV without having to make any substantial revisions of his theory. We will not attempt to uncover whether or not any of these forms of context-sensitivity could reasonably be ascribed to other proponents of CV, such as Raffman, Shapiro, and Fara. For present purposes it suffices to observe that all of them could adopt Nonindexical CV without distorting any essential elements of their theories.

4 Blindness A general worry about CV is that it seems to entail that speakers are blind to certain features of their own language. The extent to which such blindness

Vagueness and Non-Indexical Contextualism

15

is problematic depends on what form it takes, so let us first distinguish between some varieties of blindness: Strong Content Blindness (SCB): Speakers are blind to the fact that what content is expressed by an utterance of a vague sentence is (partly) determined by the context. Weak Content Blindness (WCB): Speakers are ignorant of what content is expressed by an utterance of a vague sentence. Strong Truth Blindness (STB): Speakers are blind to the fact that the truthvalue of what is expressed by an utterance of a vague sentence is (partly) determined by the context. Weak Truth Blindness (WTB): Speakers are ignorant of the truth-value of what is expressed by an utterance of a vague sentence. It seems obvious that speakers do not take vague expressions to be contextsensitive in the way that CV claims, since otherwise, CV would hardly be controversial at all.24 If this is right, then, Indexical CV entails (SCB) and Non-indexical CV entails (STB). Is this a problem? What would we say if we were confronted with a speaker who was blind to the context sensitivity of an overt indexical? For instance, consider a speaker who does not readily recognize that the content of ‘here’ depends on features of the context. Would such a speaker count as fully understanding ‘here’? It seems not. Such blindness would lead us to think that the speaker does not have an adequate grasp of ‘here’. Indeed, if you are not aware that the truth-value or content of sentences containing this expression depends on the context of utterance in a certain way, you are unlikely to be very successful in communicating with it. Now, speakers of vague languages arguably have an adequate grasp of vague predicates in this sense, but they do not readily recognize them to be context dependent in the way that CV claims. So, if the point just made about ‘here’ were to hold in the case of vague expressions, there would be at least prima facie evidence against CV. However, it is far from clear that we can generalize from the case of ‘here’ and other overtly context sensitive expressions. After all, vague expressions as conceived by CV differ from overt indexicals in that their context sensitivity is less obvious, so it seems more likely that ordinary speakers might be ignorant of it without compromising their grasp of these expressions. Moreover, everyday communication with vague predicates does not require that speakers recognize this special kind of context sensitivity in the same way as in the case of overt indexicals. So, it is not at all clear that SCB and STB pose any significant problem in this respect. Let us turn to the other forms of blindness. According to CV, ordinary speakers are typically unaware of the shifts that occur in the relevant contextual factors, and they are also ignorant of what context they are in.25 In

16

Jonas Åkerman and Patrick Greenough

other words, CV ascribes a form of context blindness to ordinary speakers. Now, since context blindness entails that speakers do not know what context they are in at a given moment, and that contexts may shift in a way that alters the content/truth-value without their noticing, Indexical CV will entail WCB and Non-indexical CV will entail WTB. In fact, that the truthvalue depends on context in this way is entailed by both forms of CV, so both of them will be committed to WTB.26 However, only Indexical CV entails WCB. WTB is more local than WCB in two respects. First, WTB holds only for the borderline area, since when it comes to the polar cases, there is no shiftiness or ignorance of truth-value.27 No matter what the extension of the vague predicate used is, the boundary will always be inside the zone of variation. However, WCB holds outside the borderline area as well. If the property ascribed to the polar cases depends upon unknown contextual factors, which can shift at any moment and thus effect a change in the property expressed by the vague predicate, the speaker will not know which property she ascribes to the polar cases, even though she can know that her ascriptions are correct. Second, according to most versions of CV, the judgements that competent speakers make in the borderline area are bound to be correct, either because the judgement itself determines that the judgement is true, or because the same facts that determine the judgements also determine that the judgement is true.28 On either of these alternatives, the speaker reliably makes correct judgments within her own context. Thus, there is scope for claiming that speakers (at least implicitly) know the truth-values of her own (present) uses of vague sentences, and that WTB is restricted to judgements made outside one’s own context. Are WCB and WTB problematic? When it comes to overt indexicals, corresponding forms of blindness are not problematic in general. For instance, we can be ignorant of the content expressed by a sentence like ‘It is quiet here’, because we are ignorant of the values of the relevant contextual factors, viz. the place of the utterance. Patrick might be locked up in the trunk of a car, and have no idea of where he is. He could still utter the sentence, and refer to the place where he happens to be, but he will be ignorant of what proposition he expresses by this utterance (Cf. Kaplan 1989, p. 536). Instances of (weak) content blindness that result from such ignorance of what the context is should not lead us to question the speaker’s linguistic competence. What about ignorance of truth-value? Such ignorance is equally harmless, at least when the reason that we are ignorant of the truthvalue of what we say is that we are ignorant of the (non-linguistic) facts that determine the truth-value. So, blindness due to ignorance of the (non-linguistic) factors that determine content or truth-value need not compromise a speaker’s grasp of the expression in question. Does that mean that WTB and WTC are equally

Vagueness and Non-Indexical Contextualism

17

unproblematic with respect to this? That depends. The reason that Patrick can be ignorant of the content of ‘It is quiet here’ and still count as having an adequate grasp of ‘here’ is that he would have known what the content was had he had the required access to the relevant contextual factors. In other words, Patrick knows the character (in Kaplan’s sense) of ‘here’, and that is what constitutes his grasp of this expression. The reason that ignorance of truth-value is not problematic in general is that we often have a grasp of the proposition expressed, so that if we had access to the relevant (non-linguistic) truth-determining factors, we would know the truth-value. So, ignorance of content need not compromise the speaker’s grasp of an indexical given that she knows the character, and ignorance of truth-value need no compromise a speaker’s grasp of a sentence in general given that she knows the proposition expressed. Are these strategies available to CV? Well, given that WTB is restricted to judgements outside one’s own context (for the reasons given before), it could be claimed that speakers actually display knowledge of the Non-indexicalist’s v-standards neutral proposition (conceived of as a function from circumstances to truth values), since when the speaker is in a given context, she will make judgements that are true at the circumstance determined by that context. So, the Non-indexicalist can maintain that speakers have an implicit grasp of the proposition expressed. Could the Indexicalist claim that speakers display implicit knowledge of the character in a similar way? No. Since WCB is not restricted to judgements outside one’s own context, they will be ignorant even of the propositions expressed by their own (present) utterances.29 This leaves the proponent of Indexical CV with the problem of explicating what the speakers’ grasp of vague expressions consist in. If vague expressions are indexicals, and they do not know the character, in virtue of what do they know their meaning?

5 Speech reports We take ourselves to be able to use indirect speech to report what other people have said in different contexts. However, if the contents of vague expressions are in constant flux due to contextual variation, how is this possible? This is the gist of Rosanna Keefe’s recent objection to CV: ‘S said that a is F’ will almost never be strictly true. For the context of this report will be different from that in which S made the utterance, due to the [difference in v-standards]; correspondingly, ‘a is F’ will express a different proposition in those two contexts. We will generally be lacking a way to report speech, and the indirect speech reports that we do make will almost always be inaccurate. (Keefe 2007, p. 286) It seems that this challenge is only effective against Indexical CV. According to Non-Indexical CV, the content does not vary between the

18 Jonas Åkerman and Patrick Greenough

context of the original utterance and the context of the report, so there is no problem with indirect speech reports. In an intuitively correct report, the same propositional content is expressed in the report and the original utterance, despite the context change, so the report is not rendered false by Non-indexical CV. The linguistic vehicle used in making the utterance can also be used to report the utterance. 30 Thus, Non-Indexical CV – but not Indexical CV – can easily allow for correct and reliable indirect speech reports. 31 But why take the v-standards neutral proposition to be the content expressed rather than the v-standards neutral proposition together with the v-standards determined by the context? After all, only the latter is truthconditionally complete. Well, in fact, the Non-Indexicalist can acknowledge that there are two different levels of content: the relativized content and the complete content, where the former is just the v-standards neutral proposition p, while the latter consists of both p and the v-standards determined by the context. Following Recanati (2007), we can then go on to claim that the relativized content is the explicit content, and this is what needs to be preserved in a correct speech report. In other words, when we ask about what the speaker said, all we need to care about is what proposition the speaker expressed. In that case, the relativized content is enough. However, when we want to evaluate the utterance, we need the complete content.32 So, according to Non-indexical CV, a speech report may leave out important aspects of the utterance, namely the v-standards determined by the context, without omitting anything that is crucial for fulfilling the purpose of reporting, which is just to convey the explicit content of the utterance, i.e., the v-standards neutral proposition that the speaker expressed. This means that there is a sense in which the report is incomplete, but this incompleteness does not lead to any problems as regards the correctness of the report. While Keefe’s objection does not work against Non-Indexical CV, the challenge for Indexical CV remains, and it is unclear how it can be met. However, the Non-indexicalist framework seems to generate some counterintuitive consequences, which must be dealt with before we can conclude that Non-indexical CV is superior when it comes to handling speech reports. The remainder of this section will be devoted to this task. Suppose Julia utters the following: ‘Sebastian said that Charles is tall, but it is false that Charles is tall, so Sebastian said something false’.33 Suppose further that Sebastian uttered ‘Charles is tall’ in C, and thereby expressed the v-standards neutral proposition p, and that Julia’s reporting context is C’. Moreover, let C and C’ differ only with respect to the v-standards, and let p be true in C but false in C’. According to Non-indexical CV, both Julia’s report and her claim that it is false that Charles is tall are correct. Intuitively, her reasoning is fine, so she should be in a position to conclude that Sebastian said something false. However, according to Non-Indexical

Vagueness and Non-Indexical Contextualism

19

CV, Sebastian’s utterance was true when he made it, so Julia should not be able to (correctly) draw this conclusion. In order to solve this puzzle, we need to distinguish between the truthvalue of what Sebastian said and the truth-value of his utterance. In other words, we should distinguish utterance truth from propositional truth.34 ‘Sebastian said something false’ should be taken as a claim about the complete content of his utterance, while ‘It is false that Charles is tall’ should be interpreted as a claim about the proposition he expressed, i.e., the explicit content.35 Given this distinction, the falsity of the proposition that Sebastian expressed in the reporting context does not entail that Sebastian’s utterance was false, so Sebastian’s utterance can be true, and Julia’s reasoning thus turns out to be flawed. But now a different problem emerges: according to the account just given, Julia could correctly say ‘Sebastian said that Charles is tall, but it is false that Charles is tall; however Sebastian said something true’. Since most ordinary speakers would arguably find this utterance self-contradictory, this example provides yet another challenge for Non-Indexical CV.36 In order to cope with this objection, we should simply deny that the utterance is inconsistent. It only seems inconsistent as long as we fail to make the crucial distinction between utterance truth and propositional truth. Moreover, Julia’s utterance is misleading in that it fails to make explicit that there is a significant difference between the original context and her own. If Julia had said ‘Sebastian said something true in that context’ or something to the same effect, it would have seemed less contradictory. Once we take the contextual variation into account, and distinguish between utterance truth and propositional truth, the appearance of inconsistency vanishes. Finally, we should not be worried or surprised if ordinary speakers do not notice the contextual variation or fail to make the crucial distinctions, and thus have a hard time seeing that there really is no inconsistency, since such theoretical knowledge is rarely needed in everyday communication with vague language.

6 Conclusion We have argued that there is an important distinction between two different kinds of CV: Indexical CV and Non-indexical CV. We have also argued that non-indexical CV is far less problematic than indexical CV, at least with respect to the issues we have considered here. Of course, this is not sufficient to settle the score between Indexical CV and Non-indexical CV, or to vindicate any form of CV. Many outstanding issues remain: What should CV say about higher-order vagueness? Is a relativistic theory of vagueness superior to Non-Indexical Contextualism? Does CV allow us to keep classical logic in its entirety? Is CV at bottom of a form of epistemicism? But these remain issues for another day.

20

Jonas Åkerman and Patrick Greenough

Acknowledgements Parts of this chapter were presented at the Seventh Arché Vagueness Workshop, November 2006, University of St. Andrews, The Arché Audit, June 2007, a workshop on vagueness at the University of Navarre, Spain, June 2007, in Sydney, August 2008, and in Stockholm, September 2008. For valuable feedback thanks to Elizabeth Barnes, Herman Cappelen, Maria Cerezo, Mark Colvyn, Richard Dietz, Mikael Janvid, Dan López de Sa, John MacFarlane, Teresa Marquez, Martin Montminy, Aidan McGlynn, Paula Sweeney, Sebastiano Moruzzi, Kathrin Glüer-Pagin, Peter Pagin, Diana Raffman, Sven Rosenkranz, Mark Sainsbury, and Stewart Shapiro, Levi Spectre, Åsa Wikforss, Crispin Wright, and Elia Zardini. This chapter was completed while one of the authors (Greenough) was a postdoctoral research fellow in the Epistemic Warrant Project at ANU 2007–8 and a visiting fellow at the Centre for Time, University of Sydney. Thanks to all the philosophers at both ANU and Sydney for their immense (philosophical) hospitality.

Notes 1. CV is defended by Kamp (1981), Raffman (1994 and 1996), Soames (1999 and 2002), Fara (2000), and Shapiro (2003 and 2006); see also Lewis (1979). For criticism of certain forms of CV see Williamson (1994 and 2002), Stanley (2003), Heck (2003), Greenough (2005), and Keefe (2007). 2. The label ‘Non-Indexical Contextualism’ is due to MacFarlane (2007 and 2009). For a non-indexical model of ‘knows’ see Kompa (2002) and Brogaard (2008). For a non-indexical model of predicates of taste see Lasersohn (2005). For an extensive defence of the general framework see Recanati (2007). 3. Stanley (2003) has argued that CV is flawed because (i) The sorites paradox can be run using verb-phrase ellipsis, (ii) indexical expressions are invariant under verbphrase ellipsis, and (iii) any solution to a verb-phase ellipsis version of the sorites should be on all fours with a solution to the standard formulation of the paradox. Even if Stanley’s argument is cogent it is only directed at indexical forms of CV and so might be thought to provide a further reason to prefer Non-Indexical over Indexical CV. However, as it turns out, matters are somewhat more complicated. We hope to address this issue in future work. 4. Williamson (1994, p. 214) says ‘the lack of natural boundaries for vague words makes context-dependence hard to avoid, but that is an empirical correlation not an a priori law’. According to GCV, there is such a law. 5. Even if the weights of everything remain fixed, the predicate ‘is heavy’ can nonetheless change in extension (or property expressed) across different context of utterance. Hence, the qualification ‘relative to a fixed subject matter for the sentence in question’. 6. A context of utterance (hereafter ‘a context’) is taken to be a sequence of parameters: world, speaker, time, location, orientation, etc. 7. We are ignoring the ‘answering-machine’ data discussed by Sidelle (1991) and Predelli (1998). However, there seems to be examples which are not threatened by such data, like ‘Today is today’. (We owe this example to Martin Montminy.) 8. See Williamson (1994, p. 215).

Vagueness and Non-Indexical Contextualism

21

9. For this reason, in speaking of the vagueness of ‘is tall’ it is better to speak of the vagueness of the predicate ‘is tall with respect to the operative comparison class’, e.g., ‘is tall for a Ugandan Pygmy’. For convenience, we shall usually omit such relativisation in what follows. 10. Many standards-sensitive terms involve no relativisation to either a paradigm case or comparison class, e.g., ‘The Earth’s atmosphere’, ‘now’, ‘nearby’, ‘roughly 10,000’, ‘looks somewhat triangular’, ‘is browner than’. 11. Cf. Keefe (2000, p. 173). 12. We don’t propose GCV2 as a reductive analysis since a full account of the right hand-side may itself, at some point, deploy the concept of vagueness. 13. GCV4 is cognate to both Raffman’s principle IP* which says that ‘for any n, if patch x is F then patch x’ is F, relative to a pair-wise presentational context (1994, p. 68), and Fara’s salient-similarity constraint which says that ‘if two things are saliently similar, then it cannot be that one is in the extension of the predicate, or in its anti-extension, while the other is not’ (Fara 2000, p. 57). Cf. Soames (1999, pp. 214–16) and Shapiro (2003, pp. 42–3). 14. Greenough (2003) dubs this symptom ‘epistemic tolerance’. 15. ST entails that the predicate ‘is F’ is tolerant in all contexts. So, in the present context, for all x, if x is F then x’ is F. That is, the major premise of the standard sorites paradox follows from ST. Given classical logic, and the fact that the first member of the series is F and the last member is not-F, then the major premise is false and so ST is false. Arguably, not all forms of CV should take the major premise to be false. See Greenough (2005) and Åkerman and Greenough (2009) for relevant discussion. 16. Lewis (1979, p. 345) does not think that context-sensitivity is exhausted by content context-sensitivity when he says that ‘[T]he constituents of an uttered sentence [ ... ] may depend on the score for their intension or extension’ (our italics). 17. This distinction was drawn in Greenough (2005, p. 172) before the present authors encountered MacFarlane on Non-Indexical Contextualism (MacFarlane 2007, 2009). In Greenough (2005) and in Åkerman and Greenough (2009), we also distinguish between ‘Boundary-Shifting CV’ and ‘Extension-Shifting CV’. 18. An immediate consequence of Truth CV is that, insofar as one thinks that the sentence ‘I am here now’ is always true relative to a context, then this sentence is not extensionally vague (even though it may remain vague what has been said by an utterance of this sentence since we do not know the exact extension of ‘here’ as used in any context). 19. A further form of Content CV entails that vagueness-related context-sensitivity is to be modelled using an ‘unarticulated constituent’ strategy. For reasons of space, we shall not dwell on the form that such a strategy may take. 20. Though in his 1999 and 2003 Soames merely speaks of extensions shifting as the context shifts. 21. The term ‘hidden indexical’ is due to Cappelen and LePore (2005, pp. 8–9). 22. Cf. Kaplan (1989, p. 522), Lewis (1980). 23. Sentence-context pairs cannot shift in truth-value and yet such pairs seem to be vague. But if vagueness consists in context-sensitivity then CV seems to be in trouble. Call that the simple objection to CV. See Åkerman and Greenough (2009) for two ways out of the problem. 24. Note that we are talking about the special context-sensitivity that CV claims to be constitutive of vagueness, not humdrum forms of context-sensitivity, which are irrelevant to the problems of vagueness, but are nevertheless exhibited by vague expressions. See Section 2.

22

Jonas Åkerman and Patrick Greenough

25. See Raffman (1996, p. 188). 26. As noted in Section 3, Content-Context Sensitivity entails Truth ContextSensitivity (at least if we ignore sentences which are always true/false when uttered). 27. See GVC3. 28. The former would be a form of judgement dependence, while the latter could be called co-determination. Shapiro (2003, 2006) defends a version of judgement dependence, while Raffman (1994) seems to defend a version of co-determination. In Raffman’s account, it is somewhat unclear whether or not there is a genuine determination relation between the judgements and the truth-values or whether they are co-determined by psychological states of the speaker, but that is not of great importance here. Soames (1999, p. 209) claims that speakers have a discretion to include object in the borderline area in the extension or anti-extension of the vague predicate. 29. It is worth noting that if WCB were restricted in the same way as WTB, speakers would be able to know what property their use of a vague predicate expressed, and since property determines extension (according to Indexical CV) and they can know all the relevant facts about the objects in the sorites series, this would mean that they could know the location of the boundary. But this would conflict with epistemic tolerance (see Section 2). 30. Modulo the required substitutions of indexical expressions. 31. For the same reasons, Non-indexicalists can also ascribe disquotational knowledge of meaning and content to ordinary speakers, contrary to what Keefe claims. 32. Recanati (2007, pp. 82–6) invokes this distinction to rebut Mark Richard’s famous objection against Temporalism (see Richard 1981). The basic idea is that belief reports are ambiguous between one interpretation which requires the complete content, and another one which requires the relativized content. The most obvious similarity with the present application is that it is when we are interested in the truth-conditional properties of the belief that we need the complete content. Similarly, it is when we focus on the truth of someone’s utterance that we need the complete content. 33. This example is due to Peter Pagin. 34. As was done in Section 3. 35. The relation between utterance truth (i.e., truth of sentence-context pairs) and propositional truth is captured in principle T (see Section 3). 36. Thanks to Martin Montminy for pointing out this problem.

Bibliography Åkerman, J. and Greenough, P. (2009) ‘Hold the Context Fixed, Vagueness Still Remains’, in R. Dietz and S. Moruzzi (eds) Cuts and Clouds (Oxford: Oxford University Press). Brogaard, B. (2008) ‘In Defence of a Perspectival Semantics for “Knows” ’, Australasian Journal of Philosophy, 86, 439–59. Cappelen, H. and LePore, E. (2005) Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act Pluralism (Oxford: Blackwells). Graff Fara, D. (2000) ‘Shifting Sands: An Interest-Relative Theory of Vagueness’, Philosophical Topics, 28, 45–81. Originally published under the name ‘Delia Graff’. Greenough, P. (2003) ‘Vagueness: A Minimal Theory’, Mind, 112, 235–81.

Vagueness and Non-Indexical Contextualism

23

—— (2005) ‘Contextualism about Vagueness and Higher-Order Vagueness’, Proceedings of the Aristotelian Society, Supplementary Volume, 79, 167–90. Heck, R. (2003) ‘Semantic Accounts of Vagueness’, in JC Beall (ed.) Liars and Heaps (New York: Oxford University Press). Kamp, H. (1981) ‘The Paradox of the Heap’, in U. Mönnich (ed.) Aspects of Philosophical Logic (Dordrecht: Reidel). Kaplan, D. (1989) ‘Demonstratives’, in J. Almog and H. Wettstein (eds) Themes from Kaplan (New York: Oxford University Press). Keefe, R. (2000) Theories of Vagueness (Cambridge: Cambridge University Press). —— (2007) ‘Vagueness Without Context Change’, Mind, 116, 275–92. Kompa, N. (2002) ‘The Context Sensitivity of Knowledge Ascriptions’, Grazer Philosophische Studien, 64, 79–96. Lasersohn, P. (2005) ‘Context Dependence, Disagreement, and Predicates of Personal Taste’, Linguistics and Philosophy, 28, 643–86. Lewis, D. (1979) ‘Scorekeeping in a Language Game’, Journal of Philosophical Logic, 8: 339–59. —— (1980) ‘Index, Context, and Content’, in S. Kanger and S. Öhman (eds) Philosophy and Grammar (Dordrecht: Reidel). MacFarlane, J. (2007) ‘Semantic Minimalism and Nonindexical Contextualism’, in G. Preyer and G. Peter (eds) Context-Sensitivity and Semantic Minimalism: New Essays on Semantics and Pragmatics (Oxford: Oxford University Press). —— (2009) ‘Nonindexical Contextualism’, Synthese, 166, 231–50. Predelli, S. (1998) ‘I Am Not Here Now’, Analysis, 58, 107–15. Raffman, D. (1994) ‘Vagueness Without Paradox’, Philosophical Review, 103, 41–74. —— (1996) ‘Vagueness and Context Relativity’, Philosophical Studies, 81, 175–92. Recanati, F. (2007) Perspectival Thought: A Plea for (Moderate) Relativism (Oxford University Press). Richard, M. (1981) ‘Temporalism and Eternalism’, Philosophical Studies, 39, 1–13. Searle, J. (1978) ‘Literal Meaning’, Erkenntnis, 13, 207–24. —— (1980) ‘The Background of Meaning’, in J. Searle, F. Kiefer, and M. Bierwisch (eds) Speech Act Theory and Pragmatics, 221–32. Shapiro, S. (2003) ‘Vagueness and Conversation’, in JC Beall (ed.) Liars and Heaps (New York: Oxford University Press). —— (2006) Vagueness in Context (Oxford: Oxford University Press). Sidelle, A. (1991) ‘The Answering Machine Paradox’, Canadian Journal of Philosophy 21, 525–39. Soames, S. (1999) Understanding Truth (Oxford: Oxford University Press). —— (2002) ‘Replies’, Philosophy and Phenomenological Research, 62, 429–52. —— (2003) ‘Higher-Order Vagueness for Partially Defined Predicates’, in JC Beall (ed.) Liars and Heaps (New York: Oxford University Press). Stanley, J. (2003) ‘Context, Interest-Relativity, and the Sorites’, Analysis, 63, 269–80. Travis, C. (1985) ‘On What Is Strictly Speaking True’, Canadian Journal of Philosophy, 15, 187–229. —— (2008) Occasion-Sensitivity (Oxford: Oxford University Press). Williamson, T. (1994) Vagueness (London: Routledge). —— (2000) Knowledge and Its Limits (New York: Oxford University Press). —— (2002) ‘Soames on Vagueness’, Philosophy and Phenomenological Research, 65, 422–28.

2 Semantics and the Place of Psychological Evidence Emma Borg

1

Introduction

My aim in this chapter is to explore one point of similarity between an approach to semantic theorizing known as minimal semantics and an earlier approach to meaning proposed by Paul Grice. Before I set out the structure of the chapter in more detail, however, I would like to say something by way of an introduction to minimal semantics. Minimal semantics follows in the tradition of formal approaches to meaning of the kind put forward by Frege, early Wittgenstein, Carnap, Davidson, and others. Although the term ‘formal semantics’ is itself somewhat vague, it seems that what unites theorists on this side of the divide is, first, the belief that semantic content attaches to objects which can be formally described (so sentences or, more probably, sentences relativized to contexts of utterance) and, second, the expectation that linguistic meaning will be amenable to study via scientific methods of inquiry, broadly construed. This formal approach is of course diametrically opposed to the other main tradition in philosophy of language: speech act or use theories of meaning. The classic advocates of speech act theories include Austin, later Wittgenstein and Searle, but more recently the approach has re-emerged under the label of ‘contextualism’, as advocated by Recanati, Travis, and the relevance theorists Sperber and Wilson, and Carston, amongst many others. The age-old debate between formal theories and speech act theories has thus metamorphosed into the debate between minimalism and contextualism in the contemporary domain.1 To give us a slightly more detailed picture of semantic minimalism, in this chapter I’m going to take the approach to be characterized by the following two claims:2 1. There are minimal contents (propositions/truth-conditions): these are contents which are maximally free from contextual effects and provide the literal meanings of sentences. According to minimal semantics, 24

Semantics and Psychological Evidence

25

then, context is semantically relevant only when introduced by a standardly context-sensitive syntactic element, e.g., indexicals, demonstratives, tense markers (Cappelen and Lepore 2005; Borg 2004a; Soames 2002). 3 2. Semantic content is not speech act content. This claim comes in somewhat different forms, for instance one might hold that minimal content is a proper part of speech act content (see Soames 2002 and Cappelen and Lepore 2005) or one might hold that they are simply different kinds of entities (see Borg 2004a; this point of difference is discussed in Borg 2007). Minimalists are committed to claim (2) because of claim (1), for if there are such things as minimal contents for sentences, which are maximally free from contextual effects, it is clear that these are not the kinds of things which get communicated in normal conversational exchanges. To take an example, the minimalist is going to claim that a sentence like ‘That apple is red’ just means that that apple is red. However, what gets communicated by an utterance of this sentence will typically be a pragmatically enriched proposition, like that apple is red on most of its skin. Thus the minimalist is committed to drawing a sharp distinction between semantic content and pragmatic content or speaker meaning. It seems then that minimalism adopts a broadly Gricean perspective on semantics (though we should note that Grice himself steered clear of the terminology of semantics and pragmatics): the entities amenable of semantic analysis are sentences, semantic content is taken to be permeable by contextual features in certain highly constrained ways (e.g., reference determination for indexicals and demonstratives, and resolution of tense markers and ambiguity) and a clear distinction between sentence meaning and speaker meaning is made. However, given this shared perspective between the two accounts, we might expect that any objection to Grice which arises due to something in the common ground will also serve as an objection to minimalism. It is this point that I want to explore in this chapter. Specifically, I want to explore the objection that neither theory fits well with relevant psychological evidence concerning linguistic understanding. The structure of the chapter is thus as follows: in the next section I’ll sketch a range of options for the relationship between semantics and psychology, and locate Grice and minimalism together on this spectrum. Then we will turn in Section 2 to look at the general objection that the two accounts fail to fit with relevant psychological evidence and we will see exactly what form this charge has taken against the two accounts. In Section 3 I’ll suggest how I think a minimalist should respond to this challenge and finally, in Section 4, I’ll close by examining how this response might help the minimalist avoid another objection to her approach, recently put forward by Lenny Clapp.

26 Emma Borg

2

The relationship between semantics and psychology

In addition to the points of similarity given before it also seems that Gricean and minimal semantics share a fundamental assumption about the way in which semantics and psychology hang together. To see this I’d like to sketch very briefly what I take to be the three main options for relating semantics and psychology.4 As we will see, each of these options can be read in two distinct ways: a metaphysical version and a (weaker) epistemic variety. I’ll suggest that both Gricean and minimal semantics share a common metaphysical outlook, but that both accounts then seem to run into problems with the associated epistemic claim. So, our question now is: how might we construe the relationship between semantics and psychology? 2.1 Independence 2.1.1 Metaphysical independence According to this view there is no constitutive or dependence-based relationship between a correct semantic theory and the states of mind involved in language comprehension in ordinary agents. Constructing a semantic theory is taken to be one kind of enterprise while the construction of a theory of language processing is something quite different. Prima facie, this position may seem less than compelling. After all, if some piece of information, I, plays no role whatsoever in an agent’s coming to grasp the semantic content of an expression, E, we might wonder what could make I a genuine semantic fact about E at all. The thought behind such a rejection of metaphysical independence is that it is simply implausible to hold that the semantic facts about a human language are ones which no human ever cognizes. However, on reflection, I think such immediate scepticism about metaphysical independence is unfounded and that the position could yet turn out to be a plausible option. For we might envisage a semantic theory as constrained to capture knowledge that would suffice for understanding, regardless of whether it fits the actual cognitive processing of any language user. It would, it seems, advance our understanding of language to have a theory which could suffice for linguistic understanding, even if we simply lacked any information about whether or not the theory accurately captured the ways in which ordinary speakers came to linguistic understanding. Indeed, this kind of metaphysical independence might be suggested by at least some of the things that Davidson says about the role of a semantic theory, where a truth theory is required to do duty as a theory of meaning and to describe knowledge capable of underpinning (aspects of our) linguistic competence rather than as making prescriptive claims about the form which that competence actually takes in ordinary subjects.5 2.1.2 Epistemic independence According to this view there is held to be no epistemic route from one domain to the other: claims about the correct form for a semantic theory do not entail

Semantics and Psychological Evidence

27

any predictions about psychological processes or the contents or structure of the mind of typical language users, and vice versa. A claim of epistemic independence might perhaps be embraced by someone who held that semantic theorizing is solely concerned with conceptual analysis and that such conceptual analysis need not be answerable to what language users typically think or do when faced with a given linguistic prompt. Semantics here would thus not be required to answer to empirical discoveries in cognitive science. 2.2

Psychological facts depend on semantic facts

2.2.1 Metaphysical dependence On this view there is held to be a constitutive relationship between the contents of the mind of the language user and the content of a natural language (and hence the content of a correct semantic theory for that language), such that the former depends on the latter. This metaphysical assumption is most famously associated with the kind of linguistic determinism proposed by Whorf and endorsed by many others. Thus for the linguistic determinist the kinds of things one can say constrains the kinds of things one can think. To give the rather hackneyed examples familiar in this area, because Eskimos have so many more words for snow than do English speakers, Eskimos are supposed to be able to think more kinds of thoughts about snow than English speakers. Or again, since some nomadic tribes lack a complex number vocabulary (for instance, the language might have words only for one, two and many) speakers of this language are unable to think complex thoughts involving number (their language prevents them from thinking ‘I’ve got four sheep’, not just from saying it). It seems that linguistic determinism is not a theory much in favour at the moment (see Pinker 2007 for discussion), but we should note that the same kind of metaphysical assumptions which lie behind linguistic determinism also lie behind certain other contemporary approaches which claim that we think in a natural language (see Carruthers 1996). 2.2.2 Epistemic dependence On this view the route to an account of mental content runs through linguistic content, since thought content is essentially inaccessible and thus cannot provide a direct object of study. Such an epistemic dependence claim might be made in conjunction with (2.2.1) or independent of it. Made without commitment to metaphysical dependence, we have the kind of picture commonly associated with the so-called linguistic turn in philosophy, which sought to place philosophy of language centre stage for the study of the mind (see Dummett 1993; Evans 1983). 2.3 Semantic facts depend on psychological facts: 2.3.1 Metaphysical dependence According to this view, facts about semantic content are determined by facts about the minds of language users. What a semantic theory aims to capture

28

Emma Borg

on this view is the knowledge which underpins grasp of linguistic meaning amongst ordinary language users. It is the mind which is taken to be the primary locus for content, thus words and sentences acquire their meaning via their relationship to mental states. 2.3.2 Epistemic dependence According to this view, the route to a correct semantic theory runs via an account of the contents of the minds of language users (thus a putative semantic theory might be confirmed or disconfirmed by psychological evidence).6 *

*

*

Now, it seems that both Grice and minimal semantics sign-up to (2.3.1). For Grice the dependence of the semantic on the psychological is clear, for he aims to explicate semantic content in terms of intentional content – that is to say meaning is ultimately to be understood in terms of speaker intentions.7 For the minimalist the allegiance to (2.3.1) is perhaps less overt, yet still the claim is that a minimalist theory aims to capture that part of our psychological make-up responsible for a subject’s competence with linguistic stimuli, thus it would still seem right to characterize the theory as signing up to the dependence of semantics on psychology. So, according to both approaches semantic content depends in some way on mental content, on what is to be found in the minds of language users; but now we might ask ‘what about (2.3.2)?’ A claim of metaphysical dependence might well be thought to carry a claim of epistemic dependence in its wake, for if semantic content depends on psychological content then, ceteris paribus, we would expect psychological evidence to be relevant to semantic theorizing. Thus, unless we posit some kind of disruptive feature which serves to muddy the path from psychology to semantics, it would seem that psychological facts ought to provide good evidence for semantic facts. However, it is with respect to the claim of epistemic dependence in (2.3.2) that both our accounts seem to run into problems. The worry is that neither Gricean semantics nor minimal semantics fit properly with relevant psychological evidence, so let’s turn to this objection now.

3

Psychological evidence runs counter to the theories

To begin with Gricean semantics: on Grice’s model it seems that literal sentence meaning is prior to speaker meaning. On hearing an utterance, a subject S is supposed to first grasp the literal meaning of the sentence uttered, then see that this flouts some general principle of good communication, finally this licenses the subject to proceed to infer some more suitable proposition as the one the speaker actually meant to convey. So to take

Semantics and Psychological Evidence

29

an example, imagine that A says ‘There is nothing to eat’. A hearer, B, can then reason as follows: (i) The sentence ‘There is nothing to eat’ literally means there is nothing to eat (ii) The proposition in (i) is trivially false. (iii) Asserting trivial falsehoods is in contravention of the general maxims of communication. (iv) I believe that A is a competent speaker and abides by conversational maxims (v) Thus I should infer some more suitable proposition as the one A means to convey, e.g., I should take A as intending to communicate that there is nothing suitable to eat. For Grice, then, it seems to be an integral part of his account that sentence meaning comes first: it is what a hearer must grasp prior to proceeding to a grasp of speaker meaning. Our question now then is: does this Gricean account fit with the psychological evidence? The first point to notice is that it obviously doesn’t fit with first-personal psychological content, for we often arrive at attributions of speaker meaning without consciously entertaining sentence meaning and then engaging in the kind of extended inferential reasoning Grice suggests. However, this realization is not necessarily problematic for Grice, for his account might still hold as an account of occurrent mental content. That is to say, although we don’t consciously engage in the kind of reasoning which Grice suggests, such a process might still provide the unconscious route to a grasp of speaker meaning. So does the Gricean picture describe the unconscious processes by which we arrive at speaker meaning? The answer to this question seems to be ‘no’, for we sometimes seem to be in a position to grasp pragmatically enriched speaker meaning before we are in a position to grasp literal sentence meaning. There are at least three kinds of case which are relevant here: non-sentential assertion, metaphor comprehension, and scalar implicatures. Turning to non-sentential assertion first: it is clear that a significant proportion of the things people say do not (at least at the surface level) reach the level of complete sentences. Thus we have exclamations like ‘Fire!’ or ‘Help!’, and comments like ‘Nice dress’, ‘Bear country’, and ‘From France’. To make the case that these or similar utterances are genuine cases of non-sentential assertion (i.e., the production of something which falls short of sentence-hood but which nevertheless conveys a complete proposition) we need to be sure that there is no syntactically present but phonetically unmarked material in the utterances. That is to say, we need to be sure that the words spoken exhaust the syntactic content of the utterance, and in at least some cases this doesn’t seem to be the case.8 Whether or not all instances of apparently sub-sentential

30 Emma Borg

assertion can be handled by mechanisms like ellipsis is a much debated point (see Stainton 2006 for extended argument in favour of genuine subsentential assertion and Stanley 2000 for an argument against it) and it is not something we can hope to settle here. Thus the point I want to make is a conditional one: if it turns out that there are such things as genuine non-sentential assertions then they seem to show that Grice’s model of how speaker meaning is recovered cannot be correct. For obviously if a speaker does not produce a complete sentence but still succeeds in communicating a complete proposition at the level of speaker meaning, then grasp of that speaker meaning cannot itself depend on a prior grasp of sentence meaning. So non-sentential assertion, if a genuine phenomenon, provides a first piece of evidence against the Gricean model. A second challenge comes from so-called direct access views of metaphor recovery (e.g., Gibbs 2002), where it is held that we are at least sometimes able to recover metaphorical meaning for words and phrases before we are in a position to grasp complete sentence meaning. That is to say, at least sometimes subjects proceed to a metaphorical interpretation of part of a sentence before they have heard the sentence uttered in its entirety. So for instance, where we have talk of ‘icy glares’ or ‘green shoots of recovery’ the claim is that we proceed directly to a metaphorical interpretation of the phrases before we hear the whole sentence the phrases are embedded in. Once again, if this is right then it seems to cause problems for Grice’s account because it runs counter to the priority claim: hearers are not, contrary to what the Gricean account seems to demand, waiting to process a complete sentence prior to working out speaker meaning. Finally, this idea that pragmatic effects must, at least sometimes, occur at a local rather than a sentential level also seems to be demonstrated by some experiments concerning the recovery of scalar implicatures. A scalar implicature occurs when a speaker opts to use a weaker or stronger item on a given scale and thereby pragmatically conveys that the alternative terms on the scale do not hold. So for instance, though the lexical entry for ‘some’ is held to be that familiar from first-order logic, namely some and possibly all, many utterances of ‘some A’s are B’s’ convey the pragmatically enhanced reading that ‘some and not all A’s are B’s’ (e.g., ‘some delegates came to my talk’ conveys the enriched reading that some but not all of them did). Or again, ‘or’ is taken to have a lexical entry matching that for the inclusive-or in logic, namely ‘A or B or both’, but again many utterances involving ‘or’ convey a pragmatically enhanced reading, namely the inclusive-or ‘A or B and not both’ (e.g., ‘Main meals come with chips or salad’). For Grice, since such enhanced scalar readings are pragmatically enhanced instances of speaker meaning they should only be available to subjects once they have determined the literal meaning of the complete sentence in which the scalar terms appear. So, recalling the picture provided, if I hear you say ‘Some delegates came to my talk’ I should first work out the literal meaning of this

Semantics and Psychological Evidence

31

sentence, then I should see that this flouts some principle of good communication (for instance, it is not the most informative thing you could have said), finally I should infer some alternative proposition, e.g., some but not all the delegates came to your talk. However, this model seems to be contradicted by experimental findings concerning how ordinary subjects process scalar terms. So, in a set of experiments Storto and Tannenhaus tracked the eye movements of subjects when exposed to a grid of pictures and a sentence relating to the pictures which contained a scalar term. To give an example of the kind of test they ran: hearers were exposed to a two-by-three grid like the following: They were then exposed to part of a sentence, for example ‘The car or the clock is next to a ...’ and their eye-movements were tracked to this point. The result was that by this stage in the sentence the majority of subjects were already focused on the pictures in the column on the left-hand side. What this seems to show is that by this stage in sentence processing subjects were already processing ‘or’ not in its weaker inclusive sense (one or the other or both) but in its stronger exclusive sense (one or the other but not both). For it is only if ‘or’ is read exclusively in this sentence fragment that one has enough information to rule out the column on right of the grid, which depicts the same pair of objects and thus would serve to make true an inclusive interpretation of ‘the car or the clock is next to a ...’. The findings from these eye-tracking experiments, together with cases of apparent sub-sentential assertion and direct access to metaphorical interpretations, seem to show that on at least some occasions context acts to affect content before sentence meaning has been recovered. That is to say, they seem to show that pragmatic effects can occur at a local (word- or phrase-based level) as well as a global (sentence) level.9 Yet this runs counter to the priority apparently assigned to literal meaning by Grice, thus evidence about

32 Emma Borg

the psychological processing of linguistic stimuli seems to run counter to Grice’s proposal. Turning now to minimal semantics, it seems that a similar kind of challenge – stemming from the psychological evidence concerning linguistic understanding – can be made against the theory. Indeed this seems to be the basis of Recanati’s objection (2004, p.20) to the minimalist approach in terms of what he calls the ‘availability principle’: (AP) What is said must be intuitively accessible to the conversational participants (unless something goes wrong and they do not count as ‘normal interpreters’). The availability principle is one which Recanati suggests any feasible theory of semantic content must respect, yet it is a principle which minimalism clearly flouts. For, as noted when we introduced minimalism at the start of the chapter, even if there are such things as minimal contents they are not the kinds of things which speakers and hearers consciously entertain in most normal conversational exchanges. If I hear you say ‘There’s nothing to eat’ or ‘You won’t die’, the contents I am likely to consciously entertain include there is nothing to eat in the fridge or you won’t die from that cut, they don’t include there is nothing to eat (in some contextually unconstrained domain) or you won’t die. Thus minimal contents are not (usually) what conversational participants consciously entertain on hearing an utterance but also nor are they the things agents (usually) bring to consciousness when reflecting on how assignments of utterance meaning were made. If asked how I got to there is nothing to eat in the fridge I’m likely to appeal to facts like your looking in the fridge, but I’m unlikely to appeal to the minimal content the minimalist assigns the uttered sentence. It seems then that minimal contents are simply not available to normal subjects and as such they cannot, Recanati objects, play the role of semantic content. According to Recanati (2004, p. 20) the availability constraint “leads us to give up Minimalism. That is the price to pay if we want Availability to be satisfied”. So, when we turn to look at what is in the minds of subjects when they are engaged in linguistic processing it seems that what we find is not Grice’s picture of grasp of literal meaning plus an act of inference to speaker meaning, nor is it the minimalist’s minimal propositions. Whether we are appealing to conscious, first-personal content or some less immediate notion of unconscious or occurrent content, the psychological evidence seems to run counter to both theories. Yet this is problematic since, as noted in the previous section, both accounts subscribe to the view that semantic content is metaphysically dependent on psychological content. The worry is that, in the absence of a story about why one cannot move from psychological evidence to semantic theorizing, the current evidence shows that both accounts must be rejected.

Semantics and Psychological Evidence

33

4 The response So, how worried should the Gricean or the minimalist be by the suggestion that their theories fail to fit with relevant psychological evidence? Well, Kent Bach (2006b, p. 25) has argued that the Gricean shouldn’t be worried at all, for, as he writes: Grice did not intend his account of how implicatures are recognized as a psychological theory nor even as a cognitive model. He intended it as a rational reconstruction. When he illustrated the ingredients involved in recognizing an implicature, he was enumerating the sorts of information that a hearer needs to take into account, at least intuitively, and exhibiting how this information is logically organized. He was not foolishly engaged in psychological speculation about the nature of or even the temporal sequence of the cognitive processes that implements that logic. Now, on one reading, Bach’s response to the challenge of the last section is, I think, the same as the one I want to propose later on behalf of the minimalist; however, I think there is also another reading where it is perhaps a little more problematic. My worry is that talk of ‘rational reconstruction’ runs the risk of driving too great a wedge between the semantic theory and the psychological theory, for if all one is offering is a way in which speaker meaning could be recovered, with no requirement that ordinary speakers do recover meaning in this way, then we seem to be sliding away from a picture which treats semantic content as dependent on psychological content and towards an account which treats semantic content and psychological content as more or less independent of each other (i.e., moving towards option (2.1.1)). A rational reconstruction which makes absolutely no psychological speculation runs the risk of providing a theory of meaning which might be alright for Martians but simply doesn’t hold true for us. Perhaps then we could respond to the challenge here not by denuding our theory of all psychological speculation but by widening our understanding of what counts as psychological evidence. The response I want to make on behalf of the minimalist is that we view minimalism as providing a theory of the form and content of the language faculty, where this is taken to be a genuine component of the cognitive make-up of ordinary agents (so, semantics is at least in part a branch of individual psychology, as Chomsky recommended for syntax). However, minimalism is not a theory of conscious content, nor even a theory of the occurrent mental states involved in given acts of linguistic processing. Rather what minimalism specifies is the content a competent language user is guaranteed to be able to recover, given adequate lexical resources plus the proviso that attentional resources are not diverted from processing literal meaning, and agents are guaranteed

34

Emma Borg

the possibility of recovering sentence-level content because the theory they have cognized is one which trades, ultimately, in sentence-level meanings. The claim is then that, at some level of specification, there are structures in the mind/brain which represent the basic elements of the minimalist theory (word meanings, syntactic rules for constructing sentences, and semantic rules for determining sentence meanings from those word meanings and syntactic structures). Furthermore, the deductive processes posited by the theory en route to determination of sentence meaning will have to be mirrored by operations within the mind, where this is most easily understood as being mirrored in the causal interactions between brain structures.10 However, even if minimalism is the right way to characterize the knowledge underpinning our linguistic competence, it doesn’t seem that the theory need be committed to the claim that, in every instance of communicative success, the theory acts all the way to deliver sentence level content. Sometimes, it seems, we go part way towards constructing a representation of sentence-level content but we stop, either because we are distracted or, more often, because fragments of meaning are all we need to proceed to a guess about what the speaker is trying to convey. The minimalist is happy to concede that what we are really interested in, in communicative exchanges, is getting at what the speaker intends to communicate and this is often very different to the literal meaning attaching to the sentences she utters. Thus the minimalist should be happy to allow that sometimes hearers simply stop thinking about semantic content before the language faculty has had a chance to deliver sentence-level content (i.e., a hearer should stop semantic processing whenever she has enough evidence to get at whatever the speaker was trying to convey).11 Yet this doesn’t show that the theory realized in the mind of the language user is not one designed to deliver sentence-level content nor that it is one which doesn’t trade essentially in complete propositions. All it shows is that the psychological processes realizing the theory are sometimes stopped on the way to delivering sentence-meaning. So, the claim is that minimal semantics does embrace both (2.3.1) and (2.3.2), although only on a refined reading of the epistemic claim: psychological evidence is relevant to semantic theory construction, but not necessarily psychological evidence concerning how a particular utterance is processed. Minimalism is thus a theory which is open, at least in principle, to confirmation or disconfirmation by the psychological evidence, but this must be evidence about what subjects know about their language, not merely evidence about how they come to grasp what speakers are (pragmatically) trying to convey. Even if hearers may sometimes be able to grasp an instance of speaker meaning without calculating the semantic content for the particular sentence uttered, nevertheless, according to minimalism, it is possession of a theory of meaning which ultimately trades in sentencelevel contents that explains (at least in part) why subjects are in a position to recover speaker meaning at all.

Semantics and Psychological Evidence

5

35

Clapp’s naturalistic challenge to minimalism

Finally, then, I’d like to close by suggesting how the account of minimalism from the previous section can answer an objection raised by Lenny Clapp. According to Clapp 2007 minimal semantics fails to give any account of how we select a correct semantic theory. By treating speaker intuitions about content as orthogonal to theorizing about content, we deprive the account of any way to ground theory selection, for if we can’t appeal to intuitions about what speakers and hearers think sentences mean, what other grounds could we have for saying that grass is green is the right interpretation of the sentence ‘Grass is green’, as opposed to some quite different proposition? Given the account of minimalism proposed in the previous section, I hope the answer I’d like to give to this grounding question is clear: an interpretation is the right/wrong one if it matches/fails to match the one generated by the semantic theory actually possessed by ordinary speakers.12 What makes ‘grass is green’ mean that grass is green is that this is the content delivered by the semantic theory contained within a subject’s language faculty; grounding is thus achieved by appeal to the contents of the mind. Clapp, however, considers and rejects this proposal, since such minimal contents will not serve to tell us for all possible worlds whether a given sentence is true or false in that world. Take the sentence ‘The cat is on the mat’: is the minimal proposition that the cat is on the mat true or false in a world where the cat is floating a few centimetres above the mat? That we don’t know what to say in this kind of case shows, Clapp (2007, p. 259) contends, that minimal propositions are not genuine candidates for semantic content.13 However, the problem with this objection, as I’ve suggested elsewhere (Borg 2004a), is that it seems that no plausible candidates for sentence level content will meet such a constraint. For instance, if we are worried by the presence of the preposition ‘on’ we might suggest that it should be given a contextually enriched, pragmatic sharpening. Though it is a little unclear exactly how this might go, one suggestion would be that (leaving aside the issue of incomplete definite descriptions) the enriched proposition should be something like the cat is on the mat in the normal sense of being on associated with cats and mats. Now (even allowing that there are such things as ‘normal senses’ here) there is a problem of grain, for it is not obvious that such senses should attach at the level of cats and mats or to something more fine-grained. The normal way in which a cat sits on a mat might not be identical to the normal way in which a Manx cat sits on a doormat (specifically, while the former might specify tail location, the latter clearly won’t). However, leaving this problem to one side, it still seems that the original worry can resurface for this contextually enriched proposition, for we can still ask is the proposition that that cat is on that mat in the normal sense of being on associated with cats and mats true in a world where the cat has most of three legs on the mat but most of one leg off it?

36 Emma Borg

Perhaps then we should opt for something which makes the context-sensitivity of the proposition more evident, say the cat is on the mat in the contextually salient sense of on. However, even here it seems open to question whether the requirement to provide a determinate answer in all possible worlds is satisfied. For instance, imagine that you are looking for the cat and I assert ‘The cat is on the mat’, where this means that the cat is on the mat in the contextually relevant sense. Even having supplied a context it’s not clear exactly when my utterance is true or false, for instance imagine the cat is wholly on the mat but the mat has been moved slightly to the left. Is what I say true, or is it false since it seems that the contextually relevant sense of ‘on’ here should specify something about the cat’s location? And assuming that we want to say the utterance is true in this case, we can then ask how far to the left can the mat shift and what I say remain true: is my utterance still true if the cat is on the mat in a different room, a different house? Intuitions vary here, but the point is simply that appealing to ‘contextually relevant senses’ doesn’t seem to help for there is no reason to think that the addition of a contextually relevant sense will result in a proposition which tells us, for all possible worlds, whether or not the proposition is true or false at that world. For, given any contextually salient sense, it seems we can always gerrymander a situation where the answer is simply unclear, just as we could for the original minimal proposition. The problem we are homing in on here is, I think, that it is only maximally specific contents which are capable of collapsing knowledge and verification (i.e., ensuring that if one knows a truth condition one can verify whether it is satisfied in every possible world) but there is no reason to think that the literal meanings of sentences are given by maximally specific propositions. So, if no plausible candidates for sentence-level content are capable of meeting the proposed condition on semantic contents (i.e., that they tell us for all possible worlds whether they are true or false at that world) then I think the minimalist is justified in getting off this particular bus before it gets going. Clapp is quite right to think that minimal contents don’t suffice to tell us, for all possible worlds, whether they are true or false at those worlds, but so long as they do allow us to tell, for a vast range of clear-cut cases which truth-value holds, that is all we can (and should) ask for from semantic contents. As I’ve suggested elsewhere, to think otherwise seems to me to yield to verificationist urges which are best kept in check in the semantic realm.14

6

Conclusion

Semantic minimalism owes much to its Gricean predecessor. Specifically both approaches make serious use of the claim that semantic content is not speech act content and treat the objects of semantic theorizing as minimally contextually affected (i.e., allowing only a highly constrained set of pragmatic processes to affect semantic content, such as reference determination

Semantics and Psychological Evidence

37

for indexicals and demonstratives). I’ve argued here that minimalism and Gricean semantics also share another point of contact, for both approaches adopt a view of semantic content as metaphysically dependent on psychological content. Given this, however, I suggested that both accounts were likely to prove vulnerable to objections concerning the extent to which psychological evidence supports the theories. For if semantic content depends on psychological content then, ceteris paribus, we would expect psychological evidence to be relevant to semantic theorizing. This apparent vulnerability was, I suggested, at the heart of the challenge to Grice from apparent cases of the recovery of speaker meaning without the recovery of sentence meaning, and to minimalism from Recanati’s ‘availability principle’. However, I have tried to argue here that, even if the opponents of Grice and minimalism are right and the theories cannot be taken to specify either conscious level content nor unconscious, occurrent content in at least some cases of communicative success, still this need not entail the rejection of the theories. For what they should be taken as proposing is an account of the form and content of the knowledge possessed by competent language users, not necessarily an account of the processing strategies deployed on each and every occasion en route to grasp of speaker meaning. What makes a subject a competent speaker of a language, and thus what underpins her communicative success in general, is (according to these approaches) tacit knowledge of a theory which ultimately trades in sentence-level contents. That sometimes successful communication might occur without calculation of such content is not, then, something the theories must rule out. So minimalism should, I think, be construed as a theory of what’s in the mind, not what is conscious nor even necessarily what plays an inferential role in arriving at the interpretation of a given speech act. This is not to say that the theory cannot ultimately be confirmed or disconfirmed by psychological evidence, but to recognize that there is no direct route from evidence about the way in which a specific instance of communicative success comes about to the general form of a cognized semantic theory. I think this broad appeal to psychological evidence is sufficient to avoid Clapp’s naturalistic challenge, since it grounds a correct semantic theory in the mind of the language user. With respect to Clapp’s ensuing challenge that minimal contents cannot tell us for all possible worlds whether a sentence is true or false in that world, I conceded this point but suggested that it was a constraint which semantic content should anyway not be asked to meet.

Acknowledgements Thanks to audiences at the Joint Session of the Aristotelian and Mind Associations, Bristol 2007 (particularly Lenny Clapp) and the University of Manchester, January 2008. Research for this chapter was made possible by the award of a Philip Leverhulme Prize.

38

Emma Borg

Notes 1. See Borg (ms, chapter 1). 2. Personally, I’d add a third clause to the definition of minimalism: Semantic content is delivered by a modular (computational) language faculty. This entails that there can be no appeal to the intentional states of current speakers at the semantic level. This is something I argued for in Borg (2004a and 2004b). However, the modularity aspect of minimalism is not something which other advocates have endorsed and it is not crucial to the debate in this chapter. 3. Note that Soames post-2002 is not an advocate of the minimal approach. This point is explored further in Borg (ms, chapter1). Minimalism is to be distinguished from theories which claim semantic content must be entirely free from contextual effects, e.g., Katz (1977) and Bach (2006). In permitting only standard context-sensitive expressions the approach also differs from the kind of hidden-indexical account proposed by Jason Stanley, e.g., Stanley (2005). Note however that in sharing the minimalist’s assumption that all contextual effects on semantic content must be syntactically marked, I’d take the hidden indexical view to be aligned more closely with minimalism than contextualism, but see Cappelen and Lepore (2005) for a different take. 4. These options are not meant to be exhaustive (for instance, some kind of twoway dependency could be envisaged). However, I do think they cover the main positions in logical space. 5. Davidson (1984). We should note, however, as Äsa Wikforss pointed out in conversation, that it would probably be a mistake to take metaphysical independence as the official Davidsonian line, given other remarks he makes, such as his requirement that conversational participants construct passing theories of one another (Davidson 1986). 6. At its most extreme, a claim of epistemic dependence might amount to what Davies (2006) disparagingly calls ‘cognitive scientism’ – the view that all the relevant facts are to be revealed via experimentation and thus that semantics is nothing over and above the collation and ordering of experimental findings. However it seems that (2.3.2) need not be read in such strong terms. 7. Though see Avramides (1989) and Garcia-Carpintero (2001) for caveats concerning the usual reductive reading of the dependency claim in Grice. 8. Common cases where syntactically elided material is allowed include question and answer contexts; for instance, if asked ‘Would you like a drink?’ and you respond ‘No thanks, I wouldn’t’ your answer is commonly held to contain the syntactically present but unvoiced material like a drink which is easily recoverable from the immediate linguistic environment. 9. See also Sauerland (2004) for an alternative kind of argument against the global Gricean account of pragmatic influence, and Russell (2006) for a response. 10. Of course there are serious and seriously difficult questions here concerning how we move between an abstract statement of a semantic theory and an account of psychological processes (let alone talk of brain structures); see Davis (1987) for an illuminating discussion of this point. However I hope that the overarching point – that there is no simple move from an assumption about the kind of semantic theory realized in the mind of a subject to claims about the kinds of processes that a subject must undergo on a specific occasion of communicative success – can be made without too deep an excursion into these murky waters.

Semantics and Psychological Evidence

39

11. There are, it should be noted, implications of this point for the kind of modularity picture I advocated in Borg (2004). To put things crudely, it can’t be the case that the language faculty remains entirely encapsulated until the point of outputting a sentence-level content. Rather the picture is one where the outputs of the language faculty are available at incremental levels, so that, as it were, other modules or central-processing systems can ‘see’ the construction of sentence-level meaning stage by stage and can utilize the sub-sentential fragments of meaning which are going into the construction of sentencemeaning. The modularity claim will then be that, although pragmatic and semantic interpretation processes run in parallel (rather than the kind of sequential picture seen in the original rendition of Grice’s view), with pragmatic processes able to operate on sub-sentential clauses before the semantic analysis of the sentence is complete, still no pragmatically enhanced reading is permitted to feed back into the semantics module to effect the semantic analysis of the sentence. 12. The answer here then is the same as that suggested to the over-generation problem for T-theory accounts proposed by Larson and Segal (1992). 13. As Clapp notes, this general line of argument follows Searle (1978). In a footnote (n.10) Clapp puts this challenge as follows: ‘Borg’s view merely replaces the question “Why is P, and not P*, the semantic content of ‘The cat is on the mat’?” with “Why is P, and not P*, the semantic content of [the Mentalese sentence] THE CAT IS ON THE MAT?” To meet the naturalistic challenge Borg must now answer this latter question.’ However, put this way the challenge sounds rather like a request to provide a naturalistic theory of mental content (say like Fodor’s theory of asymmetric dependence for which see Fodor 1990) and while this is a perfectly fair request it doesn’t seem that the lack of such a theory is an embarrassment solely for minimalism. 14. The question of the correct constraints on genuine truth-conditions is taken up again in Borg (2009).

Bibliography Avramides, A. (1989) Meaning and Mind: An Examination of a Gricean Account of Meaning (Cambridge, Mass.: MIT Press). Bach, K. (2006) ‘The Excluded Middle: Semantic Minimalism Without Minimal Propositions’, Philosophy and Phenomenological Research, 72, 1–9. —— (2006b) ‘The Top Ten Misconceptions about Implicature’, in B. Birner and G. Ward (eds) Drawing the Boundaries of Meaning: Neo-Gricean Studies in Pragmatics and Semantics in Honor of Laurence R. Horn (Amsterdam: John Benjamins). Borg, E. (2004a) Minimal Semantics (Oxford: Oxford University Press). —— (2004b) ‘Formal Semantics and Intentional States’, Analysis, 64, 215–23. —— (2007) ‘Minimalism versus Contextualism in Semantics’, in G. Preyer and G. Peter (eds) Context-Sensitivity and Semantic Minimalism: New Essays on Semantics and Pragmatics (Oxford: Oxford University Press). —— (2009). ‘Must a Semantic Minimalist Be a Semantic Internalist?’, The Proceedings of the Aristotelian Society Supplementary Volume, 83, 31–51. —— (ms) Semantics from Syntax: In Defence of Minimal Semantics. Cappelen, H. and Lepore, E. (2005) Insensitive Semantics: A Defense of Demantic Minimalism and Speech Act Pluralism (Oxford: Blackwell). Carruthers, P. (1996) Language, Thought and Consciousness: An Essay in Philosophical Psychology (Cambridge: Cambridge University Press).

40 Emma Borg Clapp, L. (2007) ‘Minimal (Disagreement about) Semantics’, in G. Preyer and G. Peter (eds) Context-Sensitivity and Semantic Minimalism: Essays on Semantics and Pragmatics (Oxford: Oxford University Press). Davidson, D. (1984) Inquiries into Truth and Interpretation (Oxford: Clarendon Press). —— (1986) ‘A Nice Derangement of Epitaphs’, in E. Lepore (ed.) Truth and Interpretation (Oxford: Blackwell). Davies, M. (1987) ‘Tacit Knowledge and Semantic Theory: Can a Five Per Cent Difference Matter?’, Mind, 96, 441–62. —— (2006) ‘Foundational Issues in the Philosophy of Language’, in M. Devitt and R. Hanley (eds) The Blackwell Guide to the Philosophy of Language (Oxford: Blackwell). Dummett, M. (1993) Origins of Analytical Philosophy (London: Duckworth and Cambridge, Mass.: Harvard University Press). Evans, G. (1982) The Varieties of Reference (Oxford: Clarendon Press). Fodor, J. (1990) A Theory of Content and Other Essays (Cambridge, Mass.: MIT Press). Garcia-Carpintero, M. (2001) ‘Gricean Rational Reconstructions and the Semantics/ Pragmatics Distinction’, Synthese, 128, 93–131. Gibbs, R. (2002) ‘A New Look at Literal Meaning in Understanding What Is Said and Implicated’, Journal of Pragmatics, 34, 457–86. Grice, P. (1989) Studies in the Way of Words (Cambridge, Mass.: Harvard University Press). Horn, L. (2006) ‘More Issues in Neo- and Post-Gricean Pragmatics: A Response to Robyn Carston’s Response’, Intercultural Pragmatics, 3, 81–93. Katz, J. (1977) Propositional Structure and Illocutionary Force (New York: Thomas Y. Crowell). Pinker, S. (2007) The Stuff of Thought: Language as a Window into Human Nature (London: Allen Lane). Recanati, F. (2004) Literal Meaning (Cambridge: Cambridge University Press). Russell, B. (2006) ‘Against Grammatical Computation of Scalar Implicatures’, Journal of Semantics, 23, 361–82. Sauerland, U. (2004) ‘Scalar Implicatures in Complex Sentences’, Linguistics and Philosophy, 27, 367–91. Soames, S. (2002) Beyond Rigidity (Oxford: Oxford University Press). Stainton, R. (2006) Words and Thoughts (Oxford: Oxford University Press). Stanley, J. (2000) ‘Context and Logical Form’, Linguistics and Philosophy, 23, 391– 434. —— (2005) ‘Semantics in Context’, in G. Preyer and G. Peter (eds) Contextualism in Philosophy: Knowledge, Meaning, and Truth (Oxford: Oxford University Press). Storto, G. and Tanenhaus, M. (2004) ‘Are Scalar Implicatures Computed Online?’, Proceedings of WECOL.

3 Naturalism in the Philosophy of Language; or Why There Is No Such Thing as Language John Collins

With the progress of science, the realm of physics has so expanded that it seems to be limited only by the limitations of the method itself. Einstein (‘Fundaments of Theoretical Physics’, p. 324)

1

Introduction

My brief is to articulate a purely methodological construal of naturalism, apply the notion to the case of language, and, upon that basis, argue that naturalistic inquiry into language has no need for the notion of language.

2

Naturalism

In contemporary philosophy, naturalism, understood as a methodological doctrine, is often allied, if not conflated, with the metaphysical doctrine of materialism or physicalism.1 I shall argue that the metaphysical thesis is of doubtful coherence and, fortunately, nothing a naturalist might want to say depends upon any such doctrine. The naturalism I seek to promote is a purely methodological attitude that constitutively eschews a presumption of metaphysical fixity, either in regard to that which is meant to be naturalistically kosher or that which is ‘real’ enough to be ripe for naturalization. Indeed, my target is the tendency to pursue philosophically naturalistic programs as if we already knew the resources available to the naturalizing theory (‘Causes and objects in space-time will do’, as it were) and as if that which is to be naturalized is itself in good standing, immune to fundamental alteration, fragmentation, or even elimination given the progress of our science. Once the presumed fixity is removed, there appears to be precious little left to materialism beyond a commitment to naturalistic inquiry. 41

42

John Collins

It will be noted that such a stance is also at odds with those who are motivated to unhinge the ‘natural’ from scientific inquiry on the grounds that some or other domain (intentionality, consciousness, normativity, etc.) is at odds with materialism. Thus, on the view to be propounded, if a naturalism that appears to imagine a fixed material natural order is not the right conception anyhow from a properly naturalistic perspective, then any position that merely rejects a materialist construal of naturalism might still be justly impugned on unanswered naturalistic grounds. The natural is not merely the familiar.2 Before we get to the positive methodological construal of naturalism, let us see why it should be free of any materialist commitments. There are, suffice it to say, many ways of formulating materialism. For present purposes, I’m just concerned to reject the idea of fixity advertised before, i.e., the notion that science has provided us with the fixed resources by which we may judge what is real. I shall be content with registering a fundamental complaint, after Chomsky and others, which, to my mind, has yet to be properly addressed. Materialism, understood as the scientific world-view of naturalism, is at its most anaemic when it merely denies any kind of mind-body dualism. It is unclear, however, whether the world-view can be more full-blooded than mere monism. If not, then materialism amounts to nothing more than a commitment to the unity of nature, which is a regulative ideal, I take it, not a metaphysical doctrine, at least not one vindicated by any scientific pursuit. Chomsky (1968/72, 1988, 2000) has offered a compelling consideration for such a deflation. He points out that, post-Newton, we don’t have a fixed conception of body in terms of which we might find mind (understood in the broadest sense) something alien; indeed, post-Faraday, it is not even clear if physical theory should sanction things (cf., Crain and Mellor 1990; Van Fraassen 1996; Strawson 2003; Ladyman and Ross 2007). In other words, notwithstanding much philosophical dispute on the matter, while mental phenomena are indeed not presently integrated into the domains of lower level sciences, it remains unclear whether materialism can be coherently formulated as a metaphysical doctrine in terms of which mental phenomena are inherently suspect. This situation need not perturb the naturalist (as opposed to the metaphysician). A naturalist may view science as an on-going activity that has no world-view at all, let alone one according to which the mind is peculiarly inexplicable. In response to this kind of consideration, Smart (1978), Melnyk (1997), and Lycan (2003) have suggested, in their somewhat different ways, that the portion of higher physical theory plausibly relevant to the naturalization of the mind might well be complete (‘Quarks or strings don’t matter’ might be their slogan); or at least our epistemic commitment to materialism should be equal to our commitment to the relevant portion of physical theory. In this sense, there is a mind-body problem after all, with body indexed to the

Naturalism in Philosophy of Language

43

relevant physical theories. There are three principal problems with this kind of response to Chomsky. First, it is not clear that there is a particular (higher) portion of physical theory that is uniquely germane to the naturalization of the mind. Penrose (1989) among numerous others, for example, has speculated on the relevance of quantum mechanical effects to the structure of mental properties. Surely a materialism/naturalism should not rule out such hypotheses, whatever their plausibility, in the name of materialism. Second, and more generally, the very idea of portions of science being complete in the relevant sense is somewhat dubious. All disputants here agree, I take it, that the arrows of explanation go in one direction: from fundamental physics upwards. So, although high-level theories are indeed ‘complete’ in the sense of being empirically adequate over phenomena where the primitive concepts apply (e.g., Newton’s laws do hold over phenomena where the classic ideas of mass and force apply), just how we understand that theory at the end of inquiry, as it were, will be a function of our fundamental theory (it will be a limit of the final theory, if all goes well). In this light, what might appear to be quite incompatible with a theory at a particular stage of inquiry, might not be incompatible at all once that theory is integrated with lower level theories (cf., the resolution of the conflict between electromagnetic theory and inertial mechanics under special relativity). In this sense, the isolated problematic phenomena require a new conceptual understanding rather than some new knowledge about elementary particles, say. Third, a complaint of a lack of integration of mental phenomena with any portion of physical theory, completed or not, must, it seems, be predicated on a reasonably clear theoretical conception of that which remains isolated, viz., mentality. 3 For comparison, consider the explanatory gap between quantum mechanics and general relativity. The gap is so serious and has given rise to so much fruitful work precisely because we have excellent theories of both domains and know precisely the nature of the inconsistency. Compare, also, the historical case of chemistry. The in principle reduction of chemistry to quantum mechanics is one of the glories of twentieth century science. The reduction is only in principle, for it is not as if chemistry can be done in terms of the field equations of electrodynamics; rather, we now understand the chemical bond in terms of the field equations and can explain, in those terms, the stability of some of the most elementary molecules (not something as complex as, say, DNA). In a sense, then, we know that chemistry depends upon the physics: chemistry is the way it is because physics is the way it is. This is the relevant sense of reduction. Such reduction, however, was only possible given the highly articulated chemical theory of valence and the periodic grouping of the elements, which in fact inspired the developments in physics.

44

John Collins

In this light, at our current level of understanding, talk of a lack of fit between physics and the theoretically intractable phenomena of consciousness and intentionality should strike us as odd, for it is not so clear if there are any uniform phenomena in view. For example, intentionality putatively covers a veritable heterogeneity of states, from the systems of insect navigation studied by Gallistel (1990) to the visual indexing hypothesized by Pylyshyn (2002 and 2007) to one’s thinking about Vienna or Zeus. There might be a unity to these phenomena, but there is surely no naturalistic ground to insist that there is. Likewise, the great variety of qualitative experience does not, on the face of it, signal a unity to be discretely explained (compare the difference between hearing a noise and feeling a pain). In short, the signature properties of the mental might well fragment under inquiry. For sure, the mind remains obscure, and to that extent we have an anaemic mind-body problem, a problem that looks decidedly less metaphysically portentous if phrased in terms of cognitive science being in its infancy and remaining separated from the more developed sciences. Until the phenomena in question are much more sharply delineated, talk of gaps strikes me, at least, as misleadingly premature. I should say that I suspect that all Smart et al. wish to claim is that mental phenomena remain isolated from the kind of integrated explanation one finds elsewhere in the natural order and that quantum mechanics, say, is hardly likely to be pertinent. Quite so! My present point is merely that formulations of this thought require no metaphysical gloss, which does nothing, it seems, other than encourage scepticism of the valuable work that is proceeding, as if science in this domain should be uniquely held to the highest metaphysical standards that have never been adopted in the physical sciences. In this sense, then, the natural order is simply that about which we can get deep integrated theories, that which will be true or ‘completed’ at a certain limit, with the hope of eventual overall integration.4

3

Methodological naturalism

In opposition to a naturalism that thinks in terms of a fixed base or a mentalism immune to elimination or fundamental distortion, I want to commend a methodological naturalism. Let us characterize this notion by way of the following features: i. Openness to scientific advance: naturalism involves no presumptions about the categorical resources available to kosher explanation. Our putative fundamental categories – object, space, cause, etc. – may wax and wane. ii. Possibility of closure: naturalism involves no presumption that every salient phenomenon submits to explanation.

Naturalism in Philosophy of Language

45

iii. Conceptual invention/elimination: the concepts employed in successful scientific inquiry are typically invented for purpose and have only a loose correlation with commonsensical notions, which are often eliminated or retained just for motivational purposes. iv. Methodological monism: naturalism involves no presumption that distinct domains require distinct theoretical or evidential approaches. Any such differences there might be are to be arrived at, not a priori determined. v. Reduction/explanation: science typically proceeds, not by classical reduction, but by ‘in principle reduction’, where we can explain why a domain of phenomena is the way it is, without any predicate-topredicate map or a practicable deduction. These features are largely negative, in that they only tell us that naturalistic inquiry is not lumbered with certain constraints that lie outside of the animating principles and methods of on-going inquiry. This is not to insist, in a relativist vein, that there are not any a priori constraints that transcend given theoretical frameworks and epochs. In this regard, we may venture a more positive understanding, which perhaps only currently applies to the hard sciences, although one that may be happily understood to be an ideal for all scientific inquiry: vi. Domains/models: the domains of natural sciences are determined by on-going inquiry and invariably involve a selecting of relevant phenomena largely determined by the theories themselves so that a theory has first application to an idealized model. vii. Integration: the ultimate goal of science is the explanatory integration of different domains (consilience); at a given moment in time, there is only a defensible presumption that a putative lower domain is explanatorily in order relative to higher domains, e.g., the physics of the time was inconsistent with developments in both biology (evolutionary theory) and chemistry (chemical bonding); it was the physics that was wrong. viii. Principles in place of metaphysics: the hard sciences seek unifying principles (symmetries/invariances) in ‘hidden parameters’ (i.e., not expressed in the description of the phenomena; compare Cartesian kinematics to the Newtonian explanations in terms of force and mass) instead of laws that describe phenomena. The principles require no metaphysical underpinning or elaboration; their content flows from the mathematical structure that is rigid over the phenomena.5 Now, I do not insist that any materialist would deny any of these claims; my point is only that once they are properly heeded, the metaphysical glosses commonly employed should be ditched, at least on naturalistic grounds. To spell this out in detail would require much greater space than is presently

46

John Collins

available. The principal claim, however, is that the various naturalist programs, whether about the metaphysics of mind, intentionality, epistemology and so on require no conception of a fixed naturalizing base or a fixed conception of what is to be naturalized. Let us now employ these considerations to the specific case of language.

4

Language as a natural phenomenon

Our commonsense conception of language covers a host of disparate phenomena. Any endeavour to gain theoretical traction on this manifold is obliged to fractionate the phenomena and idealize away from the massive interaction effects that produce our normal linguistic behaviour; in other words, our theories determine the relevant domains. The first move in this ‘divide and conquer’ direction in recent times was Chomsky’s (1965) distinction between competence and performance. A whole range of factors enter into performance, many of them perfectly general, such as memory, attention, and communicative intention. Chomsky’s distinction, in part, was an effort to isolate the hypothesized unique linguistic system that underlies certain peculiar features of our performance, and, in the first instance, we are interested in those features simply because they are the ones that submit to theoretical understanding (the drunk looks for his keys under the street light because that is where the light is). So understood, competence designates the capacity to pair discrete, structured messages or meanings with sounds (or some other vehicle, such as hand gestures in sign language) over an unbounded range. The science is concerned with the structure and development of this capacity; its integration with other cognitive capacities; its realization in the human brain; and how it has evolved in the human species alone. Now, it might be that there is no ‘box in the head’ that supports competence so understood; whether there is or not is an empirical question. What seems certain, however, is that progress depends upon some fractionation of the complexity of language, and competence/performance is one natural way to distinguish between the particular and the general over the linguistic domain. More recently, Chomsky has distinguished between the faculty of language in a narrow sense (FLN) and the faculty of language in a broad sense (FLB).6 The thought is that FLN is that which is peculiar to humans (minimally, peculiar in terms of its integration with other systems) and FLB is the set of wider systems that are integrated by the operation of FLN and are likely to be shared to some significant degree with other organisms. In line with the agenda of the minimalist program in linguistics, Chomsky has suggested that a useful methodological assumption is to set FLN to be as simple as possible to meet the empirical conditions of language acquisition and usability, i.e., the realization of FLB. In other words, FLN consists of a general recursive operation (Merge) and any features to which the operation is sensitive. FLN is thus specified as a function (in intension) that maps from selections of lexical items to pairs of structures that are

Naturalism in Philosophy of Language

47

usable or legible at the interfaces with external systems. This agenda gives rise to three leading questions: 1. How perfect or well-designed is FLN to meet the interface conditions of FLB? 2. What is the character of the interfaces and what effect, if any, do they have on FLN? 3. What is the character of the external systems that constitute FLB? Inquiry into these questions is intended to orientate the more traditional concerns of how the system ensemble develops in the child (ontogeny) and how it evolved in the human species (phylogeny).

5

Is there such a thing as a language?

This question might seem bizarre. My intent in asking it, however, is to bring into doubt the familiar philosophical understanding of language as a category distinct from the cognitive resources that support the phenomena we commonly designate as ‘linguistic’. In other words, in a naturalistic spirit, we need to ask whether the notion of a language as a non-cognitive entity (of some description) plays any substantial role in linguistic explanation. I shall argue that it does not. This negative conclusion is not intended as a piece of metaphysical legislation, nor shall I argue for the claim on the basis of the real worries to do with fixing the identity conditions for languages. In other words, along with Burge (1989), Wiggins (1997), and Williamson (2007), I am happy, for the nonce at least, to think that the idea of a mind-independent language is coherent; my only objection is that such an entity looks to be explanatorily vacuous. It bears emphasis that this negative conclusion will hold for idiolects as well.7 It should be a truism that each competent speaker employs a unique linguistic competence.8 In this regard, the idea of an idiolect is less abstract than a shared public language. Even so, an idiolect is an externalist notion, just one that is rightly sensitive to the massive variability between speaker/ hearers. So, while the move to idiolects is certainly an advance, the moral of the arguments that follow will be that we should go the whole hog and ditch the residual externalism. Let us adopt the following naturalistic reality principle: (NRP) At a given stage of inquiry, a category is taken to be (naturalistically) real iff it is either successfully targeted by naturalistic inquiry or essentially enters into the explanations of such inquiry. So, in classical (Newtonian) mechanics, absolute velocity is not ‘real’, its measurement being relative to a particular inertial frame of reference. After the maturation of electromagnetic field theory, lines of force and potentials

48

John Collins

ceased to be ‘real’; for neither essentially enters into the field-theoretic explanations and their measurement, again, is arbitrary. Likewise, after the development of the special theory of relativity, length, shape, simultaneity, etc. ceased to be ‘real’ qua frame-dependent. My submission is that language, construed in an externalist manner, fails to meet NRP, its identification being arbitrary and non-essential to the explanation of the salient phenomena. If this is so, then while languages so construed might be real in some or other sense (in the same way, perhaps, games or pieces of music might be), they are not naturalistically respectable. This thesis, it bears emphasis, does not rule out externalia (sound waves, hand gestures, inscriptions, etc.) playing a role in the acquisition of language and its future usability. The crucial point is that such entities do not offer properties that are either necessary or sufficient for the characterisation of the linguistic structures that enter into current explanations; otherwise put, externalia, meeting as yet obscure conditions, might be causally necessary for the acquisition of language, but they offer no constitutive conditions for the structures posited in current linguistics, i.e., they are not linguistic types. The externalia are not necessary because linguistic structure can be realized in a wholly internal manner, as in private monologue. They are not sufficient because (i) the richness of linguistic structure far outstrips any external signature and (ii) the apparently unlimited heterogeneity of the externalia recruitable in linguistic performance does not admit generalizations mappable onto the linguistic categories; for example, there are no independently specifiable features common between tokens of the sound type /dog/ and tokens of the inscription type ‘dog’ such that both are ambiguous between verbal and nominal categorisation. In other words, the relevant categories are invariant over external differences and so cannot be identified with externalia without eliding that which does remain invariant over the recruitment of externalia, viz., the cognition of the competent speaker/hearer. To be sure, we should have no complaint against one who takes language to be an external phenomenon as a remark on our Lebenswelt, if you will. The issue for natural science is how externalia are invested with linguistic significance. For answers we must look to the peculiar cognitive design of humans, which is invariant over the various targets of our cognitive investments. In riposte, it might be ventured that the linguistic structure is external but simply not readily available to our consciousness; it is some sort of theoretical achievement to discern it (cf., Devitt 2006). This position is perilously close to the behaviourist claim that our behaviour is under the control of external variables, albeit unobvious ones. One can always claim that linguistic structure is external, much as one can always claim that our behaviour is under the control of subtle as the Lord external variables, but if the structure is identifiable only through the cognitive resources of the speaker/ hearer, much as the putative stimuli for our behaviour are only identifiable

Naturalism in Philosophy of Language

49

through the paired responses, then the externalia loose any independence as a proper parameter in any serious explanatory practice. If the structure is to be depicted as genuinely external for our best science, then, lest it become an explanatory dangler, we should be able to identify it independently of the relation it putatively bears to our cognition. We needn’t insist that anything external must be perceptually available, but we must insist that the externalia be something more than a reflection of the cognitive design with which we have already credited the subject. Again, commonsense can continue on its merry way. The assumption that language is an external phenomenon appears to be most often made upon the basis that our conception of linguistic phenomena is answerable to our common conception of what languages are, however vague that might be, as if, that is, there is no science of language in the offing. Other times, the assumption is more elaborated, as a version of Platonism, or nominalism, or behaviourism.9 Rather than chase all these mice down their respective holes, I shall be content to indicate how linguistic explanation is concerned with internal cognitive states alone, i.e., an external language is essential to no extant explanation of any linguistic phenomena. Thus, just how the putative externality of language is unpacked will not affect my overall conclusion. So, let’s look at some key examples, and then draw the general moral. 5.1 Acceptable/grammatical If a theory of language is concerned with the property of being grammatical, then it would seem that the theory is concerned with language (sets of sentences). It would appear, therefore, that the very idea of a ‘grammar’ involves appeal to a language.10 The notion of being grammatical, however, insofar as the notion is of any use at all, is theoretical, not descriptive, i.e., a grammar should explain why we judge the way we do and not some other way; its task is not the mere classification of a corpus.11 Let’s assume that all the linguist has to go on are speakers’ judgements or intuitions of acceptability (whether a string is judged to be fine or not, independent of any reasons offered).12 That is the data set. It is a theoretical task to discern and fractionate what contributes to those intuitions, i.e., what explains the data. We shouldn’t assume that such intuitions are about the language, for an informant need have no conception of a language at all and might well be wholly lacking the relevant concepts. It suffices that informants have robust intuitions of un/acceptability. Why? Because even if every informant thought that their intuitions were about a language – about, say, whether a string was part of the language or not – that would be of no obvious interest to the theorist. Again, the intuitions are our data. We are not treating the informant as a co-theorist who has information about the language; after all, we have no possible access to the language other than through the native informants. It is part of our inherited

50 John Collins

philosophical intuition that there is a language, but it is difficult to see how such an assumption plays a role in our construal of linguistic data, i.e., what needs to be explained. Further, we can no more assume that speakers have an immediate access to the principles that account for their linguistic understanding than we can assume that they have access to the principles that underlie their visual capacity. Thus, the notion of being grammatical does not apply to strings (ink marks, acoustic waves, etc.) in the first instance; rather, a theory of the language faculty (a theory of being grammatical, if you will) brings with it a delineation of the structures that the FLN outputs. How these structures affect the pattern of acceptability, which we take to be a FLB interaction effect, is what we want to discover. Thus, unless we uncritically accept the way linguists sometimes talk, we should find ourselves with no reason to appeal to the notion of an external language or the property of being grammatical, if understood as something other than a way of demarcating the FLN output. Let’s be concrete. Consider the following data: (1) a. Bill loves Mary and Sam b. Bill loves Mary and WHO? c. *Who does Bill love Mary and? The data here are that (1)a-b are acceptable, but (1)c is not. In other words, the data are not the strings themselves, but speaker/hearers’ construals of them, the judgements they elicit in us. We present the data as strings, but we could equally have presented the data in other media or simply thought of them in our head. What unifies the various data sets is that they elicit the same pattern of judgement. Now, the purpose of the theory is not the mere classification of strings but an explanation of the data. Does the direct attribution of ‘un/grammatical’ to the strings (understood as tokens or types) exhibited in (1)a-c explain the data pattern? Transparently not, for the attributions merely describe the data. The only thing that could explain the data is that which explains why speakers judge as they do, and such an explanation is, at best, essentially postponed by attributing properties to the strings, for the speaker/hearers must be cognisant of those very properties if they are to be on explanatory duty. An externalism of grammatical properties, therefore, looks to be explanatorily supererogatory; the externalism confuses what the language faculty (FLN and FLB) enables – the projection of structure onto sounds/marks – with the target of explanation itself – the capacity to project, inter alia. If only because of the recalcitrance of a certain (philosophical) commonsense here, it bears emphasis that the way we would normally speak of a sentence being grammatical, or ill-formed, etc. is utterly irrelevant. The only point of interest is whether the explanations of the phenomena depend

Naturalism in Philosophy of Language

51

upon our taking such claims seriously. They do not; indeed, far from being presupposed by explanation, the externalism precludes explanation, or merely postpones it. 5.2 Infinity It is often remarked that natural languages are ‘infinite’, which suggests a conception of languages as sets of strings. The infinity at issue, however, is simply an artefact of presenting a grammar as a recursive device and ontologically committing to the set so specified. The actual phenomena are that speaker/hearers’ competences are unbounded and their linguistic behaviour is typically novel. The infinity of a language is not a phenomenon in any sense whatsoever. Unbounded novelty, on the other hand, is explained, if we take the speaker/hearers’ cognitive resources to contain a recursive device (specifiable as such). It is redundant to the explanation to posit, in addition, the set of structures generated (in the formal sense), much as it is redundant for a zoologist to posit the set of possible tigers, or a geologist to posit the set of possible volcanoes. The set of all and only the strings of English, say, assuming English is decidable, is not a natural phenomenon and nor is it presupposed by the relevant explanations.13 It bears emphasis that this conclusion does not refute linguistic Platonism. The Platonic linguist is free to posit languages as abstract objects to which competent speakers are somehow related (e.g., via ‘intellectual intuition’). My present point is merely that such a posit of an infinite object is explanatorily empty. Again, the empirical phenomena can only be accounted for if we credit the speaker/hearers with the recursive principles. Taking the principles to be generative of a set (a language) that is the real target of inquiry, as opposed to a needless abstraction, postpones the explanation of how speaker/hearers can enjoy unbounded competence and be continuously novel in their behaviour. 5.3 Structural construal The preponderate data that concern linguists pertain to structural construal: ambiguity, referential dependence, inferential connection, semantic argument structure, etc. Our data in all such cases are speaker/hearers’ acceptability judgements. It is the theorist’s task to fractionate the data so that different aspects of FLB (inclusive of FLN) might be apportioned responsibility. It is natural, however, to speak of a sentence being ambiguous or of a pronoun being bound by an antecedent, etc. Upon reflection, it is clear that strings with such properties are not our data. Just consider ambiguity. The data here are that speakers associate with the relevant string type two or more readings. It makes little sense, for instance, to classify a string as ambiguous even though no one could be brought naturally to associate distinct meanings with it, as if ambiguity were a property of a language quite independent of the users of it; it is the structure of the association and the

52

John Collins

constraints upon it that interest the theorist. Our conception of the strings themselves remains as it was, unadorned by our theorizing. Consider, for example, the following datum: (2) The duck is ready to eat Competent English speakers recognize (2) to be two-ways ambiguous, i.e., they reliably associate with it, on presentation, just two readings. (2) can mean that the duck is ready to be eaten by someone or other or that the duck is ready to eat something or other. In other words, the ambiguity turns on what we take the arguments of eat to be (who is eating what?), neither of which is explicitly encoded in (2). According to a standard analysis, we can present the disambiguating structures as follows: (3) a. The duck i is [OPi ready [PRO to eat ]]14 b. The duck i is [OPk ready [PROi to eat ]] The structures disambiguate the string presented in (2) by providing explicit arguments for the infinitive to eat. Now, the structures in (3) explain why (2) is ambiguous. Where we construe ambiguity to be a cognitive phenomenon, this explanatory relation is straightforward. Competent speakers associate the two syntactic structures with the one phonological form, as witnessed by the phonological null items in the structures in (3). How would the explanatory relation be understood were ambiguity taken to be a property of the string itself, independent of the responses of particular speaker/hearers? It is difficult to say. Clearly, the structures in (3) are not ‘hidden’ in the string presented in (2), as if they described some underlying quasi-physicochemical nature of the string itself. Less absurdly, one might be tempted to take the structures in (3) to be somehow not ‘real’, but more abstract functional descriptions of the divergent roles tokens of the string type displayed in (2) can play, corresponding to its two readings. This move is problematic in a number of respects. First, the phonologically null items posited in the structure are not mere abstracta. There is good evidence that they play a role in sentence processing, which would be impossible if the items were not mentally represented in the positions indicated.15 Second, we want the structures to explain the ambiguity, not merely describe the phenomena; that is, explain why, for instance, (2) is just two ways ambiguous, as opposed to three ways, etc. Again, the structures are only explanatory on the assumption that they are cognitively realized. Third, the very notion that the string is what is real, with the syntax being somehow a higher-level abstraction, turns the proper direction of explanation on its head. The string is real enough, but only as a sound wave or ink marks or a pixel array, which has no more inherent interest than the screech of a braking car, or the wind’s etching on a beach. It is

Naturalism in Philosophy of Language

53

only the recruitment and production of strings by minds that gives them a linguistic life. Thus, the fact we want explained is not the mere string or the occurrence of its tokens, but that the string has the linguistic properties it has, but such properties are simply not in view unless we take the string to be structured in the first place by a cognitive system designed to produce or consume it. It is also worth highlighting that explanation of ambiguity always involves an explanation of negative cases. Consider, for example, this further datum: (4) Mary shot the elephant from Africa If we exclude the reading where the elephant is a projectile, (4) is twoways ambiguous; the adjunct from Africa can modify the DP the elephant (it originates from Africa) or the VP shot the elephant (the shooting was from Africa – the elephant might have originated from India). The explanation offered, therefore, should explain why it is that the adjunct cannot modify Mary. Ambiguity has its own precision in the sense that a sentence is always n-ways ambiguous, for some definite n. A high-level description would not explain why just n-ways; after all, why shouldn’t someone just use (4) in such a way as to say that Mary originated from Africa? This option looks to be excluded, not merely contingently, but due to constraints on how we can interpret the string. In general, to attribute the disambiguating structures to the string itself, however that might be understood, does not explain why speaker/hearers accept differential construals of the strings. After all, since the disambiguating structures are not perceptible, if the strings were to have the relevant structures, we would be obliged to credit the speaker/hearers with the means of discerning them. Here we see the fault in the first externalist step of putting the structure outside the mind, for we appear to be obliged to posit just the same structures in the mind for the ambiguity data to be explained. Similar remarks hold for the other linguistic phenomena mentioned previously. 5.4 Integration Whatever we take language to be, we must accommodate its unique evolution in the human line, its peculiar developmental path in each normal human, its realization in the human brain, and its integration with other faculties of the mind. An externalist conception of language is not inconsistent with these lines of inquiry, but it is clearly orthogonal; for, on the one hand, we have inquiry into the biology of language and on the other hand we have a conception of language as divorced from biology, neither informing its investigation nor being presupposed by it.16 It is open to an externalist to turn this consideration on its head and claim that the integration of linguistics into the biological sciences leads to

54

John Collins

absurdity. Some such tactic has been employed by Katz (1981), who appeals to linguistic duplicate Martians, and by Katz (1990) and Katz and Postal (1991), who have advanced a ‘veil of ignorance’ argument. The general thought is straightforward. We can easily conceive of aliens who appear to share our language, but who are radically different from us as regards their biology. In this scenario, it would seem that the same linguistic theory would hold for us and the Martians; indeed, we can stipulate that it will. Even so, by internalist scruples, we and the aliens could not share a language, because our internal states differ as much as you like from those of the aliens. As a variant on the thought, imagine that cognitive neuroscience discovers that there is in fact a longest sentence that the human brain can represent, i.e., neurophysiology does not realize a recursive device. In this case, the linguist would apparently be committed to the claim that there is a longest sentence of English, which appears to contradict the most basic fact about, say, co-ordination structures. In short, the internalist is guilty of making claims about the biological nature of language from a position of ignorance about biology. These considerations are perfectly in order, as far as they go, but they go no way to establishing the independence of language from human biology. They merely spell out banal consequences of a naturalistic approach to language.17 As regards the aliens, it is of course logically possible that an organism could duplicate an English speaker/hearer while possessing a radically different biology. In itself, that is wholly uninteresting, as interesting as the fact that the manifest properties of water could (logically) be realized by XYZ. If we presume that the possibility is more narrowly nomological, then the interest remains negligible; for we should, trivially, seek an explanation of the competence that is apparently shared by the aliens and us. If none is forthcoming, then the alien-human concordance would be an empirical mystery; after all, why should the alien speak any language, let alone English? Similarly, if cognitive neuroscience were to discover that competence was bounded, then that would pose an empirical problem: how is it that there appears to be no finite bound on co-ordination, say, while neuroscience tells us otherwise? To leap from this conceivable conundrum to the conclusion that there are, nonetheless, infinitely many sentences somehow externally realized (in Plato’s heaven, perhaps) is, forgive me, silly; for now we would be faced with the much harder empirical problem of how any human can understand any sentence at all, bother infinitely many. Katz’s (1981) appeal to ‘intellectual intuition’ is, at best, a label for the supposed problem, not an answer. The motivating assumption of both thoughts is that biology might prove to be in conflict with certain apparently constitutive features of our current conception of language. This thought is true enough but no more threatens the empirical study of language than does the possibility of analogous

Naturalism in Philosophy of Language

55

conflicts in any other area of serious inquiry, such as the conflict between early twentieth century chemistry and physics. *

*

*

The previous considerations should lead us to conclude, I think, that the familiar philosophical notion of an external language is not naturalistically respectable, for it is neither targeted nor presupposed by the explanations current in linguistics; indeed, it remains unclear what conceivable connection the idea of an external language has to actual linguistic phenomena. We remain free to posit languages as abstractions and are equally free to have whatever ontological attitude we like towards them, and generate ontological conundrums much as we do about money or games or pieces of music. It is not the aim of science, however, to cleave to our quotidian ontological scheme. Schiffer (2006) has distinguished between the philosopher’s and the linguist’s sense of language, the difference being that the philosopher is concerned with the use of language, while the linguist is concerned with the cognitive representation of language. Schiffer is correct, of course, to claim that two such perspectives aren’t necessarily in conflict. The problem is that it is unclear what purpose the distinction serves other than to mark a difference of interest, for Schiffer does not question the results of contemporary linguistics. There are, however, not two sets of phenomena in the world, one for the philosopher, one for the linguist. Thus, even if the philosopher is just concerned with use, that is not a license to make assertions about language acquisition, linguistic dispositions, compositionality, and all the other empirical claims philosophers often make, right or wrong, independent of any empirical inquiry. Further, we should surely want an account of the acquisition and cognitive representation of language to bear upon issues of the use of language; after all, what is it to use language other than to deploy one’s linguistic cognitive resources? The issue is not about perspectives, but the facts, and explanation of the linguistic facts does not, it seems, require the philosopher’s perspective on use that involves an external language. That linguistics should have no concern for languages is really not that odd a conclusion if one bears in mind the methodological naturalistic scruples; in particular, that areas of inquiry are free to depart from any intuitively sanctioned ontology. Linguistic phenomena remain to be explained, but such phenomena do not require the idea of a language for their identification or their explanation.

Acknowledgements My thanks go to Tom Greaves, Oskari Kuusela, Guy Longworth, Sarah Sawyer, and especially Georges Rey.

56 John Collins

Notes 1. See, among many others, Hellman and Thompson (1975), Lewis (1983), Field (1992), Jackson (1998), Chalmers and Jackson (2001), and Kim (2005). 2. McDowell (1994), for instance, seeks to overcome a Kantian dualism of reason and cause by rejecting what he calls ‘bald naturalism’, the claim, roughly, that all phenomena are causally explicable. It is, however, perfectly rational to reject such a quaintly wooden conception of naturalism without ipso facto being relived of the Kantian problematic, for reason giving still remains presently inexplicable to scientific inquiry, regardless of the status of causation. The status of causation has been moot ever since Newton’s impugnation of ‘hypotheses’, notwithstanding the common appeal to the notion as if it were the natural relation par excellence (see Koyré 1965 for discussion of Newton on causation, and Russell 1917; Redhead 1990; and Maudlin 2007 for more contemporary scepticism). 3. Levine (2001, pp. 20–1) seeks to answer Chomsky, not by offering a fixed conception of body, but by the thought that only non-mental properties are basic, where mentality is supposed to be made sufficiently clear in terms of its essential marks: intentionality and phenomenal consciousness. First, I don’t think there is anything clear at all about mentality in Levine’s sense. Second, most kinds are non-basic, including all of chemistry, biology, and most of physics. Thus, why specify the mental as being non-basic? It would, presumably, be as equally true to say that only non-biological properties are basic. 4. For the classic account of this kind of model of progress, see Cassirer (1923). In more recent times, a similar account is an aspect of structural realism; see Worrall (1989), Ladyman (1998), and Ladyman and Ross (2007). 5. I mean ‘rigid’ in the sense of Einstein (1950/54, p. 350, cf., Weinberg 1993), where a theory is rigid if alteration of its mathematical constants leads to absurdity, not merely empirical disconfirmation. E.g., we might explain why gravity is inversely proportional to the square of the distance as opposed to the cube of the distance; the value of the relation ceases to be a mere fact of the way the world happens to work. 6. See Hauser et al. (2002) and Chomsky (2005). For wider discussion, see PiatelliPalmarini et al. (2009). 7. For recent discussion of the idiolect notion, see Heck (2006) and Higginbotham (2006). 8. Semantic and phonetic variability between speaker/hearers of the same ‘language’ is now a commonplace, but analogous reflections should also lead us to acknowledge the same for syntax. As Kayne (2000, p. 7) notes: ‘it is entirely likely that no two speakers of English have exactly the same syntactic judgements’. Of course, it does not follow that the linguist should dwell on the variable as opposed to that which is shared. 9. See, e.g., Katz (1981) and Katz and Postal (1991) for Platonism, Devitt (2006) for (a version of) nomianlism, and Quine (1960) for behaviourism. 10. This kind of argument is central to Katz (1981), Soames (1984), Katz and Postal (1991), and Devitt (2006) even though they draw distinct conclusions. Curiously, it never occurs to any of them to ask whether the explanations couched for English data, say, require a substantive appeal to English as opposed to the cognitive design of the speaker/hearers. See the text that follows in the chapter. 11. See Collins (2008, chapters 2, 3) for discussion of this point. 12. Of course, the linguist has much broader data sources, e.g., deficit studies, parsing, acquisition, neurology, etc.; see Jenkins (2004). We may sideline these

Naturalism in Philosophy of Language

13.

14.

15. 16.

17.

57

sources, for they are all cognitive and so cut in the direction of my arguments. Restricting ourselves to judgements is a concession to the externalist. Correlatively, Katz (1981) and Katz and Postal (1991) lay great store by the thought that language gives rise to necessary truths, which cannot be accommodated by the properties of our contingent cognitive-biological make-up. First, the apparently necessary truths are only in the realm of semantics, not syntax, or phonology, etc. Thus, Katz has no argument via necessity for those properties being non-biological, even if semantics is. Second, Katz needs a metaphysical notion of necessity (as opposed to a semantic notion via analyticity, say) for it to do the desired anti-cognitive work, but it is wholly unclear what metaphysical necessity has to do with linguistics. It is surely enough to explain how the semantic relations strike all speakers in the same way. It is not the burden of linguistics to explain the metaphysics that might attend this phenomenon. Analogously, it is not the business of mathematics to explain the necessity that attaches to it. Take ‘OP’ to be an empty operator that moves from object position to form a fused predicate that takes the duck as its single argument, with ‘PRO’ designating an arbitrary eater, if unindexed. Thus, the bracketed material means something like ‘being ready to be eaten by someone/something’. See, e.g., Fodor (1995) and Featherson (2001). Katz (1981) and Devitt (2006) independently claim that cognitive inquiry into language presupposes an independent conception of the language. Both appear to assume (i) that linguistic data are non-cognitive, a claim we are revealing to be badly awry, and (ii) that processing exhausts the realm of the cognitive, such that that any linguistic principle or generalization will immediately be non-cognitive qua not a processing rule. Neither Katz nor Devitt provides an argument for such a narrow conception of cognitive inquiry. The remarks that follow are an echo of Fodor’s (1981 and 2001) thought that linguistics seeks internal constraints on grammars, not mere descriptive adequacy. Fodor goes wrong, though, in arguing that such an approach entails a representational conception of the language faculty. See Collins (2004) for discussion.

Bibliography Burge, T. (1989) ‘Wherein Is Language Social?’, in A. George (ed.) Reflections on Chomsky (Oxford: Blackwell). Cassirer, E. (1923) Substance and Function (Chicago: Open Court). Chalmers, D. (1998) The Conscious Mind: In Search of a Fundamental Theory (Oxford: Oxford University Press). Chalmers, D. and Jackson, F. (2001) ‘Conceptual Analysis and Reductive Explanation’, Philosophical Review, 110, 316–60. Chomsky, N. (1965) Aspects of the Theory of Syntax (Cambridge, Mass.: MIT Press). —— (1968/72) Language and Mind, extended edn (New York: Harcourt Brace Jovanovich). —— (1988) Language and the Problems of Knowledge: The Managua Lectures (Cambridge, Mass.: MIT Press). —— ( 2000) New Horizons in the Study of Language and Mind (Cambridge: Cambridge University Press). —— (2005) ‘Three Factors in Language Design’, Linguistic Inquiry, 36, 1–22. Collins, J. (2004) ‘Faculty Disputes’, Mind and Language, 19, 503–33. —— (2007) ‘Meta-Scientific Eliminativism: A Reconsideration of Chomsky’s Review of Skinner’, British Journal for the Philosophy of Science, 58, 625–58.

58 John Collins —— (2008) Chomsky: A Guide for the Perplexed (London: Continuum). Crane, T. and Mellor, H. (1990) ‘There Is No Question of Physicalism’, Mind, 99, 185– 206. Devitt, M. (2006) Ignorance of Language (Oxford: Oxford University Press). Einstein, A. (1940/54) ‘Fundaments of Theoretical Physics’, in A. Einstein, Ideas and Opinions (New York: Three Rivers Press). —— (1950/54) ‘The Generalized Theory of Gravitation’, in A. Einstein, Ideas and Opinions (New York: Three Rivers Press). Featherson, S. (2001) Empty Categories in Sentence Processing (Amsterdam: John Benjamins). Field, H. (1992) ‘Physicalism’, in J. Earman (ed.) Inference, Explanation and Other Frustrations (Berkeley: University of California Press). Fodor, J. (1981) ‘Some Notes on What Linguistics Is About’, in N. Block (ed.) Readings in the Philosophy of Psychology, Vol. II (Cambridge, Mass.: Harvard University Press). —— (1987) Psychosemantics: The Place of Meaning in the Philosophy of Mind (Cambridge, Mass.: MIT Press). —— (2001) ‘Doing Without What’s Within: Fiona Cowie’s Critique of Nativism’, Mind, 110, 99–148. Fodor, J.D. (1995) ‘Comprehending Sentence Structure’, in L. Gleitman and M. Liberman (eds) Language: An Invitation to Cognitive Science, 2nd edn (Cambridge, Mass.: MIT Press). Gallistel, R. (1990) The Organization of Learning (Cambridge, Mass.: MIT Press). Hauser, M., Chomsky, N., and Fitch, T. (2002) ‘The Faculty of Language: What Is It, Who Has It, and How Did It Evolve?’, Science, 298, 1569–79. Heck, R. (2006) ‘Idiolects’, in J. Thomson and A. Bryne (eds) Content and Modality: Themes from the Philosophy of Robert Stalnaker (Oxford: Oxford University Press). Hellman, G. and Thompson, F. (1975) ‘Physicalism: Ontology, Determination, and Reduction’, Journal of Philosophy, 72, 551–64. Higginbotham, J. (2006) ‘Idiolects: Their Languages and Ours’, in E. Lepore and B. Smith (eds) The Oxford Handbook of the Philosophy of Language (Oxford: Oxford University Press). Jackson, F. (1998) From Metaphysics to Ethics: A Defence of Conceptual Analysis (Oxford: Oxford University Press). Jenkins, L. (ed.) (2004) Variation and Universals in Biolinguistics (Amsterdam: Elsevier). Katz, J. (1981) Language and Other Abstract Objects (Oxford: Blackwell). —— (1990) The Metaphysics of Meaning. Cambridge, Mass.: MIT Press. Katz, J. and Postal, P. (1991) ‘Realism vs. Conceptualism in Linguistics’, Linguistics and Philosophy, 14, 515–54. Kayne, R. (2000) Parameters and Universals (Oxford: Oxford University Press). Kim, J. (2005) Physicalism or Something Near Enough (Princeton: Princeton University Press). Koyré, A. (1965) Newtonian Studies (London: Chapman and Hall). Ladyman, J. (1998) ‘What Is Structural Realism?’, Studies in the History and Philosophy of Science, 29, 409–24. Ladyman, J., Ross, J., and Ross, D., with Spurrett, D., Collier, D., and Collier, J. (2007) Every Thing Must Go: Metaphysics Naturalized (Oxford: Oxford University Press). Levine, J. (2001) Purple Haze: The Puzzle of Consciousness (Oxford: Oxford University Press). Lewis, D. (1983) ‘New Work for a Theory of Universals’, Australasian Journal of Philosophy, 61, 343–77.

Naturalism in Philosophy of Language

59

Lycan, W. (2003) ‘Chomsky and the Mind-Body Problem’, in L. Antony and N. Hornstein (eds) Chomsky and His Critics (Oxford: Blackwell). Maudlin, T. (2007) The Metaphysics Within Physics (Oxford: Clarendon Press). Melnyk, A. (1997) ‘How to Keep the “Physical” in Physicalism’, Journal of Philosophy, 94, 622–37. McDowell, J. (1994) Mind and World (Cambridge, Mass.: Harvard University Press). Papineau, D. (2001) ‘The Rise of Physicalism’ in C. Gillet and B. Loewer (eds) Physicalism and Its Discontents (Cambridge: Cambridge University Press). Penrose, R. (1989) The Emperor’s New Mind: Concerning Computers, Minds, and the Laws of Physics (Oxford: Oxford University Press). Piatelli-Palmarini, M., Uriagereka, J., and Salaburu, P. (eds) (2009) Of Minds and Language: A Dialogue with Noam Chomsky in the Basque Country (Oxford: Oxford University Press). Pylyshyn, Z. (2002) Seeing and Visualizing: It’s Not What You Think (Cambridge, Mass.: MIT Press). —— (2007) Things and Places: How the Mind Connects with the World (Cambridge, Mass.: MIT Press). Quine, W.V.O. (1960) Word and Object (Cambridge, Mass.: MIT Press). Redhead, M. (1990) ‘Explanation’, in D. Knowles (ed.) Explanation and Its Limits (Cambridge: Cambridge University Press). Rey, G. (2001) ‘Physicalism and Psychology; a Plea for a Substantive Philosophy of Mind’, in C. Gillet and B. Loewer (eds) Physicalism and Its Discontents (Cambridge: Cambridge University Press). Russell, B. (1917) ‘On the Notion of Cause’, in B. Russell, Mysticism and Logic (London: Unwin). Schiffer, S. (2006) ‘Two Perspectives on Knowledge of Language’, in E. Sosa and E. Villanueva (eds) Philosophy of Language: Philosophical Issues, Vol. 16 (Oxford: Blackwell). Smart, J. (1978) ‘The Content of Physicalism’, Philosophical Quarterly, 28, 339–41. Soames, S. (1984) ‘Linguistics and Psychology’, Linguistics and Philosophy, 7, 155–79. Strawson, G. (2003) ‘Real Materialism’, in L. Antony and N. Hornstein (eds) Chomsky and His Critics (Oxford: Blackwell). Weinberg, S. (1993) Dreams of a Final Theory (London: Vintage Press). Wiggins, D. (1997) ‘Languages as Social Objects’, Philosophy, 72, 499–524. Williamson, T. (2007) The Philosophy of Philosophy (Oxford: Blackwell). Worrall, J. (1989) ‘Structural Realism: The Best of Both Worlds?’ Dialectica, 43, 99–124.

4 Referring When Push-Comesto-Shove Kevan Edwards

1

Direct reference, Neo-Fregeanism, and the threat of collapse

The anchoring focus of this chapter is a cluster of complaints that have been raised against reference-based approaches to semantics, in particular against the view defended by Scott Soames (2002). I am going to lump the complaints that I have in mind under the heading of the Threat of Collapse (or the Threat, for short). At the heart of the Threat of Collapse is the accusation that various moves referentialists make in dealing with well-known problems undercut the motivations for a reference-based semantics. I am going to propose a way of disarming the Threat. The proposed solution will involve stepping back and saying something about features of a broader framework in which (I believe) a reference-based semantics ought to be embedded. In effect, I am going to use the Threat of Collapse as a jumping off point – an excuse, really – for making some suggestions about how the Direct Reference theorist ought to construe the relationship between language and thought.1 Thus, my aim is two-fold: (i) to make plausible the claim that the Direct Reference theorist has more than sufficient resources to avoid the Threat of Collapse, (ii) to lay some suggestive foundations for future work that embeds Direct Reference in a, roughly speaking, pragmatist conception of the relationship between language and thought.2 Here is the plan for the chapter: The rest of this first section will introduce some features of the dialectical geography, including an abstract outline of how the Referentialist can hope to steer clear of the Threat. Section 2 will gesture towards a way of thinking about the relationship between language and thought. Section 3 will build on the previous section, focusing on a more specific proposal regarding the nature and theoretical motivations for a notion of literal meaning, or, more precisely, literal usage. Section 4 will return to a specific case of the Threat of Collapse, and show how it can be resolved.

60

Referring When Push-Comes-to-Shove

61

An important caveat: In recent years, there has been an explosion of work on issues at the intersection of philosophy of mind and philosophy of language. However, where possible, I am going to side-step potentially distracting details and focus on issues that are germane to the specific topic at hand. I beg the reader’s forgiveness for various cut corners and ignored details. 1.1

Direct Reference

As indicated in the opening line of the chapter, Scott Soames is going to be my chief protagonist. Soames is an arch defender of a Millian or Direct Reference account of the semantics of proper names, demonstratives, and more recently (ms) natural kind terms. Briefly, I’ll consider an expression to be Directly Referential iff its semantic content is exhausted by reference.3 Not surprisingly, Direct Reference finds its roots in the work of J.S. Mill (1858) and, arguably, Russell (1905).4 The considerations that have done the most to pave the way for Direct Reference are a cluster of so-called anti-descriptivist arguments typically credited to Kripke (1980).5 A standard taxonomy divides these arguments into three categories: the semantic, epistemic, and modal arguments.6 Very roughly, the heart of these arguments is a cluster of intuitions to the effect that the semantic, epistemic and modal profiles of (e.g.) proper names fail to pattern after those of descriptions. To give a bit more detail, the intuitions suggest that (i) there are no particular descriptions or clusters of descriptions that a speaker must associate with familiar referring expressions in order to count as a competent user of those expressions; (ii) various statements that a descriptivist semantics is apt to classify as necessary are in fact contingent, and (iii) various statements that a descriptivist semantics is apt to classify as a priori are in fact a posteriori. For example, “was the teacher of Alexander” is a plausible candidate for (part of) the descriptive content of “Aristotle”, but “Aristotle was the teacher of Alexander” is neither necessary nor knowable a priori, and someone can be a perfectly competent user of “Aristotle” while failing to associate this particular piece of descriptive information with the bearer of the name. Call the intuitions underlying these arguments the Kripkean Intuitions. Kripke’s conclusion in Naming and Necessity is that names are rigid designators; that is, that a name refers to the same entity across possible worlds (or, at least, across worlds in which the entity exists). The claim that prima facie referring expressions are rigid falls short of a full commitment to Direct Reference, since rigidity can be squeezed out of the right kind of complex semantic structure. For example, descriptions can be rigidified by an ‘actually’ operator or by (somehow) being forced to take wide-scope over modal operators. Enter so-called Direct Reference theorists, such as Braun (2006, 2005, 2008), Kaplan (1989), Salmon (1986, 1989b, 1989a), Soames (1997 and 2002), and Thau (2002). Soames (2002), in particular, argues at length and

62 Kevan Edwards

in detail that Kripke-style arguments can be extended to support a fullblown Direct Reference thesis. Direct Reference theorists have done much to swing the tide in favour of reference-based approaches to semantics. Nevertheless, it is important to acknowledge that the standard Kripkean arguments, extended or otherwise, point not to positive features of the view, but to the apparent failure of the alternatives.7 Moreover, the alternatives – approaches that dominated the fashion trends in philosophy of language in the interval between (roughly) Mill and Kripke – are not without their own motivations. At the core of these motivations is none other than the accusation that a significant class of intuitions militate against a reference-based semantics. Here, then, is the first hint of a serious tension. 1.2

Neo-Fregeanism

For the purposes of setting up a simplified geography of competing views, call an approach Neo-Fregean if it contrasts with Direct Reference by claiming that something other than reference is part of the semantic content of the relevant class of expressions.8 The most obvious motivations for Neo-Fregeanism stem from the uncontroversially impoverished nature of reference. Famously, while sentences of the form a = b are trivially true, uninformative, and knowable a priori, sentences of the form a = b have the potential to be non-trivial, informative, cognitively significant, and only knowable a posteriori, including cases where a, b are replaced with co-referring names.9 Things get even worse when one considers well known intensional contexts, notably belief ascriptions. Again, in the right context, the intuition can be very powerful that someone believes that Hesperus is a planet while also believing that Phosphorus is not a planet.10 Arguably even worse, uncorrupted intuition suggests that someone can believe that Hesperus is a planet while simultaneously failing to believe that Phosphorus is a planet.11 Advocates of Direct Reference have some explaining to do. To summarize, they need to explain at least the following: (i) direct intuitions about the ‘meaning’ of various prima facie referring expressions; (ii) intuitions about the potential cognitive significance (informativeness, etc.) of expressions of the form a = b (for co-referring instances of a, b); and (iii) intuitive failures of substitution (in particular, when singular terms are found under the scope of a belief operator). Call these Fregean Intuitions. Fregean Intuitions play into the hands of Neo-Fregeans. There is a lot to say – and a lot has been said – about how the Direct Reference theorist can deal with various Fregean Intuitions. One idea that Soames (e.g., 2002 and 2005) explores in some depth is that relevant descriptive information can be asserted by a speaker in uttering a sentence – for example as a result of a process of pragmatic enrichment – without it being part of the semantic content of the sentence. Moreover, it is worth noting

Referring When Push-Comes-to-Shove

63

that even if one is suspicious of Soames’ specific appeals, the general strategy of preserving an austere, Directly Referential semantics by appealing to some or another pragmatic resources has a lot of potential. 1.3

The threat of collapse

So far so good for the Direct Reference theorist. Drawing a line between semantics and pragmatics promises to yield resources for explaining various hard cases in a way that is consistent with a referentialist semantics. Enter the Threat of Collapse. In a nutshell, advocates of (what I am calling) the Threat of Collapse claim that appeals to pragmatically imparted descriptive information are fated to conflict with Kripkean Intuitions. At this point, somewhat different conclusions are drawn by different advocates of the Threat. Braun and Sider (2006) and McKinsey (2005) criticize Soames’ appeals to pragmatically conveyed descriptive information, but they do not claim that this undermines a basic commitment to a Millian semantics. In what follows, I am going to focus on Caplan’s (2007) version of the Threat.12 While Caplan doesn’t put it in exactly these terms, the line of thought he pursues amounts to the following: First, everyone in the Neo-Fregean versus Millian debate needs to account for both Kripkean Intuitions and Fregean Intuitions. Second, each side has a semantic account that is tailor-made to handle one set of these intuitions but faces prima facie problems explaining the other. Thus, nothing in these two patterns of explanatory successes and failures privileges Millian Descriptivism over a (Neo-)Fregean alternative. Caplan says the following: Even if the explanation that Millian Descriptivists offer of speakers’ intuitions about some sentence appeals to the content of that sentence, Fregean Descriptivists can offer a parallel explanation, by saying that speakers communicate the proposition that Millian Descriptivists say is the content of that sentence. As a result, Millian Descriptivism fares no better (or worse) than Fregean Descriptivism. (2007, p. 194, original italics) Caplan’s conclusion is cautious. Given Caplan’s basic line of argument, one might go a step further and conclude that the situation facing the Direct Reference theorist is in fact worse than a stalemate. Arguably, the NeoFregean can claim that their semantics does a better job of according with pre-theoretic intuitions. I myself think the Direct Reference theorist ought to concede this last point. Fregean Intuitions are very easy to generate, whereas Kripkean Intuitions can take more work to probe – it isn’t any surprise, I think, that descriptivism has seemed so attractive, and for so long. Nevertheless, for reasons that will become clear later in this chapter, I also think that this difference in the immediacy of the respective intuitions is to be expected once Direct Reference has been embedded in a larger theoretical framework. Therefore, I do not think these considerations should bear

64

Kevan Edwards

any weight in the final analysis. This, however, is to jump ahead. For now the goal is to clarify the potential force of the Threat of Collapse. The basic worry, to repeat, is that the Neo-Fregean can hope to mimic moves made by the Direct Reference theorist, and they can do so against the backdrop of a semantics that claims better to accord with pre-theoretic intuitions. 1.4 A digression about descriptivism Advocates of the Threat of Collapse, at least as I have described them, attack Millian Descriptivism. But one might wonder if the only way the Direct Reference theorist can hope to appeal to pragmatics is in terms of pragmatically conveyed descriptive information. Speaking very generally, it is natural for the Direct Reference theorist to conceive of pragmatics as a potentially very messy business. And one might expect this messiness to leave open a broad range of options when it comes to information potentially conveyed by pragmatic mechanisms. A related point, which I think is generally under-appreciated, is that it isn’t obvious that the Direct Reference theorist needs to come up with a fully general account of the source of Fregean Intuitions. Arguably, the Direct Reference theorist just needs to be able to invoke a relevant difference in the information conveyed in each particular problem case. Notice, for example, that different people can have different reasons for believing that “Hesperus” and “Phosphorus” fail to co-refer. One person might think this because they tend to apply the names at different times of the day; another because they were told as much by someone they believe to be reliable; yet another because of a general presumption that orthographically distinct names fail to co-refer. In itself, this suggestion about a lack of generality does not shed light on relevant pragmatic resources, but it does open the door for a potentially broad range of resources to be helpful. Soames himself registers doubts about the scope of the appeals to descriptive enrichment that he discusses in Beyond Rigidity. In his response to McKinsey, for example, Soames says the following: One might reasonably ask how often this sort of thing [an appeal to descriptive enrichment] occurs, and how much of a role it plays in understanding Frege’s Puzzle. These are legitimate questions about the scope and significance of the kind of descriptive enrichment I emphasize. My point is minimal; such enrichment does occur, and it plays some role in resolving Frege’s Puzzle. (2005, p. 173) In a related context, Soames gestures towards an appeal to different ‘ways’ of entertaining a proposition (also sometimes described as propositional ‘guises’). This appeal is worth mentioning insofar as it is familiar among advocates of Direct Reference, and it isn’t obvious that ‘ways’ or ‘guises’ need to be understood descriptively. (For related discussion, see Crimmins 1992; Thau 2002.)

Referring When Push-Comes-to-Shove

65

In spite of Soames’ acknowledgement that there might be alternatives to descriptive enrichment, it is striking, and to some extent suspicious, that Soames does little to explore such alternatives. More importantly, it is not obvious that the basic logic of the Threat of Collapse won’t resurface in a different guise (sorry, I can’t resist the pun). As I have described it, the core problem behind the Threat of Collapse is the appearance of a fundamental tension between the austerity of the semantic resources that best account for Kripkean Intuitions and the richness of the resources required to explain Fregean Intuitions. The Threat of Collapse presents an interesting and potentially robust problem for the referentialist. 1.5 The two-source solution Simply put, I think that Kripkean Intuitions and Fregean Intuitions stem from two different sources, and that the appearance of tension between the two sets of intuitions is merely an appearance.13 On the one hand, Kripkean Intuitions stem from the semantic content of the expressions in question; indeed, I’ll sketch an account according to which these intuitions stem from the literal meaning of the expressions in question. On the other hand, I will suggest that Fregean Intuitions have their source in potentially much richer facts about what the speaker of an expression intends to express/communicate in uttering the expressions in question. The basic structure of the two-source solution is obvious enough; what one wants is some reason to think it can be implemented in a way that is both plausible and consistent with Direct Reference. In short, the Direct Reference theorist needs a larger framework that leaves obvious room for a relatively profligate view of pragmatics and yet carves out an important role for an austere, directly referential semantics. I’m going to rough out such a framework in the next two sections. In Section 2, I will help myself to a broadly speaking pragmatist or use-theoretic conception of natural language. Much of this I am going to borrow from recent work by Stephen Neale (ms, 2004a, 2004b). Neale articulates a “collection” of theses that he groups together under the heading of Linguistic Pragmatism.14 In what follows, I will focus on several ideas that are especially relevant for present purposes, and add a few details of my own. I am going to diverge from Neale on one crucial issue. The notion of literal meaning that I am going to introduce is different in emphasis, if not in substance, from Neale’s own appeals to what he calls blueprints. This point of divergence is important enough to warrant its own section (Section 3).

2

Linguistic Pragmatism

The core idea behind Linguistic Pragmatism (henceforth “LP”) is that natural language is best understood as a tool cooperatively exploited by speakers and hearers (mutatis mutandis for writers and readers) for the purposes

66

Kevan Edwards

of expressing/communicating meaningful thoughts. Thus, LP involves a strong claim about the semantic priority of thought over language. A point of central importance to LP is the idea that the primary explanatory notions at work in understanding language are not what a sentence means (or says, or whatever) but what a speaker uses a sentence to say/communicate/express/ etc. I think Neale gets these issues almost exactly right, and I will quote him at length. Saying and implying are things people do. Following ordinary usage, the speaker is taken to be the understood subject, so to speak, of the verbs ‘say’ and ‘imply’, the verbs in talk about ‘what is said’ and ‘what is implied’. (Similarly, with verbs such as ‘communicate’, ‘convey’, and ‘get across’.) We should be initially suspicious of talk about what uses of sentences say (imply, communicate, etc.) and talk about what sentences-relative-tocontexts say (imply, communicate, etc.), unless such talk is taken to be straightforwardly translatable into talk about things that speakers are doing. [ ... ] At the same time, we should be open to the idea that new, technical uses of the verbs ‘say’ (‘imply’, ‘communicate’, etc.) may need to be defined, or at least developed, in the course of our inquiries, such uses earning their keep because of ineliminable theoretical work they do. (2004a, p. 183, original italics) One minor point of possible contention is the extent to which Neale thinks that pre-theoretic intuition (“Following ordinary usage ...”) respects the priority of speaker meaning. I have my doubts about this. For present purposes, however, the issue is moot; even if the primacy of speaker meaning is not pre-theoretically intuitive or countenanced by ‘ordinary usage’, it is well motivated by the extent to which it is a core feature of an imminently plausible account of natural language use. Thus far I have been steering clear of any need to distinguish between what is communicated and what a speaker intends to communicate. As a matter of fact, I assume – and here I am still following Neale – that notions such as ‘what is said’ or ‘what is communicated’ are not primary in the explanatory order. Again, I’ll quote Neale at length: In the first instance, we should separate (i) what A intended to say by uttering X on a given occasion, and (ii) what a rational, reasonably well-informed interpreter in B’s shoes would think A intended to say by uttering X on that occasion (which is not to say there are not problems with the idea of a rational, reasonably well-informed interpreter in B’s shoes). In cases where (i) = (ii), we can talk freely about what the speaker said. In cases where (i) ≠ (ii), certainly we could argue about which of (i) or (ii) or some third thing has the ‘right’ to be called what is said, but what would be the point? First, what third thing distinct from (i) and (ii) could be of any significance

Referring When Push-Comes-to-Shove

67

to a theory of interpretation? There is simply no rôle for a transcendent notion of what is said upon which (i) and (ii) converge when all goes well. It is the coincidence itself of (i) and (ii) that constitutes success, and it is the potential for such coincidence, independently of some third thing, that gives sense to the very idea of saying. (Contrary to linguistic appearances the concept of intending to say is, in fact, more basic than the seemingly simpler concept of saying.) (2004a, p. 182, original italics) To summarize: any notion in the vicinity of ‘what is communicated’ or ‘what is said’ is at best something to be derived from the dual notions of (i) what a speaker intends to say/communicate/etc. and (ii) what a speaker reasonably can believe about their audience’s ability to recover these intentions. Importantly, these notions come together in at least two ways: First, when communication is successful, a speaker’s intentions line up with what their audience can and does recover. Second, speakers are constrained in their choices of utterances by their beliefs about how speakers will understand various sentences they could utter. It is here that linguistic conventions enter the LP picture: speakers and hearers must cooperatively take themselves to be operating with a shared set of conventions about how language can be used.15 The next question to ask – given the present topic of the source of Kripkean and Fregean Intuitions – is how intuitive data fits into the LP framework. Neale (ms) acknowledges the following: As the pragmatist sees things, our intuitive judgments about what A meant, said and implied, and judgments about whether what A said was true or false in specified situations constitute the primary data for a theory of interpretation, the data it is the theory’s business to explain. One needs to be careful in interpreting what Neale is asserting here. Two important points: First, remember Neale’s insistence that the primary explanatory target is not what sentences mean, but what speakers mean in uttering these sentences. Second, in passages like the preceding one, Neale does not intend to rule out notions along the lines of ‘what is said’ or, for that matter, the literal meaning of a sentence. Rather, I take Neale to be committed to the following two claims about the priority of speaker-meaning: (i) Any notion that goes beyond what speakers mean, for example by assigning meaning to sentences (or to their constituents), needs to earn its keep by making a substantial – if not practically ineliminable – explanatory contribution. (ii) Any notion that goes beyond what speakers mean needs to be explained in terms of – or as I prefer to put it, extracted from – the basic use-theoretic framework at the heart of LP. So, intuitive judgments are a (if not the) primary source of data for the Linguistic Pragmatist; however, within the LP framework, there is the

68

Kevan Edwards

potential for (at least) two sources of intuitive judgments. The core data for a theory of interpretation is judgments properly treated as intuitions about what speakers mean (/say/imply/etc.); but the approach also leaves room for more constrained notions of, for example, the literal meaning of a sentence, as long as such a notion can both earn its keep and be derived from within a use-theoretic framework. Before saying more on the topic of literal meaning, I want to discuss several potentially distracting details. Some of these details I mention only for the purpose of setting them aside; however, I also want to introduce a few substantive assumptions that I will take for granted in what follows. 2.1

Assertion, what is said, and literal meaning

Let’s go back, for a moment, to Soames’ notion of descriptive enrichment. Recall that Soames’ appeals to descriptive enrichment, as well as his subsequent responses to McKinsey (2005) and Sider and Braun (2006) are couched in terms of information that is not part of the semantic content of a sentence but nevertheless is asserted by a speaker (in uttering the sentence in a particular context). Soames also makes similar claims about what is said by a speaker in uttering a sentence. These are potentially controversial claims. Indeed, I think it does much to cloud the discussion between Soames and McKinsey that they disagree about whether descriptive enrichment is a fact about what is asserted (mutatis mutandis for ‘what is said’) or if, as McKinsey (2005, p. 166) puts it at one point, “the descriptive assumptions that underlie the use of proper names typically remain in the background, and play at most a reference-fixing role”. This specific debate is not one into which I wish to enter, nor are various other points of contention surrounding different accounts of ‘what is said’ and related notions.16 I am going to sidestep these issues by jumping right to a relatively simple-minded contrast between the overall information that speakers intend to communicate in uttering a sentence and a potentially much more constrained conception of the literal meaning of a sentence. Sketching how a highly constrained conception of literal meaning can be extracted from the basic LP framework is the burden of the next section of the chapter. Just to keep track of the point of all this, recall that the aforementioned distinction is intended to serve as an implementation of the two-source solution. In particular, I want to claim that Kripkean Intuitions plausibly are construed as intuitions about the literal meaning of a sentence, whereas Fregean Intuitions have their source in something potentially much richer. 2.2 Communication/expression Language is used not only for communication but also in cases where there isn’t an intended audience – consider a frustrated parent who blurts out an obscenity. The basic spirit of LP is focused on communication, and it will be easier to take this for granted in what follows (and thereby drop

Referring When Push-Comes-to-Shove

69

the annoying use of “communicate/express”). Simplifying the discussion is one motivation here, but I also think the case can be made that expressive uses of language are parasitic on communicative uses. For example, it isn’t obvious how to make sense of the use of an obscenity as an expression of frustration unless the same expression could be used in a relevantly similar context to communicate the same content.17 2.3 Language and thought; words and concepts A pragmatist or use-theoretic approach to natural language does not entail a pragmatist approach to the semantic content of thought. To the contrary, the basic outline of LP is consistent with an anti-pragmatist conception of the content of psychological states.18 Although it will not play a central role in what follows, I myself am inclined to assume that thoughts have a roughly language-like structure, at least to the extent that whole proposition-sized thoughts break down into robust (repeatable and, for the most part, context-independent) constituents, or concepts. Of course, this is to commit to the heart of the so-called Language of Thought hypothesis (Fodor 1981, 1975, 2008). I am also inclined to accept, at least as a working hypothesis, that words have semantic properties in virtue of their bearing a relation to corresponding concepts – more along these lines in a moment. Finally, I am inclined to follow the suggestion outlined previously to the effect that conceptual content should receive an independent, non-pragmatist treatment.19 I should add that while Neale flirts with these kinds of assumptions in various places, I can’t with a clear conscience burden him with a firm commitment to either the Language of Thought or the plausibility of a nonpragmatist approach to conceptual content.

3 Literal meaning as what is communicated when push-comes-to-shove Here is where I take us to be: According to LP, the primary explanatory notions underwriting claims about the semantics and pragmatics of natural language have to do with facts about how speakers use sentences to communicate thoughts. Of particular relevance is the nature of various conventions constraining the thoughts that a sentence can be used to communicate. In particular, then, anything in the vicinity of a claim about the literal meaning of a sentence needs to bottom out in such facts. I’m going to assume – and I don’t think this much is particularly tendentious – that once such facts are spelled out, they can be treated as defining a relation that holds between each sentence (type) and the thought (or set of thoughts) that grounds claims about the literal meaning of the sentence. For the sake of introducing a relatively neutral term, I will call this relation the backing relation. As I indicated, I am also inclined to accept, at least as a working assumption, that claims about the literal meanings of words are

70 Kevan Edwards

underwritten by the fact that they bear the backing relation to corresponding concepts.20 One detail: I assume it is a mistake to describe the backing relation as holding between a word/sentence and its meaning. The backing relation is more aptly described as the relation whereby natural language expressions inherit their meanings from their corresponding concepts.21 Thinking about the notion of literal meaning in terms of a relation is very natural within the LP framework; nevertheless, one might prefer to run the discussion in terms of some kind of entity that either defines or is defined by such a relation. This is more or less what I take Neale to be doing in his discussion of what he calls blueprints. I should note, however, that insofar as I want to offer an alternative to Neale’s blueprints, the relation-vs.-entity distinction is not what I take to be at issue. Here is how the notion of a blueprint arises for Neale: Qua description of semantic competence, a semantic theory for a language will explain how the syntactic structure of a sentence (or sentence fragment) X and the meanings of the individual words in X conspire to constrain what speakers can say using X. Flushing out the modal: a semantic theory for a language L will provide, for each sentence X of L, a blueprint for (a template, a schematic or skeletal representation of) what someone will be taken to be saying when using X to say something. The blueprint associated with X is its semantics, and the set of such blueprints, one for every sentence of a language L, is the semantics for L. (2004a, p. 189) It is important to note that Neale is careful about the theoretical burden that blueprints can and should bear in an overall account of linguistic interpretation. It is part and parcel of Neale’s approach that blueprints leave a lot of slack to be picked up by other theoretical resources. I think Neale is right to limit the role of anything in the vicinity of a notion of literal meaning; however, I’m worried that he doesn’t go far enough. In particular, there seems to be little to prevent blueprints themselves, at least as Neale construes them, from being infected by the obvious messiness of language use. To explain what I mean here, I need to take a step back and make some claims about plausible constraints on any workable notion of literal meaning. I take it, and I take it that Neale would agree with this, that a substantial part of the explanatory bona fides of a notion of literal meaning is that it promises to ground linguistic communication in something relatively stable across agents and contexts. The need for a stable core is especially pressing in the context of a commitment to the basic LP framework insofar as it is part and parcel of LP that language use can be very plastic, contextdependent, creative – in short: messy. The tool metaphor is instructive here: Once a tool is available, it can be used in no end of creative ways, uses that can far outstrip its original raison d’être.

Referring When Push-Comes-to-Shove

71

In spite of the obvious messiness of language use, there is substantial pressure in the opposite direction; there are some relatively uncontroversial ways in which language is a strikingly robust phenomenon. The most obvious point here is just that linguistic communication works, and that it works reliably and non-arbitrarily. To be sure, temporal, geographical and cultural discontinuities present obstacles to communicative success; but, one way or another, speakers/hearers know how to overcome these obstacles. To introduce a slogan of which I am going to make heavy use in what follows: when push comes to shove, speakers and hearers know how to retreat to a common ground, and know that doing so will help to ensure communicative success. Retreating to a common ground presupposes that there is a common ground, and that speakers and hearers recognize it (in some sense) as such. 22 Moreover, to repeat a now familiar point, the need for a common ground becomes even more pressing against the presumptively messy background of a use-theoretic conception of language. I submit that underwriting the robust nature of communication should be at the heart of any notion of literal meaning worth wanting. Given this assumption, my worry about Neale’s blueprints is that there is little reason to think they will be sufficiently robust. One might respond by pointing out that Neale’s invocation of blueprints is merely a starting point, one that needs to be filled in by, among other things, substantial empirical research.23 Moreover, Neale’s motivation for introducing the notion of a blueprint is not so much to articulate a positive account of literal meaning as to make the negative point that whatever literal meaning is, it is apt to vastly underdetermine (i) what a speaker can intend to say/communicate and (ii) what a speaker reasonably can expect their audience to understand them as intending to say/communicate. These caveats noted, I want to introduce what I take to be an alternative way of thinking about how a notion of literal meaning might impose constraints on speakers. Rather than each actual usage being within a range of acceptable uses (as the notion of a blueprint seems to suggest) or being an elaboration or extension of acceptable use (as various discussions of ‘what is said’ seem to suggest), I want to suggest that the literal meaning of a sentence imposes constraints on speakers indirectly, by determining how they would use a sentence were they to be in a kind of ideal context (what I will call – in spite of obvious worries about circularity – a literal context). The basic idea builds on the aforementioned suggestion that language users share a common ground to which they retreat when push comes to shove; in particular, they retreat in the direction of a shared expectation that sentences will be used literally. Let me add some flesh to this picture. I assume that in most conversational contexts, speakers start out with a wide range of information that they would be happy to communicate, some pieces of information more

72 Kevan Edwards

important than others. Speakers select from among various sentences that they could utter based on their estimates of what speakers would be able to extract from these sentences (as uttered in the particular context in question). To clarify, I am not claiming that speakers need to have an explicit conception of exactly what information they intend to communicate. Nor do speakers need to have an explicit conception of exactly what information needs to be communicated for the interaction to count as a success. Nor do they need to have a direct or infallible way to figure out what or how much information has been communicated. What speakers do have, according to the picture I want to advocate, is a potentially complex set of dispositions to react in certain ways given different kinds of feedback from their audience. Most important for present purposes is the suggestion that when a speaker perceives a threat to communicative success, they are disposed to retreat to uses of language that are increasingly literal. The reason they retreat in this direction is obvious given the view that I am advocating: literal meaning is a matter of conventions that are robust, inter alia that are shared across users of a language. In short, while it might not be efficient, literal meaning is relatively safe. Without committing to details that outstrip the grain of the present analysis, more can be said about what it means to retreat to increasingly literal uses of language. In some cases, merely restating an initial sentence can stave off the threat of communicative failure. Consider a situation in which Ann and Ben are sitting at opposite ends of a table. The table is very long and thus Ann has a legitimate reason to wonder if Ben is able to pass various condiments located at his end of the table. Consider the following dialogue, the heart of which is a restatement of an initial sentence: Ann: Can you pass the salt? [Ben picks up the salt and pauses, evidently reflecting on whether he can get the salt down the table without getting up from his chair] Ann: No, don’t bother; I was really just asking if you can pass the salt. In other cases, a mutually assumed shift towards an increasingly literal context will render a previous sentence misleading, but the speaker can get away with refining it in various ways. Cases of refinement include the familiar notion of a pragmatic implication being cancelled: Casey: Did you hear that Jackson and Jillian had children? David: Yes, but I’ve heard two names for the baby; did they have twins? Casey: No, Jackson and Jillian had children, but not together. In yet other cases, the threat of communication breakdown is so strong, or the shift to an increasingly literal context does so much to undermine the

Referring When Push-Comes-to-Shove

73

felicity of the previous utterance, that the speaker needs to retract and replace their previous sentences, as in the dialogue that follows: Evan: I went to have a look at the aircraft carrier last night and, wow, that thing is a mile long! Frida: Really? that seems unlikely. Evan: No, not really, but it is a lot longer than I expected. A clarification: the fact that speakers restate, refine, and retract and replace does not entail that any particular use of a sentence is purely literal. What matters to the approach is that speakers/hearers are disposed to retreat in the direction of literal usage, not that they ever are pushed all the way to a context where only strictly literal usage is felicitous. Note that a retreat to increasingly literal usage is driven by practical considerations, but that practical considerations cut in two directions: On the one hand, there is pressure to retreat to literal usage when communication threatens to break down. On the other hand, setting up a literal context – a context where hearers assume that the speaker is speaking literally, and speakers can rightly assume that their hearer so assumes, and so on – can be laborious and inefficient. Thus there is little reason to think that typical cases where an utterance is restated, refined, or retracted and replaced are anything other than steps in the direction of literal usage. The potential unattainability of a purely literal context is a virtue of the account. It helps to make sense of how difficult it is to isolate exactly what a speaker has in mind in uttering a particular sentence, even in highly rarefied contexts where a speaker is under substantial pressure to ensure that communication is successful. This is an example of what I take to be the most obvious, overarching virtue of the push-comes-to-shove approach to literal meaning, namely that it promises to capture the best of two prima facie competing worlds. On the one hand, a relatively austere notion of literal meaning promises theoretical tractability and robustness across agents and contexts.24 On the other hand, the fact that actual conversational contexts are rarely, if ever, strictly literal leaves room for the obviously messy features of language use (context-dependence, plasticity, creativity, etc.) Exactly what distinguishes contexts of increasingly literal use? I can’t hope to offer a fully worked out account here (nor do I have one); but I can offer two broad suggestions regarding how the use of an expression can fail to be literal. First, a speaker can fail to use an expression literally for secondary practical reasons. Speaking very generally, in a context where speaker and hearer share a sufficient number of background beliefs, a speaker can assume (and assume that her audience will assume) that a sentence used non-literally will succeed in communicating more information than a different sentence used (relatively) literally. Of course, this is little more than a standard gesture to the need to invoke pragmatic mechanisms (e.g., Soames’

74 Kevan Edwards

appeals to descriptive enrichment). Nevertheless, the underlying point is that various practical considerations, while contributing to the efficiency with which language can be used to communicate information, are secondary in that they tend to be overridden in contexts where communicative success is under threat. A second reason speakers can fail to use an expression literally is (again very roughly) because they are suffering from relevant epistemic deficiencies. Things get tricky here insofar as attempts to distinguish issues properly belonging to semantics from those better classified as having to do with the epistemic situation of speakers/hearers is bound to induce controversy. The basic idea, at any rate, is that various epistemic deficiencies can lead a speaker to use an expression in a way that outstrips or even violates its literal meaning. The possibility of epistemic constraints on what counts as literal usage introduces the need to shift to a theorist’s perspective. What I mean by this is that epistemic deficiencies can prevent a speaker (mutatis mutandis their audience) from recognizing that a use of an expression is, in a relevant respect, non-literal. At risk of courting controversy, let me offer an example. The example has the benefit of gesturing in the direction of how the pushcomes-to-shove approach plays into the hands of Direct Reference theorists. The following is an example of a typical problem case for Direct Reference. Gary is aptly described as failing to realize that Cilantro is Coriander (more details about Gary’s epistemic situation in a moment). In a suitable context involving a conversation with his friend Hattie, Gary utters (1): (1) Cilantro is not Coriander. According to Direct Reference, (1) is semantically equivalent to (2). (2) Cilantro is not Cilantro. There is little reason to think that raising the practical stakes will always be a way to generate contexts that expose the presumptive synonymy of expressions like “Cilantro” and “Coriander”. Perhaps no amount of pragmatic pressure would give Gary a reason to let go of his commitment to the truth of (1); whereas, if queried, he would refuse to assent to (2). Nevertheless, my suggestion is that the Direct Reference theorist can explain Gary’s inability to retreat to an increasingly literal context in terms of his suffering from various epistemic deficiencies. For example, Gary doesn’t realize that “Cilantro” and “Coriander” are names for the same plant, and he doesn’t realize that the plant on his window sill produces the same kind of seeds that are in his cupboard in the jar marked “Coriander”, and so on. These, among others, are true statements about what Gary (and perhaps also Hattie) doesn’t know.25 Importantly, there does not need to be any particular epistemic deficiency to which the Direct Reference theorist appeals in all situations in which a speaker utters (1).26 What matters is that the Direct Reference theorist can gesture towards a relevant epistemic deficiency in each particular case.

Referring When Push-Comes-to-Shove

75

Before going any further, I want to clarify that I am not tying myself to the claim that what a speaker communicates when (practical and epistemic) push-comes-to-shove defines or provides an analysis of literal meaning. Defending this claim would require a much richer story about purely literal contexts than I have to offer. Nevertheless, I hope I have made it plausible (1) that the notion of how speakers (would) react when practical and epistemic push-comes-to-shove is one with intuitive force and explanatory potential, (2) that it is relatively easy to see how to elaborate on this basic picture in potentially productive ways (for example by filling in details about the pragmatic mechanisms that speakers/hearers are under pressure to abandon when push-comes-to-shove).27 I now want to add some flesh to the suggestion that the Direct Reference theorist can hope to exploit the push-comes-to-shove framework. And I want to do this in a way that brings Kripkean considerations about an expression’s modal and epistemic profile back into the discussion. Kripkean considerations enter the picture with the suggestion that being under pressure to respect an expression’s epistemic and modal profile is partly constitutive of the notion of literal usage. As an illustration, consider a discussion between Ian and Jackie about how mathematicians tend to do their best work early in life. Given the right conversational background, one can imagine Ian using a proper name to communicate descriptive information, which I will indicate in square parentheses:28 Ian: Wiles [the discoverer of the proof of Fermat’s last theorem] did great work when he was over forty. Now imagine the conversation continuing as follows: Jackie: Wiles is an exception; what if Wiles had not proved Fermat’s last theorem? Ian: Wiles would still have done great work when he was past forty, even if he didn’t prove Fermat’s last theorem. Here is what the aspiring Direct Reference theorist can say about this conversation. First, there is an obvious shift in context between Ian’s two utterances. The shift is such that Ian can no longer use “Wiles” to express the relevant descriptive information. Ian has been pushed to consider counterfactual circumstances in which Wiles did not complete the proof. Because of this, he is under pressure to make sure that his use of “Wiles” does not presuppose otherwise. Second, the Direct Reference theorist can claim to have a principled reason for treating this shift as a step in the direction of increasingly literal usage. Recall that preserving something robust (sharable, context independent, etc.) at the core of language use is a central motivation for wanting to extract a notion of literal meaning from the LP framework. Robustness across counterfactual circumstances is an important form of robustness. Thus, it is natural for the Direct Reference theorist to claim that

76 Kevan Edwards

increased sensitivity to features of an expression’s modal profile is part and parcel of a shift towards increasingly literal usage. The same points can be made about pressure on speakers to manifest sensitivity to an expression’s epistemic profile (or, more precisely, to the epistemic profile of sentences of which the expression is a constituent). As an illustration, imagine a conversation between two historically (and philosophically) ignorant students, who are discussing whether anyone in ancient Greece would have been smart enough to pass the exam they have just suffered through: Karl: Plato [the ancient Greek philosopher] was really smart; he would have aced it. Laura: I’m sure Plato was really smart, but you don’t know that he was an ancient Greek; you just heard that from your roommate, who is a terrible source of information. Let’s assume that Laura’s utterance is enough to make Karl worried about the source of his assumption. He can no longer use “Plato” as a means of communicating the relevant descriptive information. Again, it is natural for the Direct Reference theorist to claim that this shift in context is a shift towards an increasingly literal context. And, again, this is consistent with the general phenomenon of increasingly literal contexts involving uses of language that are increasingly robust, this time across differences in an agent’s epistemic situation. A quick recapitulation is in order. I have suggested a way to extract a notion of literal meaning from the basic LP framework. For convenience, I have been assuming that expressions in natural language are backed by corresponding concepts. The backing relation, I have suggested, is best approached by thinking about how speakers are disposed to respond when pragmatic and epistemic push-comes-to-shove. Kripkean considerations entered the picture with the suggestion that increased sensitivity to an expression’s modal and epistemic profile can be understood as instances of the general phenomenon that speakers retreat to increasingly robust uses of language when push-comes-to-shove. The obvious suggestion is that the aspiring Direct Reference theorist can hope to claim that the notion of literal meaning that emerges from this picture is one that vindicates Kripkean Intuitions. More specifically, they can hope to claim that the notion of literal meaning that emerges is Directly Referential.

4 Being pragmatic about the threat of collapse I can now be explicit about how all of this bears on the two-source solution. Working within the basic framework of LP, it is relatively easy to account for Fregean Intuitions. At the core of LP is the idea that speakers and hearers exploit their linguistic resources in potentially open-ended

Referring When Push-Comes-to-Shove

77

ways. Given this framework, rich and fine-grained intuitions about ‘meaning’, cognitive significance, and so on, are surely to be expected. The substantive question facing the aspiring Direct Reference theorist is whether something can be extracted from the LP framework that is plausibly the source of the Kripkean Intuitions. My claim in the previous section was that the push-comes-to-shove approach supplies a notion of literal usage that has this feature. Actually, I think the question of whether the relevant notion of literal usage can produce the Kripkean Intuitions is dialectically misguided. The Direct Reference theorist who buys into my suggestion that Direct Reference ought to be embedded in an LP framework does not need to treat standard Kripkean Intuitions as anything other than a starting point. The important question is whether the referentialist-qua-pragmatist can point to sufficiently powerful considerations – intuitive or theory-driven – to motivate a referentialist commitment within the framework of LP. My suggestion has been that taking seriously the push-comes-to-shove approach promises to do this. Of course, insofar as a Directly Referential notion of literal meaning can be extracted from the LP framework, it isn’t surprising to find something aptly described as versions or analogues of the original Kripkean Intuitions; for example, considerations about how speakers react to various kinds of conversational pressure.29 Let’s look at a specific example, in particular a case discussed by Caplan (2007). I will follow Caplan in taking for granted the following as background: First, in the relevant context, “Cary” and “Archie” are names for the same individual, “Cary Grant” being the screen name of Archibald Alexander Leach. Second, I’ll assume that the descriptive information, the actor who played C.K. Dexter in The Philadelphia Story, is a plausible candidate for what the Millian Descriptivist might claim is conveyed by “Cary” but not by “Archie”. Caplan’s worry, then, centres on a sentence such as (3). 30 (3) Cary is an actor. The problems start with the fact that if “Cary” conveys the descriptive information the actor who played C.K. Dexter in The Philadelphia Story, then (3) should be equivalent to (4). (4) The actor who played C.K. Dexter in The Philadelphia Story is an actor. (4) is both necessary and knowable a priori; therefore, the Millian Descriptivist ought to be committed to the claim that (3) is necessary and knowable a priori. But this runs contrary to Kripkean Intuitions. Hence the conclusion that I have credited to Caplan: the Millian Descriptivist’s motivations for a Direct Referential semantics are fated to collapse given their account of the relevant Fregean Intuitions.

78

Kevan Edwards

Here is what the resources that I have outlined allow the Direct Reference theorist to say about this case: On the one hand, Fregean Intuitions are to be expected given the background framework of LP. Utterances of sentences are mutually understood by speakers and hearers to communicate a potentially wide range of information. Insofar as there is any intuitive pull towards including the actor who played C.K. Dexter in The Philadelphia Story in the information conveyed by a particular utterance of “Cary”, the referentialist-qua-pragmatist will have an easy time accommodating this. But now let’s be more careful about the kind of context that might generate such an intuition. For example, imagine that Mary and Ned have just watched The Philadelphia Story. Imagine further that Mary is enamoured with the leading star of the film and has been subjecting Ned to one after another of his films over the past few weeks. In an underhanded attempt to strike back, Ned has tracked down an article detailing Archie Leach’s marital exploits; indeed, he has conveniently left the article open on a page detailing accusations to the effect that Archie Leach spanked at least one of his five wives. Unfortunately for Ned, the article is written without mentioning the name “Cary Grant”. When Mary reads the article, she comments that “Archie Leach must have been a chauvinist pig”, but she never connects the person described in the article with the star of the films she has been watching. Given these details of the case, how can the referentialist-qua-pragmatist hold onto the claim that even though Ned uses (3) to communicate the relevant descriptive information, this information is not part of the expression’s semantics? Here is how: First, it is clear that the conversational context in question is not a purely literal context, since Mary is suffering from an obvious cluster of epistemic deficiencies. For example, Mary doesn’t realize that “Cary” and “Archie” (as used in this context) refer to the same person, she doesn’t realize that the actor in the movie she just watched is none other than the person about whom the article was written, and so on. In this particular case, Ned himself is in a better epistemic position, and he can take steps to shift the conversational context such that it is increasingly literal. For example, he can utter (5): (5) You know, the person described in the article just is the actor you are in love with. If you don’t believe me, look it up online. The resulting context is more literal than the previous context, and hence a better insight into what constitutes a literal usage of the names in question. And note that in the resulting context, the relevant descriptive information can no longer be used to draw a wedge between “Cary” and “Archie”, and hence cannot be part of the content of one but not the other. So, even if a use of “Cary” in a relevant utterance of (3) conveys the descriptive information “the actor who played ...”, the referentialist-qua-pragmatist should not countenance it as part of the name’s literal meaning. Again, the general

Referring When Push-Comes-to-Shove 79

moral is that even if typical uses of an expression communicate descriptive information, there are reasons to cleave to the claim that this information is not part of the expression’s literal meaning. Thereby, I claim, the relevant Threat of Collapse can be disarmed. What would Caplan say about this? While Caplan does not use the term “two source solution”, it is clear that he has considered the basic strategy. Here is what he says: It is implausible that speakers’ intuitions about modal profile have a different source than their intuitions about truth-value. After all, the modal profile of a sentence includes, not only whether it’s necessary or contingent, but also which worlds it’s true in and hence whether it’s actually true; and it’s implausible that speakers’ intuitions about whether a sentence is actually true should have a different source than their intuitions about whether it’s true simpliciter. Similarly, it’s implausible that speakers’ intuitions about epistemic profile have a different source than their intuitions about cognitive value. After all, the cognitive value of a sentence includes whether it’s a priori or a posteriori, as does its epistemic profile. (2007, p. 189) I want to underline two points in response, neither of them new. I’ll focus on what Caplan says about the relationship between intuitions about modal profile and intuitions about truth-value. First, as clarified in the previous section, the Direct Reference theorist qua pragmatist does not need to hold themselves hostage to the project of rescuing original Kripkean Intuitions about the cases in question; they are free to appeal to whatever considerations motivate their commitment to a referentialist semantics. I have suggested that within the general framework of LP, there is good reason to embrace the existence of what amounts to a referentialist core of literal usage. Second, and more importantly, I disagree with Caplan’s claim about the implausibility of intuitions about modal / epistemic profile having a different source than intuitions about truth value / cognitive value. For starters, standard intuitions about truth value, informativeness, and more generally about what an expression ‘means’ typically are largely pre-theoretical, whereas generating robust Kripkean Intuitions often involves some unpacking, if not the clarification of relevant theoretical notions (necessity, a prioricity, etc.). Moreover, there are some important differences between the kinds of questions that are useful in prompting each category of intuition. For instance, when asked about the ‘meaning’ of a belief ascription, it is easy to interpret the question relatively loosely, as a question about what the sentence typically is used to say about the believer. Without some reason to be careful about drawing a distinction between what is semantically expressed versus pragmatically conveyed by such an ascription, or between what the sentence itself means versus what attributers mean in uttering the sentence, it is no wonder that

80

Kevan Edwards

intuitive responses tend to be informationally rich. Contrast this with questions intended to probe intuitions about the modal and epistemic profile of a sentence. In answering these questions, an informant is forced to look for features that are shared across modal and epistemic circumstances. This has the potential to constrain rather than broaden the scope of potentially relevant information. As such, it is natural for the informant to settle on something much narrower than what guides intuitions about informativeness, cognitive significance, and even truth value. Pace Caplan, it is Kripkean Intuitions that are more restricted in the relevant respects.

5

Conclusion

The discussion in this chapter suggests conclusions at several levels of abstraction. Beginning with the most concrete, I hope to have made it plausible that if the Direct Reference theorist is willing to embed their core semantic commitments within the larger framework of a pragmatist conception of natural language, they can hope to lay claim to more than sufficient resources to avoid the Threat of Collapse. To be fair, I haven’t argued for Linguistic Pragmatism so much as take it for granted; nevertheless, the view is certainly not implausible. Moreover I think there are various signs that Linguistic Pragmatism and Direct Reference are a powerful combination. I also suggested what I take to be an alternative way of approaching the notion of literal meaning or, more precisely, literal usage. This is, no doubt, the most controversial territory explored in the chapter. Nevertheless, the push-comes-toshove approach has notable virtues, in particular the extent to which it draws on facts about how speakers are disposed to react in various conversational settings. This, in turn, focuses attention on the extent to which the referentialistqua-pragmatist can lay claim to what I have described as the best of both worlds. The Direct Reference theorist can hope to embrace a full-blooded, anythinggoes-as-long-as-it-works conception of language use while holding onto an austere conception of literal meaning. This overall package promises to explain the manifest plasticity, context-dependence, and creativity of language use without this posing a threat to the robust and non-arbitrary nature of linguistic communication. In a nutshell, the approach that I have outlined goes right to the heart of how language can be so disastrously messy and yet work so well.

Acknowledgements It should be obvious that I owe an intellectual debt to Stephen Neale and Jerry Fodor. I would also like to thank Ben Caplan for helping me to think through some of the ideas in this chapter years before I managed to get them down on paper. Finally, I would like to thank colleagues at Syracuse University for conversation and helpful feedback, in particular Kris McDaniel, Ben Bradley, Melissa Frankel, and Kara Richardson.

Referring When Push-Comes-to-Shove

81

Notes 1. Here and throughout, I use the term ‘thought’ to pick out a generic class (or classes) of psychological states. Hence my use contrasts with that of, e.g., Frege, who uses the term to pick out abstract platonic entities. 2. It is worth mentioning a point that will become clear in what follows, namely the extent to which the ‘pragmatism’ that I will advocate is carefully delimited. Not only will the approach be invoked in the service of a robust reference-based approach to semantics, but it will be consistent with a strong anti-pragmatism about the nature of the mental. 3. I am overlooking details here that have the potential to make a difference. E.g., one might distinguish between an expression’s semantic content and its semantic contribution to larger expressions of which it is a proper part (see Crimmins, 1992). Worse still, once one distinguishes between an expression’s semantic content and its semantic contribution, room opens up for accepting (e.g.) context-dependence vis-à-vis the latter while denying it vis-à-vis the former. I myself doubt there is much to be gained by splitting this particular hair, but my thinking so turns on issues – e.g., the pointlessness of preserving the letter of Direct Reference by giving up on a strict notion of compositionality – that I will not argue for here. At any rate, Caplan (2007), who is going to be my main antagonist, does not distinguish semantic content from semantic contribution; indeed, he treats the semantic content of a proper name to be its contribution to propositions expressed by sentences of which it is a constituent. 4. Russell is more obviously a Millian when viewed in abstraction from (i) his particular views about how much underlying structure is masked by the surface grammar of natural language, (ii) various epistemological commitments that serve as restrictions on the class of genuine referring expressions (e.g., the commitment to a principle of acquaintance). 5. Also see (at least!) Donnellan (1997) and Kaplan (1989a and 1989b). Davidson (2007) is a recent and very helpful collection of seminal contributions to the Direct Reference tradition. 6. See, e.g., Caplan (2007), Soames (2002), and Salmon (2005). 7. Sider and Braun make this point. They say, “The data of Kripke et al. is genuinely puzzling. It in no way undermines the old Fregean arguments against Millianism; it simply is new, conflicting data.” (2006, p. 669). 8. The suffix ‘Neo’ indicates that Neo-Fregeans need not embrace Frege’s anti-psychologism. Unlike Frege, a Neo-Fregean can, and I think should, claim that ‘modes of presentation’ – or whatever else are called on to do some or all of the work for which Frege posited senses – are most plausibly understood in terms of some or another psychological entities. 9. The famous source of this ‘puzzle’ is Frege (1952). A few details: First, following a convention employed by Soames in various places, I use bold-face font in place of corner quotes. Second, I should be explicit that I am merely gesturing towards a standard presentation of Frege’s Puzzle, and glossing over various subtleties. One subtlety has to do with the potential non-existence of the referent of (substitution instances of) a and b. A deeper issue turns on the relevant notion of ‘form’ employed in the statement of the puzzle. In the right context, someone can wonder whether “Fred = Fred” (e.g., when they are worried that the two occurrences of “Fred” are names of two different people), or even whether “that is that” (e.g., when pointing to the same thing twice without realizing it). It isn’t obvious what notion of

82 Kevan Edwards

10.

11.

12.

13.

14.

15.

form can divide these cases from similar cases that are uncontroversially trivial. One might respond that slipperiness in the relevant notion of ‘form’ only makes things more difficult for the referentialist. But this isn’t right. Insofar as the Nng responses. Even among those broadly sympathetic to Millianism there is substantive disagreement; for example, McKay (1981) argues for a straightforwardly Millian eo-Fregean draws on instances of Frege’s Puzzle to criticize the alternatives, they need to ensure that their own view does not fall prey to similar problems. Insofar as the alternatives to Millianism do face similar problems, this invites the conclusion that a semantic solution to Frege’s Puzzle was never on the cards and that (Neo-)Fregeans were wrong to wield the problem against Millians. I take it this is the moral that Kripke (1979) draws from the famous case of Puzzling Pierre. (I should note, however, that Kripke’s conclusions have provoked varyiresolution to the puzzle, whereas Loar (1987) suggests that the puzzle motivates an ‘impure’ version of Millianism.) I assume a standard story where “Hesperus” and “Phosphorus” are two names for the same heavenly body, namely Venus, even though competent speakers (and belief attributers) may not realize they co-refer. Sometimes I will use italics to highlight that-clauses, in particular when they are used in belief attributions. One reason for doing this is that it serves as a reminder that belief reports being reported by the theorist occupy a potentially uneasy middle ground between pre- and post-theoretic usage. For example, a committed Millian might offer a pragmatic explanation for why speakers utter various sentences that semantically express contradictory propositions. If so, the Millian is as free as anyone to utter such sentences in standard conversational contexts, but they might be reluctant to do so in contexts where attention is focussed on questions about literal semantic content. “Arguably even worse” because according to the Direct Reference theorist the last intuition involves not just a report of inconsistent beliefs but a contradiction in the mouth of the belief reporter. See Kripke (1979). Why am I focusing on Caplan? First, I find Soames’ own response to Sider and Braun and McKinsey close to definitive and don’t want to repeat what has been said elsewhere. Second, and more importantly, Caplan’s discussion is much more in the spirit of unearthing a fundamental tension in reference-based approaches, and I think unravelling the strands contributing to this tension is both important and illuminating. Note that explaining away the appearance of conflict is not, on its own, enough to dissolve the force of the Threat of Collapse. The Direct Reference theorist needs not only to distinguish two sources for the intuitions in question, but to say something to privilege the source of the Kripkean Intuitions as deserving the title of ‘semantics’. Although I won’t explicitly focus on this second issue in what follows, an appropriate asymmetry between the two sources of intuitions will be an obvious feature of the diagnosis that I will offer. Citing Neale is convenient insofar as he brings together various theses under a common heading. However, Neale himself (2004a, p. 178) is quick to cite a long list of people who are “operating in a broadly pragmatist spirit”, including (but not limited to): Austin, Bezuidenhout, Carston, Chomsky, Crimmins, Evans, Heal, Fodor, Grice, Quine, Perry, Récanati, Searle, Sellars, Sperber and Wilson, and (Peter) Strawson. There is little reason to think that a complete account of linguistic interpretation will be anything other than extremely complex, if not theoretically intractable.

Referring When Push-Comes-to-Shove

16. 17. 18.

19.

20.

21. 22.

23.

24. 25.

83

Neale describes at least two pragmatists, Chomsky and Fodor, as deep pessimists who “have argued that asking for a theory of interpretation is tantamount to asking for a ‘theory of everything’, a complete cognitive psychology, because virtually anything can impinge upon the holistic process of interpretation.” (2004a, p. 178 n. 25). See, as a starting point, Bach (2001), Carston (2002), Recanati (2001). For all I know, expression might be best understood as a kind of communication with oneself. One indication that LP is consistent with a non-pragmatist conception of thought is the fact that Neale describes Fodor as an ally when it comes to LP. Fodor is perhaps the arch anti-pragmatist when it comes to mental content: In his recent LOT-2, Fodor writes, “In my view, pragmatism is perhaps the worst idea philosophy ever had” (2008, p. 9). Fodor’s target here is pragmatist conceptions of the nature of intentional states; in particular, the intentional properties of these states. Admittedly, questions about how this story might go are fraught and I cannot hope to do justice to them here. For what it is worth, here is a quick summary of my own views on the subject: (1) In the context of a presumptive pragmatist reduction of linguistic content to mental content, questions about the potential naturalization of mental content become all the more pressing. (2) The basic geography of naturalistically-respectable accounts of mental content comes down to (i) one or another version of Conceptual Role Semantics (e.g., those articulated by Block 1986; Harman 1982; Loar 1981; Peacocke 1992) and (ii) one or another theory of representation (e.g., those articulated by Dretske 1981; Fodor 1992; Millikan 1993). (3) CRS faces a list of deep problems, including various difficulties catalogued by Fodor (1998, 2004) and Fodor and Lepore (2002), as well as analogues of the Kripkean Intuitions (Edwards, ms). One reason to think this might be an oversimplification is the potential for sentences in natural language to be backed by thoughts containing constituents that do not correspond to anything ‘articulated’ in the surface structure of the sentence. See note 20 for more about conceptual content. Another way in which language is robust, one that is also in prima facie tension with the manifest flexibility, context-dependence, and creativity of language use, is the extent to which language is both productive and systematic. The natural explanation for these properties – perhaps the only plausible explanation – is that language and/or thought is both syntactically and semantically compositional, and arguably in a particularly strict way (Fodor and Lepore 2002). There are suggestions in Neale (ms) that any workable notion of ‘what is said’ will need to be carefully circumscribed. The point, as I understand it, is that ‘what is said’ needs to be able to underwrite notions of commitment and contract across a potentially broad range of contexts (conversational or otherwise), and that the relevant constraints on ‘what is said’ bottom out in constraints imposed by the relevant blueprints. Austerity also makes it easier for a semantic theory to live up to a compositionality requirement (see note 22). Depending on how details of the view are worked out, it can be important to the Direct Reference theorist that these statements about Gary’s relevant epistemic deficiencies are describable in terms that do not presuppose distinctions that go beyond those sanctioned by a commitment to Direct Reference.

84 Kevan Edwards 26. See the related discussion in Section 1.4. 27. There are interesting questions about what kinds of considerations motivate shifts toward increasingly non-literal usage. For example, one can imagine conversations where a metaphor is discharged in terms of a more literal explanation but then speakers/hearers shift back to using the metaphor because it is more efficient. More interesting is the possibility of situations in which speakers are happy to be interpreted as intending to communicate a range of propositions but efficiency demands that they not be held hostage to spelling out each and every one. Perhaps typical utterances of ‘I’m ready’ [to go to the party / to leave the house / to go out on the town / ?] might be like this (to pick a favourite example of Stephen Neale’s). 28. After writing this chapter, I came across the fact that Kripke (1977) draws on similar conversational considerations as partial support for a ‘unified’ Russellian account of descriptions. The two appeals are nicely complementary. Conversational considerations suggest that both descriptive uses of referring expressions and so-called referential uses of descriptions can be explained in terms of pragmatic features of language use, rather than in terms of the literal meaning of the expressions in question. (The referential versus attributive distinction is due to Donnellan 1970). 29. There is at least one way in which the referentialist-qua-pragmatist needs to reconfigure standard intuitions about the semantic, modal, or epistemic profile of prima facie referring expressions: According to the pragmatist, talk of what sentences (inter alia their constituent expressions) mean is at best short-hand for talk of conventional constraints on what speakers can hope to communicate in uttering a sentence. 30. I’ve simplified Caplan’s sentences by glossing over a point about which Caplan is rightly careful; namely, the possibility of counterfactual circumstances in which Cary/Archie fails to exist, and hence where (3) is trivially true.

Bibliography Bach, K. (2001) ‘You Don’t Say?’, Synthese, 128(1/2), 15–44. Block, N. (1986) ‘Advertisement for a Semantics of Psychology’, in P.A. French, T.E.J. Uehling, and H.K. Wettstein (eds) Midwest Studies in Philosophy (Minneapolis: University of Minnesota Press). Braun, D. (2005) ‘Empty Names, Fictional Names, Mythical Names’, Nous, 39(4), 596–631. —— (2006) ‘Illogical but Rational’, Nous, 40(2), 376–79. —— (2008) ‘Complex Demonstratives and Their Singular Contents’, Linguistics and Philosophy, 31(1), 57–99. Braun, D. and Sider, T. (2006) ‘Kripke’s Revenge’, Philosophical Studies, 128, 669–82. Caplan, B. (2007) ‘Millian Descriptivism’, Philosophical Studies, 133(2), 181–98. Carston, R. (2002) ‘Linguistic Meaning, Communicated Meaning and Cognitive Pragmatics’, Mind & Language, 17(1–2), 127–48. Crimmins, M. (1992) Talk about Beliefs (Cambridge, Mass.: MIT Press). Davidson, M. (2007) On Sense and Direct Reference: Readings in the Philosophy of Language (Boston: McGraw-Hill). Donnellan, K. (1970) ‘Proper Names and Identifying Descriptions’, Synthese, 21, 335–58. Dretske, F.I. (1981) Knowledge and the Flow of Information (Oxford: Basil Blackwell).

Referring When Push-Comes-to-Shove

85

Edwards, K. (ms) ‘Keeping Reference in Mind’, Paper delivered at the Pacific meeting of the American Philosophical Association, April 2009. Fodor, J.A. (1975) The Language of Thought (Cambridge, Mass.: Harvard University Press). —— (1981) Representations: Philosophical Essays on the Foundations of Cognitive Science, 1st MIT Press edn (Cambridge, Mass.: MIT Press). —— (1992) A Theory of Content and Other Essays (Cambridge, Mass.: MIT Press). —— (1998) Concepts: Where Cognitive Science Went Wrong (Oxford: Oxford University Press). —— (2004) ‘Having Concepts: A Brief Refutation of the Twentieth Century’, Mind & Language, 19(1), 29–47. —— (2008) LOT 2: The Language of Thought Revisited (New York: Oxford University Press). Fodor, J.A. and Lepore, E. (2002) The Compositionality Papers (Oxford: Oxford University Press). Frege, G. (1952) ‘On Sense and Reference’, Translations from the Philosophical Writings of Gottlob Frege (Oxford: Basic Blackwell). Harman, G. (1982) ‘Conceptual Role Semantics’, Notre Dame Journal of Formal Logic, 23(2), 242–56. Kaplan, D. (1989) ‘Demonstratives’, in J. Almog, J. Perry, and H. Wettstein (eds) Themes from Kaplan (New York: Oxford University Press). Kripke, S. (1977) ‘Speaker’s Reference and Semantic Reference’, Midwest Studies in Philosophy, 2, 255–77. —— (1979) ‘A Puzzle about Belief’, in A. Margalit (ed.) Meaning and Use (Dordrecht: D. Reidel). —— (1980) Naming and Necessity (Cambridge, Mass.: Harvard University Press). Loar, B. (1981) Mind and Meaning (Cambridge; New York: Cambridge University Press). —— (1987) ‘Names in Thought’, Philosophical Studies, 51, 169–85. McKay, T. (1981) ‘On Proper Names in Belief Ascriptions’, Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition, 39(3), 287–303. McKinsey, M. (2005) ‘Critical Notice of Scott Soames, Beyond Rigidity’, Canadian Journal of Philosophy, 35(1), 149–68. Mill, J.S. (1858) A System of Logic, Ratiocinative and Inductive, 5th edn (New York: Harper & Brothers). Millikan, R.G. (1993) White Queen Psychology and Other Essays for Alice (Cambridge, Mass.: MIT Press). Neale, S. (2004a) ‘Pragmatism and Binding’, in Z. Szabo (ed.) Semantics and Pragmatics (Oxford: Oxford University Press). —— (2004b) ‘This, That, and the Other’, in A. Bezuidenhout (ed.) Descriptions and Beyond (New York: Oxford University Press). —— (ms) ‘Linguistic Pragmatism’, Unpublished manuscript. Peacocke, C. (1992) A Study of Concepts (Cambridge, Mass.: MIT Press). Recanati, F. (2001) ‘What Is Said’, Synthese, 128(1–2), 75–91. Russell, B. (1905) ‘On Denoting’, Mind, 14(56), 479–93. Salmon, N. (1986) Frege’s Puzzle (Cambridge, Mass.: MIT Press). —— (1989a) ‘How to Become a Millian Heir’, Nous (23), 211–20. —— (1989b) ‘Illogical Belief’, Philosophical Perspectives (Atascadero: Ridgeview). —— (2005) Reference and Essence, 2nd edn (Amherst, N.Y.: Prometheus Books).

86

Kevan Edwards

Soames, S. (1997) ‘Direct Reference, Propositional Attitudes, and Semantic Content’, in P. Ludlow (ed.) Readings in the Philosophy of Language (Cambridge, Mass.: MIT Press). —— (2002) Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (New York: Oxford University Press). —— (2005) ‘ ”Beyond Rigidity”: Reply to McKinsey’, Canadian Journal of Philosophy, 35(1), 169–78. —— (ms) ‘What Are Natural Kinds?’, Forthcoming in Philosophical Topics. Thau, M. (2002) Consciousness and Cognition (Oxford; New York: Oxford University Press).

5 Semantic Normativity in Context Anandi Hattiangadi

1

Introduction

There are two parallel, contemporary debates in the philosophy of language about the relationship between literal, linguistic meaning, and use. The ‘contextualism’ debate concerns the extent to which the literal meanings of sentences in a language determine the truth conditions or propositional contents of utterances, and the extent to which extra-linguistic context determines utterance content. The normativity debate concerns whether the literal meaning of a sentence in a language has normative implications for a speaker’s utterance of the sentence. Both debates concern the adequacy of the sort of formal semantic theory pioneered by Frege and the early Wittgenstein, and later championed by Davidson and Kaplan, among others. Both debates turn on challenges to the formal semantic tradition inspired by the later writings of Wittgenstein. The contextualist argues that formal semantic theories cannot adequately determine the truth conditions, or propositional contents, of utterances. The normativist holds that formal semantic theories must be normative – they must have implications for whether an utterance of a sentence is correct or incorrect. Semantic normativity is famously employed by Saul Kripke, to argue that there are no semantic facts to correspond to the statements of any given semantic theory. Crucially, Kripke’s sceptic argues that there is no account of what the semantic facts consist in which can adequately ground the normativity of meaning (Kripke 1982; Hattiangadi 2002, 2007). It would be surprising if these debates, and the issues involved, had no bearing on one another. Yet, to my knowledge, the relations between them have not been explored. In this chapter, I will argue that the contextualism debate has a bearing on the normativity debate. Many of the examples used in contextualist arguments can be used to give powerful reasons to reject the view that meaning is normative. I will also tentatively suggest that the semantic normativity debate has a bearing on contextualism.

87

88 Anandi Hattiangadi

2

Formal semantic theories and the contextualist challenge

There is a compelling picture of meaning and understanding, according to which a semantic theory assigns meanings to the words of a language, and specifies functions from the meanings of words, and their mode of combination, to the meanings (truth conditions) of sentences. For example, a semantic theory of English would contain the axiom that the word ‘snow’ means snow – it is satisfied by x if and only if x is snow. Similarly, the theory would say that ‘white’ is satisfied by something if and only if it is white, and given these meanings and logical form, the sentence ‘snow is white’ means that snow is white; it is true if and only if snow is white.1 In general, such a theory will contain lexical axioms of the form: For all x: e is satisfied by x in L if and only if Fx Here, e is an expression of the language, such as ‘snow’, L is the language in question, such as English and F expresses a property in virtue of which e is satisfied by x, such as the property of being snow. I use ‘is satisfied by’ here as a general term covering different word-world relations, such as, for example, reference. The previous lexical axioms, together with compositional functions, generate assignments of meanings for sentences, such as that a sentence s means that p, where p is the proposition expressed by the sentence (i.e., ‘snow is white’ means that snow is white). A truth conditional semantic theory will specify the meanings of sentences by specifying the truth conditions of the sentences. That is, it will generate theorems of the form: s is true in L if and only if p. Where p is a proposition expressed by the sentence, s in the language, L. Semantic theories in this tradition have a number of distinguishing characteristics. First they are formal, which means that they begin with a syntactic description of sentences and assign them to equivalence classes on the basis of syntax. Second, strictly speaking, they assign a meaning to expression-types and sentence-types of the language, and derivatively, assign meanings and determine truth conditions of token sentences and utterances. Third, these theories are compositional. That is, the meaning of a sentence-type is a function of the words of which it is composed and their mode of combination. The meaning of a sentence type is thus invariant across contexts of utterance. The meaning assigned by a semantic theory to a sentence-type in the language is commonly called ‘literal meaning’. What implications does a formal semantic theory have for the content of an utterance of a given sentence? On the face of it, a formal semantic theory will simply assign a propositional content (truth condition) to the sentence,

Semantic Normativity in Context

89

on the basis of the meanings of its parts and their mode of combination, and the content of the utterance will just be equivalent to the content of the sentence. Matters get more complicated, however, when a sentence contains an indexical expression, such as ‘I’. Following Kaplan (1989), the formal theorist can distinguish between the ‘character’ of the indexical and its ‘content’. The character is the context-invariant, literal meaning of the indexical. It is a rule which tells you what, in the context of utterance, is the semantic value of ‘I’. So, if I say ‘I am a woman’, the character of ‘I’ determines that Anandi is the referent of the indexical, and the content of the sentence is that Anandi is a woman. The content of the sentence will be different, of course, if you utter the sentence, but the character will remain constant. This is compatible with the formality of semantic theories because, although the context plays a role in determining the content of the token sentence uttered, the influence of context is syntactically triggered and semantically determined. A formal semantic theory thus assigns a literal meaning to a sentence, which determines the content of an utterance of that sentence, together with syntactically mandated input from the extra-linguistic context. Utterance content is just the content of the sentence uttered with indexicals evaluated relative to the context of utterance. So, if you utter the sentence ‘snow is white’, the content of your utterance is that snow is white, and it is true if and only if snow is white. When we communicate using language, however, we do much more than utter sentences. To use a well-worn example, if you ask me what I think of a candidate for an academic appointment in philosophy, and all I say is ‘he has nice handwriting’, in that context, I communicate or ‘implicate’ (Grice 1989) that he is not a very good philosopher. The propositions implicated by a sentence uttered in a given context are not determined, or even obviously constrained, by either the syntax or the semantics of the sentence uttered, and hence, fall outside the purview of semantics. This gives us distinctions between: 1) the literal meaning of a sentence, which is a function of the meanings of its parts and their mode of combination, and which remains invariant across contexts of utterance; 2) the content of a token sentence which, according to a formal semantic theory, is determined by the literal, linguistic meaning of the sentence evaluated relative to a context of utterance; 3) the content of an utterance, which, according to a formal semantic theory, just is the content of the sentence uttered; and 4) the ‘speaker meaning’ of an utterance; what the speaker intends to convey by the utterance of the sentence, in context. The contextualist challenge to formal semantic theories concerns the adequacy of formal semantic theories in specifying 3 – the contents of utterances of sentences, or ‘what is said’ by an utterance of a sentence in context. According to contextualists, there are at least some sentences, s, such that even if you specify the meaning of every word in s, specify all of the

90 Anandi Hattiangadi

compositional meaning rules relevant to s, disambiguate every ambiguous expression, precisify every vague expression, and fix the referents for every indexical expression in s, you would still fail to specify the propositional content, or truth conditions, of an utterance of s.2 The contextualist debate concerns how we should explain intuitions about what is said by an utterance of a sentence s, when those intuitions deviate from the utterance content determined by our best semantic theories, and when those intuitions seem to vary in relation to utterance context. The literature on contextualism in semantics has reached mammoth proportions. There is no way to do justice to the literature, and the subtle variations between positions, in this short essay. So, I’m afraid I will have to make do with a drastic over-simplification. Broadly speaking, there are three camps. Contextualists hold that the semantic contents of at least some utterances are context-sensitive – we arrive at the contents of these utterances by a pragmatic process of ‘saturation’ or ‘free-enrichment’, which is not triggered by the overt syntax of the sentence, and which is not determined by the semantics of the sentence (e.g., Bach 1994, 2005; Carston 2002; Recanati 2004; Sperber and Wilson 1986; Perry 1993; Searle 1978; Stanley 2000; Stanley and Szabó 2000; Travis 1985). Some contextualists postulate surprise indexicals, hidden variables, or unarticulated constituents, in order to systematize the role played by context in determining the truth conditions of utterances (e.g., Perry 1993; Stanley 2000; Stanley and Szabó 2000). Not all contextualists hold that all utterances are context sensitive, and not all contextualists agree about how contextual features are factored in, but they all agree that at least in some cases, the literal meaning of a sentence, even when evaluated relative to a context of utterance, fails to determine the propositional content, or truth conditions, of an utterance. Charles Travis captures the spirit of contextualism when he says, ‘what words [literally] mean plays a role in determining when they would be true; but not an exhaustive one. For that role leaves room for variation in truth conditions, with meanings fixed, from one speaking of words to another’ (Travis 1996, p. 451). In response to the contextualist challenge, minimalists argue that a formal semantic theory is adequate to specify the truth conditions, or semantic content, of utterances (Borg 2004; Cappelen and Lepore 2005). They claim that what a formal semantic theory delivers is a ‘minimal’ propositional content of an utterance. On this view, it is just not the role of a semantic theory to explain everything there is to be explained about communication (Borg 2004). On this view, the rich propositional utterance contents assigned by contextualists are not part of the semantic contents of utterances, and hence are outside the purview of semantics. So-called semantic intuitions that are sensitive to changes in the context of utterance show only that what is communicated (or implicated) by an utterance can be sensitive to utterance context but what is said by the utterance of a given sentence, its ‘semantic content’ (once resolved for ambiguity, indexicality, vagueness, etc.) remains invariant across different contexts of utterance.

Semantic Normativity in Context

3

91

The normativity debate

The normativity debate also concerns the relationship between literal, linguistic (context-invariant) meaning and use. The locus classicus for the view is Kripke’s influential discussion of Wittgenstein’s rule following considerations (Kripke 1982), where he famously claimed that ‘the relation of meaning and intention to future action is normative, not descriptive’. On Kripke’s view, Wittgenstein’s rule following considerations pose a fundamental challenge to the sort of formal, semantic theory described before – they show that if we assume such a semantic theory, there can be no such thing as meaning anything by any word (Kripke 1982, p. 55). The normativity of meaning plays a crucial role in Kripke’s sceptical argument. Assuming the normativity of meaning, Kripke’s sceptic purports to show that there can be no adequate, non-circular, account of what the semantic facts consist in (Hattiangadi 2007). Since the publication of Kripke’s book, a large number of philosophers have endorsed the view that meaning is normative (Boghossian 1989, 2003; Brandom 1994; Blackburn 1984; Gibbard 2003; Glock 1996; McDowell 1993; McGinn 1984; Millar 1998, 2002, 2004; Soames 1989; Whiting 2008; Wilson 1994; Wright 1984; Zalabardo 1989). What, exactly, does semantic normativity amount to? Paul Boghossian expresses one version of the view clearly in the following passage: Suppose the expression ‘green’ means green. It follows immediately that the expression ‘green’ applies correctly only to these things (the green ones) and not to those (the non-greens). The fact that the expression means something implies, that is, a whole set of normative truths about my behaviour with that expression: namely, that my use of it is correct in application to certain objects and not in application to others. (Boghossian 1989, p.513) Here, the normative implication – that in English, ‘green’ applies correctly to the green things – is said to follow from the fact that ‘green’ means green. That is, if a formal semantic theory assigns the meaning green to the word ‘green’ – if it says that x satisfies ‘green’ if and only if it is green – it thereby assigns a normative status to utterances of the word ‘green’. It implies, in particular, that uses of the word ‘green’ in application to green things are correct, while uses of the word ‘green’ in application to non-green things are incorrect. What is it to apply a word? Obviously, you don’t go around shouting out words in isolation from sentences, and even if you did, it would not be entirely clear whether an utterance of an isolated word could be regarded as correct or incorrect. What would make it correct or incorrect for you to shout out ‘green’? Does there need to be something green present, for it to be correct? Do you need to demonstrate something green? Would it be

92

Anandi Hattiangadi

enough to think of something green? I don’t know how to answer these questions. So, I think it is best to just avoid them. To make the normativist’s view maximally plausible, we can restrict the uses for which meaning has normative implications to utterances of well-formed sentences. However, it is still not clear whether an utterance of a question counts as an application of an expression – to ask whether unicorns are green does not seem to be to ‘apply’ green. So, it is most plausible to interpret the normativist as holding that meaning has normative implications at least for utterances of (wellformed) declarative sentences. Under what conditions is the utterance of a declarative sentence correct, or incorrect? Suppose I say ‘snow is green’. Is this utterance correct? Surely, it is not, because ‘green’ is applied to something non-green. However, if this is what makes it incorrect, then the word ‘snow’ must have been applied correctly – to snow – otherwise, the application of ‘green’ could have been either correct or indeterminate. So, the only clear cases in which an utterance is correct in every respect are those in which all of the words in the sentence are used correctly, i.e., when the sentence is true. That is, the normativist maintains that for any utterance u and sentence s: N: u of s is correct if and only if s is true. This is a ‘criterion of correctness’ for utterances. It is not intended as an analytic statement: ‘correct’ or ‘semantically correct’ does not mean the same as ‘true’, as normativists often take pains to point out. One argument for this is that ‘correct’ is a normative term, whereas ‘true’ is not (Wedgwood 2007). Nor is this a statement of necessary property identity – so, it is not the claim that the property of being semantically correct is identical to the property of truth. Rather, the claim is that truth is a ‘semantically correct making property’ of utterances. If you utter a true sentence, this makes the utterance correct; utter a false sentence, and this makes the utterance incorrect. The semantic normativist’s view is that the semantic correctness conditions of an utterance of a sentence are the conditions under which the sentence uttered is true, given its literal, linguistic meaning. Anti-normativists, including myself, have argued that the normativity thesis is untenable (Hattiangadi 2002, 2007, 2008; Glüer 1999a, 1999b, 2001; Glüer and Pagin 1999; Wikforss 2001). Many of these arguments have centred around the idea that what a word means implies something about what a speaker ought to say or may say. If the normativist holds that you ought to utter all the true sentences of your language, he violates the principle that ought implies can – for you can’t utter all the true sentences of your language. Some of them are just too long. If the normativist holds merely that you ought not to utter false sentences, he commits to the counter-intuitive view that lying, irony, and humour are in some sense, semantically forbidden. In this chapter, I will set aside the question whether the

Semantic Normativity in Context

93

semantic normativity thesis can be formulated in terms of the paradigmatically normative concepts of permission and obligation, and merely grant that ‘correct’ is a primitive normative notion. However, even if it is granted that ‘correct’ is a primitive normative notion, we must be careful to distinguish between what might be called instrumental and non-instrumental correctness. This parallels the distinction between instrumental and non-instrumental oughts. An instrumental ought is a conditional ought in which the antecedent and consequent of the conditional are related as end to means. For example, ‘if you intend to increase the temperature in the room, you ought to turn up the heat’, is an instrumental ought. Turning up the heat is a means to satisfying your intention, namely, increasing the temperature in the room. In contrast, a noninstrumental ought is one that lacks this feature. For example, ‘you ought to give to charity’ is non-instrumental (so long as it is not assumed that giving to charity is a means to satisfying a desire or intention). Similarly, one might distinguish between instrumental and non-instrumental correctness. So, one might say, ‘if you intend to increase the temperature in the room, the correct thing to do is to turn up the heat’. In this case, the correctness is instrumental, because the correct thing to do is only correct insofar as it constitutes a means to satisfying an intention. This distinction is crucial to the discussion of semantic normativity, because if semantic correctness is merely instrumental, then meaning is not intrinsically normative in any interesting sense. To see why, we need to first pay attention to an ambiguity in instrumental ought statements – the ought-operator can take wide or narrow scope (Cf. Broome 1999). That is, in general terms, ‘if you want to Φ, then you ought to Ψ’ can be interpreted in either of the following ways: Narrow scope: If you intend to Φ, then you ought to (Ψ) Wide scope: You ought to (if you intend to Φ, then you Ψ) If we take the narrow scope interpretation of instrumental oughts, such statements cannot be understood as genuinely normative, on pain of ‘bootstrapping’. That is, if such statements are regarded as normative, then you can acquire obligations simply by forming desires or intentions (ends), no matter how absurd or immoral these desires or intentions are, and no matter how absurd or immoral the acts are that satisfy the intentions or desires. So, if instrumental oughts are understood to have narrow scope, and if they are taken to be genuinely normative, then when Bush formed the intention to invade Iraq, he thereby acquired an obligation to do so. However, creating obligations just by forming intentions is like pulling yourself up by your own bootstraps, or lifting yourself off the ground by pulling up your socks – it can’t be done. You cannot make it obligatory or correct to do something, just by forming an intention. Thus, if

94 Anandi Hattiangadi

instrumental oughts are given the narrow scope interpretation, they must be understood as descriptive statements – statements that merely describe the (necessary or sufficient) means to satisfy certain ends. Similarly, if statements about semantic correctness are merely instrumental and the scope of ‘correct’ is narrow, then these statements will also have to be understood as descriptive statements. For, just as you can’t create an obligation by forming an intention, you can’t make it correct for you to do something just by forming an intention. On the other hand, if instrumental oughts are given the wide scope interpretation, they can be understood to be normative, because the bootstrapping problem can be avoided. On this interpretation, there are two ways to satisfy the requirement: either you can do what is needed to satisfy your ends, or you can give up your ends. So, Bush did not create an obligation just by intending to invade Iraq, since he could have satisfied the wide scope requirement either by invading, or by giving up the intention – and it is plausibly the latter that he ought to have done. So, wide scope instrumental oughts can be understood to be genuinely normative. Nevertheless, the general, wide scope requirement that one ought to take the necessary means to satisfy one’s intentions is a rule of practical rationality, not semantics. Facts about meaning, facts about heating systems, facts about the weather, can all be relevant to what you ought to do if you are to satisfy your intentions, and such facts can be relevant to what constitutes satisfying requirements of practical rationality. But this does not imply that these facts are normative. Similarly, even if there are true statements specifying what it is correct for one to do in order to satisfy one’s (semantic, or communicative) intentions, and even if ‘correct’ takes wide scope over the conditional, we will still only have special cases of general principles of practical rationality. Hence, if semantic correctness is merely instrumental, meaning is not normative in any interesting sense. In the following I will employ some of the examples that are often brought forth by contextualists of one kind or another, and argue that these examples give us powerful reasons to reject the thesis that meaning is normative.

4

Context-shifting

Context-shifting arguments rely on cases in which an utterance of a sentence, resolved for indexicality, ambiguity, vagueness and so forth, still seems to express different propositions in different contexts. Here is an example of one sentence that gives rise to intuitions about context sensitivity: (1) Every bottle is empty. (Stanley and Szabó 2000, pp. 219–20) A standard context-shifting argument with regard to (1) goes as follows: the content of an utterance of (1) cannot be determined by the meaning of (1),

Semantic Normativity in Context

95

because the contents of utterances of (1) change from context to context, even when the meaning of (1) remains fixed. Since (1) contains no overt indexical expressions, such as ‘I’, ‘there’ and so forth, there is no overt syntactic marker of context shift. Yet, what is said by (1) does shift from one context to another. Therefore, the argument goes, what is said by (1) cannot be determined by its literal, linguistic meaning and those features of the context whose relevance to determining content is mandated by overt, syntactic elements of the sentence. For example, suppose that you ask me whether there is anything more to drink at the end of a long party, and I utter (1). In that context, my utterance of (1) seems to express the proposition, (1a) Every bottle (of alcoholic drink in the house) is empty. That is, we would regard an utterance of (1) as saying (1a) and therefore, we would regard (1) as being true, even if there were full bottles of bleach in the house, and even if there were full bottles of alcoholic drink elsewhere in the universe. Now, suppose that I’ve just come home with some bottles for you to fill with your home made beer, and you ask me whether the bottles need emptying. I utter (1) (Stanley and Szabó 2000, pp. 231–32). In this context, it seems as though my utterance of (1) expresses the proposition, (1b) Every bottle (that I bought) is empty. In this case, again, we would regard (1) as true, even if it turned out that there were lots of full bottles of alcoholic drink in the house, and elsewhere in the universe, because we would take an utterance of (1) in this context, as expressing (1b). There is nothing in the overt, syntactic structure of (1) that corresponds to the parts of the expressed propositions given in parentheses in the example. In (1), the quantifier expression, ‘every’ is un-restricted, yet it seems to be restricted, in different ways, in different contexts of utterance. If it is true that an utterance of (1) expresses (1a) in one context, and (1b) in another context, then the proposition expressed by utterances of (1) cannot be determined by the linguistic meaning of (1) alone (resolved for ambiguity, etc.), for that would determine a context-invariant truth condition for (1), and therefore, a context-invariant content for both utterances of (1). Hence, the contextualist argument goes, the context of utterance must be relevant to determining the ‘semantic content’ or what is said by an utterance. This example can be converted to show that the normativity of meaning must be mistaken. N was the view that an utterance of a sentence is correct, if and only if the sentence uttered is true. Now, let s be (1). Given just the

96

Anandi Hattiangadi

meanings of the words in (1) and their mode of combination, the sentence (1) expresses the proposition, (1c) Every bottle is empty. Since (1) contains no overt indexicality, the context-invariant content of the sentence will be (1c). However, if the content of any utterance of (1) is (1c), then it will be false if there are any non-empty bottles in the universe. And it is plausible to assume that on most of the ordinary occasions in which one might utter (1), there is at least one non-empty bottle in the universe. Assume that in both of the scenarios described, there is at least one non-empty bottle in the universe, and hence, that (1) is false in both contexts. N then implies that utterances of (1) in both contexts are semantically incorrect. However, intuitively, it is correct to utter (1) in either of the above-mentioned contexts. If I utter (1) in either context, I succeed in communicating a true proposition: (1a), in the first context, and (1b), in the second. If there is nothing semantically wrong with uttering a false sentence when one succeeds in communicating a true proposition, N must be false. The correctness of an utterance of a sentence does not depend on whether the sentence uttered is true. This seems to follow regardless of whether one holds, with the minimalist, that the content of any utterance of (1) is (1c), or whether one holds, with the contextualist, that the contents of utterances of (1) vary in relation to the context of utterance. For, the normativist holds that the correctness of an utterance depends on the truth of the sentence uttered, not on the truth of the content of the utterance. And the content of the sentence uttered is (1c), regardless of whether one endorses the minimalist or contextualist view of the content of the utterance. The only view that might help the normativist in this case is one that postulates a covert indexical in the logical form of (1), which would make the context-sensitive contents, (1a) and (1c) a function just of the literal meaning of the expression, evaluated relative to the contexts of utterance. This suggests that the move the normativist ought to make is to say this: N*: u of s is correct if and only if p is true. Where p is the proposition expressed by the utterance, in context.3 Assuming some form of contextualism, the content of an utterance of (1) in the first scenario is (1a), so it comes out as true, and therefore, semantically correct. The content of an utterance of (1) in the second scenario is (1b), so it too comes out as true, and therefore semantically correct. This coheres with our intuition that there is nothing semantically wrong with uttering (1) in either of the contexts described. On this view, the correctness of an utterance depends on the truth of the content of the utterance, and the content of the utterance depends, in part,

Semantic Normativity in Context

97

on the meaning of the sentence uttered, and in part, on features of the extralinguistic context of the utterance. So, ultimately, the correctness of an utterance of a sentence, s, depends on both semantic features of s, and on pragmatic features of the context of the utterance of s. This explication of N* should make it clear that it no longer amounts to the thesis that meaning is normative. The meaning of a sentence uttered is at most relevant to the correctness of an utterance of the sentence. This is particularly clear if part of what determines the content of an utterance of a sentence in context is the communicative intention of the speaker of the sentence. And this seems to be crucial to determining truth content on most forms of contextualism.4 For instance, suppose once again, that I utter (1), when you ask me if there is anything left to drink. Suppose that some form of contextualism is true, and the content of my utterance is (1a), in that context. If what makes it the case that (1a) is the content of my utterance has something to do with my communicative intention, such as my intention to communicate the thought that every bottle of alcoholic drink in the house is empty, then the correctness of uttering the sentence (1) in that context is merely instrumental. That is, if I intend to communicate the thought that every bottle of alcoholic drink in the house is empty, and given that an utterance of (1) is a means to communicating that thought, then it is correct for me to utter (1) in that context. The literal meaning of (1) is clearly relevant to whether uttering (1) is a means to communicating the thought I intend to communicate. Yet, this does not imply that meaning is normative. Similarly, if I intend to go out in the rain without getting soaked, it is true that one means to satisfy my intention is to wear my raincoat. The fact that my raincoat is waterproof is relevant to the fact that wearing it is a means to satisfying my intention to go out in the rain without getting soaked. But this clearly does not make facts about waterproofing normative. Thus, the normativity of meaning cannot amount to N*, together with some contextualist account of what determines the content of an utterance of a sentence in context. Moreover, if minimalism is assumed, then N* offers no advantage over N, since according to minimalism, the content of an utterance of a sentence is just the content of the sentence itself (evaluated relative to the context of utterance). Assuming minimalism, then, N* collapses into N.

5 Incompleteness Incompleteness arguments purport to show that, in some cases, the literal meaning of a sentence, even when evaluated relative to the context of utterance, does not determine a ‘complete proposition’ as the content of the utterance. For example, consider the following: (2) John is ready. (3) Steel isn’t strong enough. (4) It is raining.

98 Anandi Hattiangadi

The standard argument from incompleteness goes as follows. A sentence like (3) does not specify what John is ready for – is he ready for his exam? Is he ready to go out in the rain? Is he ready for supper? If I utter (2) in the context of a discussion of John’s preparedness for an exam, the content of my utterance seems to be: (2a) John is ready (to take the exam). If, in contrast, I utter (2) in the context in which we are all getting dressed to go out in the rain, the content of my utterance seems to be: (2b) John is ready (to go out). Hence, the contextualist argues, (2) displays sensitivity to context, in much the same way as (1) does. However, in these cases, the contextualist goes further and argues that the part of the contents of utterances of (2a) and (2b) that remain invariant across contexts – i.e., the parts of the propositions in (2a) and (2b) that occur outside of the parentheses – are incomplete. They are mere ‘propositional fragments’ (Bach), or ‘skeletons’ (Recanati) which need ‘enrichment’ or ‘saturation’ by features of the context of utterance to determine complete propositions. There is no such thing as just being ready, simpliciter. If (2) just says that John is ready simpliciter, we have no way of evaluating the truth of the utterance. Since there is nothing in the overt syntax of (2) which indicates what should go in the parentheses, the influence of context on the contents of utterances of (2) cannot be traced back to syntactic or semantic features of (2). For instance, here is Kent Bach on (3): [(3)], though syntactically well formed, [is] semantically or conceptually incomplete, in the sense that something must be added to the sentence for it to express a complete and determinate proposition. With [(3)] we need to know strong enough for what (it does not express the weak proposition that steel is strong enough for something or other) ... (Bach 1994, p. 269) The point here is that, unless features of the context of utterance play a role in completing the content of the sentence (given its linguistic meaning, and evaluated relative to the context of the utterance), the sentence expresses something less than a complete, truth evaluable, proposition, and the utterance, without completion, would express something less than a complete and determinate, truth evaluable proposition. The incompleteness argument can also be converted to provide good reasons to deny the normativity of meaning. According to the normativist, an utterance of a sentence, s, is correct if and only if s is true. If the contextualist’s argument is cogent, it seems to show that a sentence such as (4) has no truth conditions, no propositional content, when considered in isolation

Semantic Normativity in Context

99

from a context of utterance. If this is correct, and assuming N, it follows that it is indeterminate whether any utterance of (4) is correct or incorrect. However, this flies in the face of our intuitions. If you and I are standing in the midst of a downpour, and I utter (4), it hardly seems to be indeterminate whether my utterance is correct or incorrect – surely, it is correct in that context. Similarly, if I see a mirage, and undergo an hallucination, in the middle of the Sahara desert, and utter (4), it hardly seems indeterminate whether my utterance is correct or incorrect – surely, it is incorrect, in that context. Hence, N must be false, and the correctness of an utterance of a sentence does not depend on whether the sentence uttered is true. Of course, this difficulty will not arise if minimalism is true. If minimalism is true, then utterances of sentences such as (4) do express complete propositions, even without contextual enrichment. If minimalism is true, then incompleteness would no longer pose a problem for normativists. However, even if minimalism is true, sentences such as (2), (3), and (4) will still give rise to shifting intuitions about correctness. Suppose I utter (4) in the Sahara desert, and suppose that at the time of utterance, it is raining in Tokyo. If minimalism is true and (4) is true if and only if it is raining (at the time of utterance),5 then an my utterance of (4) in the desert is true in the situation described. Together with N, this implies that my utterance of (4) is semantically correct, in this situation – which is counterintuitive. An utterance of (4) in the absence of local rain seems to be incorrect, even if it is raining in a distant place. Once again, it might be tempting for the normativist to resort to N* – to claim that the correctness of the utterance depends on the truth of the content of the utterance, which in turn may well depend on contextual enrichment. However, the foregoing objections to N* will apply in this case as well. If what determines the propositional content of an utterance of a sentence is, in part, the communicative intention of the speaker, then the correctness of uttering the sentence, in that context, will be merely instrumental. If, in contrast, what determines the propositional content of an utterance of a sentence is the propositional content of the sentence itself, then N* collapses into N, and the foregoing objections apply.

6 Inappropriateness Arguments from inappropriateness concern utterances of sentences which do seem to express complete propositions, but where the propositional content of the sentence seems intuitively inappropriate to capturing our intuitions about the content of the utterance. For example, consider the following: (5) I’ve eaten. (6) The ham sandwich wants to pay.

100 Anandi Hattiangadi

First, take (5). Suppose you ask me if I’d like to stay for supper and I reply by uttering (5). Given just the literal meaning of (5), it seems that it is true if and only if I have eaten, and it would be true even if I hadn’t eaten since yesterday. Yet, intuitively, the argument goes, when I utter (5), I express the proposition that I’ve eaten recently, not just that I’ve eaten (at some time or other). In these cases, contextualists argue that the literal meaning of the sentence may determine the content of the sentence, but does not determine the intuitive content of the utterance. Now, take (6). The contextualist can rightly point out that it is somewhat difficult even to understand (6) without imagining a context of utterance. However, it is easy to understand (6) if we imagine it uttered by a harried waitress to her colleague in a busy café, where the patrons are frequently referred to by what they have ordered. In this context, the utterance of (6) seems to express the proposition that the man who ordered the ham sandwich wants his bill. The contextualist argues that there is nothing in the morphology of the sentence itself which dictates this interpretation, so contextual enrichment supplies what is needed to determine a sensible or intuitive truth conditional content for the utterance. Once again, the normativity of meaning flies in the face of our intuitions about the correctness and incorrectness of utterances of sentences such as (5) and (6). N states that an utterance of a sentence is correct if and only if the sentence uttered is true. Suppose now that you ask me to stay for supper, and I utter (5), even though I haven’t eaten since yesterday. N implies that my utterance is correct, since it is true that I have eaten, even though I haven’t eaten for a very long time. Yet, this is counterintuitive. It seems to be incorrect for me to utter (5) in this context, because to do so is to mislead you into thinking I had eaten recently. Hence, N, once again, seems to be false. The troubles with N are even more pronounced in the case of (6). Ham sandwiches don’t have desires, and don’t have money with which to pay for anything. So, (6) is almost certainly false. N implies that it would be incorrect of the harried waitress to utter (6). Yet, intuitively, there seems to be nothing wrong with the waitress’ utterance of (6) – for, by uttering (6), the waitress can communicate a truth to her colleague, namely, that the man who ordered the ham sandwich wants to pay. Once again, N seems to fly in the face of intuitions about the correctness and incorrectness of utterances in context. It goes without saying, at this point, that N* will suffer the same fate in relation to these examples as it did in relation to the previous examples.

7

Kinds of correctness

The above-mentioned cases generated conflicts between our intuitions about the correctness and incorrectness of utterances and the implications

Semantic Normativity in Context

101

of N. This suggests that N is mistaken. N*, which is prima facie more plausible, in contrast, either fails to amount to the thesis that meaning is normative, or collapses into N. How might the normativist respond? The normativist could attempt to explain away our intuitions about correctness. Perhaps our intuitions about correctness are not sufficiently fine grained, and we confuse overall correctness, which is a function of both semantic correctness and other factors, and just semantic correctness, narrowly construed. That is, in addition to semantic correctness, there might be other kinds of correctness in play. An utterance may be correct, with respect to the semantic norms, but incorrect with respect to the norms of etiquette – as would be the case if one uttered a sentence that is true, but tactless. The normativist could argue, in all of the above-mentioned cases, our intuitions are about overall correctness, or pragmatic correctness, not semantic correctness. If we were to reflect, we would discover that our intuitions about semantic correctness do confirm the normativist’s hypothesis. For instance, in the case of (6), the normativist might argue that we are misled by the intuition that the utterance is pragmatically or instrumentally correct, because she succeeds in communicating the proposition she intended to communicate. However, if we ignore this feature of the situation, we will recognize that intuitively, the utterance is semantically incorrect because the sentence uttered is literally false. This reply is unsatisfactory not just because it is obviously ad hoc, but because it turns semantic correctness into a superfluous third wheel which has no explanatory value. A semantic theory assigns a literal meaning to (6), and determines the propositional content of (6), when (6) is evaluated relative to a context. On the basis of the literal meaning of (6), the sentence is literally false. So much is implied by the semantic theory, together with the way the world is in relevant respects. This explains the intuition that the sentence uttered is false, and there does not seem to be any further work to be done by the additional thesis that the utterance is semantically incorrect: semantic correctness has no further roll to play in explaining the intuition that this sentence is prima facie false. Furthermore, semantic correctness plays no role in explaining the intuition that the utterance is nevertheless correct – for, this is entirely explained by the fact that an utterance of (6) is a successful act of communication, and carries out the communicative intention of the speaker. All we need to explain the intuition that the utterance of (6) is correct is instrumental correctness. Thus, semantic correctness is not required to explain any of our intuitions in this case, neither about the falsity of the sentence uttered, nor about the correctness of the utterance.

8

An alternative account of semantic normativity

The foregoing discussion of semantic normativity focused on N, the idea that an utterance of a sentence is (semantically) correct if and only if the

102

Anandi Hattiangadi

sentence uttered is true. Alan Millar has put forward an alternative account of semantic normativity which deviates, in significant respects, from the foregoing view. Crucially, Millar holds that if one means green by ‘green’, it does not follow that one is normatively committed to apply ‘green’ correctly, but that one is normatively committed to use ‘green’ in a way that ‘respects its conditions of correct application’ (Millar 2004, p. 126). For example, he says: The upshot is that a use is correct in the sense of being in keeping with the relevant meaning provided that it is not a misuse. It is a misuse when it fails to respect the conditions of correct (= true) application of the term, that is, when it involves using the term as if those conditions of application, which are fixed by its meaning, were other than they are. An application that is correct, in the sense of being true, may or may not be correct, in the sense of being in keeping with the relevant meaning. An application that is correct, in the sense of being in keeping with the relevant meaning, may or may not be correct, in the sense of being true. Thus correctness of use, conceived as true application, is not the same notion as the correctness of use, conceived as use in keeping with meaning. (Millar 2004, p. 163) Millar goes on to argue that to speak a language is to participate in a practice, and that it is constitutive of the practice of speaking a language that speakers of the language are committed to following certain rules. According to Millar, the rules are not of the form ‘apply e to something if and only if it is F’, but rather, ‘use e only in ways that respect its conditions of true application, i.e., only in ways that are in keeping with its meaning F’. (Millar 2004, p.166) He claims that meaning is intrinsically normative because from ‘e means F in L’ it follows, without further ado, that speakers of L are committed to following the rule: use e only in ways that respect its conditions of true application. (Millar 2004, p.170) What is it to use an expression in a way that respects its conditions of true application? Millar suggests that it is to use it to pick out the property or object that it picks out, in the language. So, if I say ‘cut down the oak tree’, referring to the one and only oak tree in the garden, I use it in a way that is in keeping with its meaning. (Millar 2004, p.166) What is it to misuse an expression, i.e., to fail to respect the conditions of correct application of the term? Millar gives the following example. Suppose I think that ‘arcane’ means ancient, as opposed to known only to the initiated. Then, Millar says, I might be liable to misusing ‘arcane’. So, for example, if I think that ‘arcane’ means ancient, and say, (7) The practice of wearing Sub Fusc for examinations at Oxford is arcane,

Semantic Normativity in Context

103

I have misused the word ‘arcane’, even if this practice is only known to the initiated. That is, I have misused the word ‘arcane’ if I use it to pick out the property of being ancient, even if, in so doing, I utter a true sentence by a stroke of luck. According to Millar, I misuse a word if ‘I use it meaning to say something to the effect that the [practice] is ancient, but the word I use is not suitable for the purpose.’ (Millar 2004, p. 162). This example strongly suggests that, on Millar’s notion of correct/incorrect use, what makes a use of an expression correct or incorrect is not whether the sentence one utters is true or false. Rather, what makes the use of ‘arcane’ incorrect in the utterance of (7) is that I was assumed to have uttered (7) with the intention of picking out the property of being ancient. That is, what is wrong with the use of ‘arcane’ in the utterance of (7) is that the sentence uttered does not express the proposition that I intended to express by uttering the sentence – I intended to pick out the property of being ancient, but the word ‘arcane’ does not pick out this property in my language, so uttering this sentence does not satisfy my intention. If, in contrast, I uttered sentence (7) with the intention of picking out the property of being only known to the initiated, I would not have misused the word ‘arcane’. Since the truth value of (7), according to Millar, remains the same whether I intend to pick out the property of being ancient or the property of being known only to the initiated, what differs between these two cases is my intention – whether I intended to pick out the property of being ancient, or the property of being only known to the initiated. This suggests the following criterion of semantic correctness along the following lines: N**: An utterance, u, of s is correct if and only if s literally means that p, and the speaker intends to express p by uttering s. N** seems to capture Millar’s view, and presents it in the form of a criterion of semantic correctness. However, there is an obvious flaw with this proposal – it does not show that meaning is normative in any interesting sense. That is, N** says that an utterance of a sentence is correct just in case the sentence means the proposition that I intend to express by uttering the sentence. The correctness of the utterance thus appears to be merely instrumental – uttering a sentence that means the proposition that I intend to express by making the utterance is nothing more than a sufficient means by which I can satisfy my communicative intention. Second, Millar’s account of misuse has counterintuitive implications in several of the cases discussed. Let’s consider some of them. Suppose that I utter (1) with the intention of expressing the proposition that every bottle of alcoholic drink in the house is empty. This is not what (1) literally means. Hence, my utterance of (1) is incorrect, by Millar’s rights – I have misused the word ‘every’ which, in the context of that sentence, is unrestricted, but the proposition I intended to express contains a restricted quantifier. Once again,

104

Anandi Hattiangadi

this flies in the face of our intuition that an utterance of (1) is correct. Now, imagine that I utter (4) in the midst of a downpour, with the intention of expressing the proposition that it is raining here and now. (4) does not literally mean that it is raining here and now, so my utterance of (4) is incorrect, on Millar’s view, even though it is somewhat difficult to see which expression I have misused. Now, consider the waitress who utters (6), with the intention of expressing the proposition that the man who ordered the ham sandwich wants to pay. (6) does not literally mean that the man who ordered the ham sandwich wants to pay. The waitress intends to pick out the man who ordered the ham sandwich with the expression ‘ham sandwich’, but that is not what ‘ham sandwich’ picks out in her language. Hence, she has misused the words ‘ham sandwich’ and her utterance is incorrect. In all these cases, N** turns out to be just as counterintuitive as N. One reply, on Millar’s behalf, might be to exploit the vagueness of the concept of ‘respecting the conditions of true application’. That is, Millar might argue that in all of the cases where we intuitively think that an utterance of a sentence is correct, the utterance does, in some sense, respect the conditions of the true applications of expressions. So, when the waitress utters (6), she is, in some sense, respecting the conditions of the true application of ‘ham sandwich’ – she simply deferred reference from the ham sandwich to the guy who ordered it. And this is perfectly kosher, even if the sandwich is not, because deferring reference is one way of respecting the conditions of true applications of expressions. The trouble with this reply, however, is that if it is employed in general, to accommodate all of the countless cases of loose talk, the relevance of literal meaning to semantic correctness will be stretched to breaking point. Thus, not only is this notion of semantic correctness instrumental, but in order to cope with the above-mentioned cases, the relevance of the literal meanings of words, to what sentences one can utter to satisfy one’s communicative intentions, becomes very weak indeed.

9

Conclusion

I have employed a number of contextualist arguments against a few versions of the thesis that meaning is normative. These arguments provide good reason to deny that an utterance is correct if and only if the sentence uttered is true. It also shows that some more plausible alternatives do not capture semantic normativity. The dialectic of this chapter might suggest that I endorse the contextualist arguments, i.e., that I assume that the cases given do show that the contents of utterances must be highly context-sensitive. However, this is not so. For, unlike contextualists, the intuitions to which I have appealed concern not the truth conditions or contents of utterances, but whether the utterances are correct or incorrect. Since I deny the normativist thesis that correctness implies truth (and vice-versa), my arguments imply that intuitions

Semantic Normativity in Context

105

about the correctness of utterances do not have implications for assumptions about the truth or truth conditions of utterances. It is plausible that our basic intuitions about the various cases are indeed somewhat vague intuitions about correctness, at least in the case of ordinary speakers of the language, as opposed to theoretical linguists and philosophers of language. Similarly, Cappelen and LePore (2005) argue that the intuitions contextualists appeal to are too fine grained to be the intuitions of ordinary speakers of the language, untutored in theoretical linguistics and philosophy of language. Even if ordinary speakers have intuitions about the contents of utterances, they don’t, and can’t be expected to have, intuitions about whether the contents they assign to utterances are implicatures, explicatures (utterance contents), or contents of the sentences uttered. On my view, the intuitions ordinary speakers have are best understood as rather vague intuitions about the correctness of utterances. If this is true, then many of the intuitions of ordinary speakers of the language about the correctness of utterances can be explained by reference to speaker meaning and speaker intention. In those cases where we think that an utterance is correct, though the sentence uttered is literally false, and in those cases where we think that the utterance is incorrect, thought the sentence uttered is literally true, our intuitions about correctness and incorrectness can be explained by our assumptions about speaker intention. This style of explanation of intuitions about correctness makes the postulation of contextually variable utterance content superfluous. This is a line of argument I’d like to explore further. But it will have to wait for another occasion.

Acknowledgements The research towards this chapter was partly conducted while I was a visiting fellow at the Swedish Collegium for Advanced Studies (SCAS), in Uppsala, Sweden, and was generously supported by the Oxford Philosophy Faculty Board. Earlier versions of this chapter were presented at SCAS, at the Department of Philosophy in Stockholm University, and with my colleagues at the Philosophy Faculty, Oxford University. I am grateful to all of the participants of all of these sessions for their insightful comments and suggestions. Special thanks to Gustaf Arrhenius, Michael Blome-Tillman, Krister Bykvist, Cian Dorr, Ant Eagle, Katalin Farkas, Kathrin Glöer-Pagin, Sten Lindström, Peter Pagin, Wlodek Rabinowicz, Folke Tersman, Gabriel Uzquiano, and Åsa Wikforss.

Notes 1. Of course, this sounds trivial, because the meta-language – the language in which the theory is stated – is the same as the object language – the language that the

106

2. 3.

4.

5.

Anandi Hattiangadi

theory is about. But it sounds less trivial when these are different. E.g., a theory expressed in English (the meta-language) of Swedish (the object language) would say that the expression ‘snö’ means snow – ‘snö’ refers to some stuff if and only if it is snow. It is just simpler to use English for both the object language and the meta-language. This characterisation is due to Cappelen and Lepore (2005). According to the covert variable views, the role of context will be constrained by a feature of the logical form, so the proposition expressed by the utterance, in the context, will be identical to the proposition expressed by the sentence uttered in that context, evaluated relative to the context of utterance. Borg (2004) argues that even if we postulate covert variables to systematize the role of context in determining truth conditional content, we will still need to appeal to communicative intentions to determine utterance content. The postulation of the covert variable only directs the hearer to search for some suitable part of the context to restrict the quantifier, but ultimately, this can only be filled in by subjective, as opposed to objective, features of the context, such as the speaker’s communicative intention. The temporal index seems to be semantically mandated by the fact that the sentence is in the present tense. But there is no location marker in the syntax of the sentence.

Bibliography Bach, K. (1994) ‘Conversational Implicature’, Mind & Language, 9, 124–62. —— (2005) ‘Context ex Machina’, in Z.G. Szabó (ed.) Semantics versus Pragmatics (Oxford: Oxford University Press). Boghossian, P.A. (1989) ‘The Rule-Following Considerations’, Mind, 98, 507–49. —— (2003) ‘The Normativity of Content’, Philosophical Issues, 13, 31–5. Blackburn, S. (1984) ‘The Individual Strikes Back’, Synthese, 58, 281–301. Borg, E. (2004) Minimal Semantics (Oxford: Oxford University Press). Brandom, R. (1994) Making It Explicit (Cambridge, Mass.: Harvard University Press). Broome, J. (1999) ‘Normative requirements’, Ratio, 12, 398–419. Cappelen, H. and Lepore, E. (2005) Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act Pluralism (Oxford: Blackwell). Carston, R. (2002) Thoughts and Utterances: The Pragmatics of Explicit Communication (Oxford: Blackwell). Gibbard, A. (1994) ‘Meaning and Normativity’, Philosophical Issues, 5, Enrique Villanueva (ed.) Truth and Rationality, 95–115. Glock, H.J. (1996) ‘Necessity and Normativity’, in H. Sluga and D.G. Stern (eds) The Cambridge Companion to Wittgenstein (Cambridge: Cambridge University Press). Glüer, K. (1999a) Sprache und Regeln. Zur Normativität von Bedeutung (Berlin: Akademie). —— (1999b) ‘Sense and Prescriptivity’, Acta Analytica, 14, 111–28. Glüer, K. and Pagin, P. (1999) ‘Rules of Meaning and Practical Reasoning’, Synthese, 117, 207–27. Grice, P. (1989) Studies in the Way of Words (Cambridge, Mass.: Harvard University Press). Hattiangadi, A. (2002) ‘Is Meaning Normative?’ Mind & Language, 21, 2, 220–40. —— (2007) Oughts and Thoughts: Rule-Following and the Normativity of Content (Oxford: Oxford University Press).

Semantic Normativity in Context

107

—— (2009) ‘Some More Thoughts on Semantic Oughts: A Reply to Daniel Whiting’, Analysis, 69, 54–63. Kaplan, D. (1989) ‘Demonstratives’ and ‘Afterthoughts’, in J. Almog, J. Perry, and H. Wettstein (eds) Themes from Kaplan (Oxford: Oxford University Press). Kripke, S. (1982) Wittgenstein on Rules and Private Language (Cambridge, Mass.: Harvard University Press). McGinn, C. (1984) Wittgenstein on Meaning (Oxford: Blackwell). Millar, A. (1998) Philosophy of Language (London: UCL Press). —— (2002) ‘The Normativity of Meaning’, in A. O’Hear (ed.) Logic, Thought and Language (Cambridge: Cambridge University Press). —— (2004) Understanding People (Oxford: Oxford University Press). Perry, J. (1993) The Problem of the Essential Indexical and Other Essays (New York: Oxford University Press). Recanati, F. (2004) Literal Meaning (Cambridge: Cambridge University Press). Searle, J. (1978) ‘Literal Meaning’, Erkenntnis, 13, 207–24. Sperber, D. and Wilson, D. (1986) Relevance: Communication and Cognition (Oxford: Wiley-Blackwell). Stanley, J. (2000) ‘Context and Logical Form’, Linguistics and Philosophy, 23, 391– 424. Stanley, J. and Szabó, Z. (2000) ‘On Quantifier Domain Restriction’, Mind & Language, 15, 219–61. Travis, C. (1985) ‘Meaning’s Role in Truth’, Mind, 100, 451–66. Whiting, D. (2007) ‘The Normativity of Meaning Defended’, Analysis, 67, 133–40. Wikforss, Å. (2001) ‘Semantic Normativity’, Philosophical Studies, 2, 203–26. Wright, C. (1980) Wittgenstein on the Foundations of Mathematics (London: Duckworth).

6 Literal Force: A Defence of Conventional Assertion Max Kölbel

But literal meaning may not (and in my view does not) go beyond truth conditions. Donald Davidson (‘Communication and Convention’, p. 269) Our difficulty arises from the fact that we have tried to characterize the activity of assertion without taking into account its being a conventional activity: the fact that a sentence expresses an act of assertion is as much a matter of linguistic convention as is its having the sense it has. Michael Dummett (Frege: Philosophy of Language, p. 300.) The aim of this chapter is to motivate and defend a conventional approach to assertion and other illocutionary acts.1 Such an approach takes assertions, questions and orders to be moves within an essentially rule-governed activity similar to a game. The most controversial aspect of a conventional account of assertion is that according to it, for classifying an utterance as an assertion, question or command, ‘it is irrelevant what intentions the person speaking may have had’ (Dummett 1973, p. 302). I understand this to mean that it is irrelevant for the issue of whether an utterance is an assertion whether the utterer has certain communicative intentions, such as the intention to utter something true, the intention to get one’s audience to believe (that one believes) what one has asserted etc. Just as one can commit a foul in football without meaning to do so, one can make an assertion, issue a command or ask a question without meaning to do so. The rules of football specify that a certain form of conduct (tackling an opponent in a certain way), carried out under certain general conditions (being a member of a team engaged in a game of football) counts as committing a foul. Similarly, I claim, the rules of language specify that a certain form of conduct (uttering an assertoric sentence), carried out under certain general 108

Literal Force

109

conditions (being a member of a speech community engaged in a conversation) counts as making an assertion. It is not part of the conventional approach defended here that there cannot also be a useful notion of assertion that is defined in terms of some suitable cocktail of communicative speaker intentions. On the contrary, a conventional account of assertion and other illocutionary acts is complemented by a pragmatic theory of the communicative intentions and actions of speakers. For this reason, I will distinguish conventional notions of C-assertion from notions of I-assertion: assertion defined in terms of the subject’s communicative intentions. What is defended here is not an analysis of any pre-theoretical notion of assertion. Rather, I am defending the claim that C-assertion, C-question etc have an important and coherent role to play in a theory of linguistic communication. The purpose of a conventional account of assertion is perhaps best explained by viewing it as the proposal that there is literal force in addition to literal content and literal sub-content. Most philosophers of language make a distinction between the literal, encoded meaning of utterance types, and the non-literal meaning which tokens (or tokenings) of those types can have on particular occasions of use.2 Without the notion of literal meaning, we would presumably be forced to deny the existence of context-invariant meanings altogether. However, most philosophers of language employ such a distinction only at the level of the content, or sub-content, of utterances, not at the level of illocutionary force or speech acts. Most theorists have no room for a notion of literal assertion, i.e., a performance that counts as an assertion in virtue of the linguistic meaning of the utterance type used. Instead, these theorists employ a notion of assertion according to which asserting is a matter of having certain communicative intentions, with the result that they see themselves forced to deny the existence of illocutionary force indicators, i.e., context-invariant meaning at the level of speech acts. In this chapter, I want to provide some motivation for admitting a notion of literal assertion and distinguishing this from assertion conceived of in terms of speaker intentions. The second aim is to dispose of some objections that have been made against this type of theory, namely Davidson’s and Stainton’s objections against Dummett’s conception of assertion as a conventional act. Sections 1–3 are introductory. I begin in Section 1 with some general considerations about essentially rule-governed action. In Section 2 I give a general characterization of conventional accounts of assertion. In Section 3 I provide a general characterization of intentional accounts of assertion. Sections 4–6 provide various motivations for employing a conventional notion of assertion and other speech acts. In Section 4 I explain the role a conventional notion of assertion can play in dealing with indirect speech acts. In Section 5 I argue that the usual reasons for distinguishing between literal and non-literal meaning concern the force as much as the content

110

Max Kölbel

of utterances. In Section 6 I show how the notion of conventional force helps avoid the sorts of problems Grice and Lewis have in grounding linguistic meaning in language use (or characterizing the actual-language relation). Section 7 deals with Davidson’s objection against conventional force indicators, namely his argument from ‘non-serious’ uses, such as in jokes or fictional utterances. Section 8 finally tackles Stainton’s alleged counterexamples to Dummett’s conventional account of assertion, namely non-sentential assertions.

1

Essentially rule-governed action

Many actions are essentially rule-governed and in that sense conventional. For example: buying a newspaper, scoring a goal or playing the ball out in football, signing an assured short hold tenancy agreement, expelling someone from the Reform Club, registering an objection to a planning application, betting ten pounds on Ronnie O’Sullivan to win. What these actions have in common is that they are, or are part of, activities that are constitutively governed by norms and conventions. They depend for their existence on certain human-made rules. For example, without the 1988 Housing Act the action of signing an assured short hold tenancy would not exist, and without the rules of football, nothing would count as playing the ball out in football.3 In my discussion, I will distinguish two different kinds of rules: implementation conventions and constitutive norms. Let me explain. Every rule-governed activity must have a medium in which it is implemented. For example, any (concrete) game of chess must be implemented in some medium, such as a wooden chess-board or constellations of digital encodings. Because any conventional action depends for its existence on certain rules, performing the action must be constituted by some other action or procedure that does not itself depend for its existence on the same rules. To see this, try to imagine a conventional action which is not implemented non-conventionally. Suppose action A exists only because of certain rules R a certain group adheres to. But there is no action independent of R that constitutes performing A. This seems incoherent, for how would the members of the group perform A? The only sense in which rules can bring about the existence of new kinds of action is this: the rules stipulate that a certain type of independently existing action is to have a certain status, a status which is further specified by a number of norms. This means that for any conventional act-type C, there must be a certain non-conventional act-type N performing which counts as performing C. What I call ‘implementation conventions’ lay down which non-conventional procedure N counts as performing C.4 Thus, for example, handing a coin to the newspaper man while grabbing a newspaper with the other hand counts as buying a newspaper. Turning the doubling die in a certain way in the course of a game of backgammon counts as doubling.

Literal Force

111

Implementation conventions thus stipulate which non-conventional procedure counts as the conventional act in question. However, they do not yet introduce any norms properly so-called, namely obligations, permissions or rights, i.e., the sort of convention that can be violated. In addition to the stipulation of what counts as performing C, there also need to be constitutive norms that make up the normative significance of C. For example, in addition to knowing what type of conduct counts as a foul in football, we also need to know what obligations and rights arise from a foul. Someone who didn’t know these things wouldn’t fully understand the conventional act of committing a foul in football. Similarly, the implementation convention which says that handing over a coin counts as buying the newspaper does not exhaust the constitutive rules concerning that activity. Buying a newspaper crucially gives rise to certain rights and obligations: the buyer acquires property rights over the copy he bought. I call these norms ‘constitutive norms’. There can be a certain arbitrariness about the implementation conventions, i.e., the rules that say which non-conventional procedures count as the conventional action in question. That’s why I call them ‘implementation conventions’. Sometimes, these conventions are not essential to an act. In other words, a different non-conventional procedure could have been taken to count as a given conventional act C, and it would still have been C. For example, backgammon could be implemented in a very different medium, including (very uncomfortably) in a purely one-dimensional medium. It would still be backgammon. Another example: handing a credit-card to the newspaper man and signing the slip he hands back to you also counts as buying a newspaper. This could have been the only procedure that counts as buying a newspaper, if the implementation conventions of buying had been different. It would still have been the act of buying a newspaper. In other cases, one particular non-conventional implementation is essential to a conventional act type or activity.

2

Asserting as rule-governed activity

Linguistic communication has an obvious similarity with the conventional activities described. It seems that in linguistic communication going through certain procedures or forms of conduct counts as making one of a range of moves, and making these moves has a certain normative significance. For example, going through the procedure of uttering ‘Hello’ (in the right tone of voice) when addressing someone counts as greeting that person, and greeting a person has a certain normative significance (e.g., failure to greet may be regarded as offensive; being greeted by someone creates the expectation that one greet back etc). Another example: going through the procedure of uttering ‘The train is late’, in normal circumstances, counts as asserting that the train is late. The normative significance of this action

112

Max Kölbel

consists in it bringing about certain rights and obligations concerning the asserter and his or her audience. For example, the assertion might introduce the conversational background information that the train is late, with the result that this may be presupposed in subsequent utterances. Another kind of normative significance is not purely linguistic: if a member of the audience of this assertion were to miss the train because the train was on time but he or she was relying on the assertion that the train was late, then this member of the audience would have a right to reproach the asserter, or to impose other sanctions. Whether and how he or she may express his or her reproach, or what sanctions would be appropriate, are ethical questions, not linguistic ones. One of the guiding ideas of this chapter is that we ought to take seriously the analogy of linguistic communication with other rule-governed activities. The following aspect of the analogy is especially important for my purposes. Consider a game like football. It is useful to distinguish objectives internal to the game from the external aims which are pursued through playing the game. Thus scoring goals and winning are objectives which are internal to the game, while entertaining oneself or others, improving one’s health or earning money are aims that are external to the game and that can be pursued by making moves within the game. Thus playing a game is not only non-conventionally implemented, it can also itself implement actions that are not themselves part of the game, such as enjoying oneself or earning money. I believe that there is a similar distinction in the case of linguistic communication. On the one hand, we make certain linguistic moves which have certain purely linguistic purposes and consequences. On the other hand, we pursue non-linguistic goals by making these linguistic moves. For example, I may pursue the linguistic purpose of manipulating the common ground of a conversation in order to serve the extralinguistic aim of transmitting information. I believe that a distinction between innerlinguistic and extra-linguistic purposes is crucial to understanding linguistic communication. My aim in this chapter is specifically to carve out a role for assertion (and other illocutionary acts) as an innerlinguistic move which cannot be fully understood in terms of communicative intentions. Just as one can distinguish between innerlinguistic and extra-linguistic purposes, one can also distinguish between purely linguistic norms and other social norms, such as legal, moral or prudential norms. It is difficult to draw this distinction and this is one reason why it is difficult to say what precise set of norms is constitutive of assertion understood as an innerlinguistic, conventional act. For example, asserting that McDonalds food is unhealthy may constitute libel, and committing libel is a conventional act that can have drastic normative consequences.5 However, intuitively the norms concerning libel are not themselves constitutive of assertion.6 A number of theorists share the view that assertion is essentially normative. (Though they do not, as I do, distinguish assertion as a conventional

Literal Force

113

act from assertion as an act performed with certain communicative intentions.) For example Williamson (1996 and 2000, chapter 11) suggests that the only norm constitutive of assertion is the norm that prohibits asserting propositions one does not know.7 Brandom (1983 and 1994) claims that asserting a proposition brings about an obligation of the asserter to justify her assertion if challenged and it also brings about a permission for the audience to rely on the proposition asserted as a premise. Brandom does not distinguish general social norms from specifically linguistic ones. The framework in my view best suited to the idea of conventional (literal) assertion is that pioneered by Stalnaker (1973, 1974, 1978, 2002), and Lewis (1979). According to this framework, assertions and other linguistic acts have a characteristic effect on the ‘context’ or ‘conversational score’ of a conversation. In my view, modification of the conversational score can be seen as the internal aim of linguistic communication and communicators pursue this internal aim in order to further their non-linguistic aims. Conventional (literal) assertion is an innerlinguistic move the normative role of which can be completely characterized by specifying the rules by which the conversational score changes. In particular, these rules will include the rule that if a proposition is asserted then it is added to the conversational score unless the assertion is challenged. However, this is not the place to defend this account in detail.8 The purpose of this chapter is merely to argue that a notion of assertion as conventional action (literal assertion) has a number of theoretical attractions and can be defended against some objections in the literature. In the current context, I will therefore operate with a schematic formulation of an account of assertion as an essentially normative, conventional action. Let’s suppose, for the rest of this chapter, that the account of the norms constitutive of assertion has the following form: (A1) For all p, necessarily, if a speaker has asserted that p, N. where ‘N’ is schematic for whatever the constitutive norms of assertion are.9 I have not yet said anything about the implementation conventions concerning assertion, i.e., those conventions that say which procedure counts as asserting p (for variable p). According to Sadock and Zwicky (1985), all natural languages contain some distinction between what they call ‘sentence-types’, namely declarative, imperative and interrogative. Many theorists call the features of sentences that indicate illocutionary force ‘moods’ (whatever those features may be). This is slightly misleading because ‘mood’ originally referred to certain inflectional properties of verbs (see Harnish 1994, p. 408–9). It is clear that there are many languages, including English, where the function of indicating illocutionary act types is not implemented by a systematic inflectional feature of

114

Max Kölbel

the principal verb, even though some systematic such modifications are quite typical. Nevertheless I will follow custom and speak of moods as the conventional indicators of force, in whatever way they are implemented. That moods exist and are somehow correlated with the force of sentences is, I believe, undeniable (see again Sadock and Zwicky 1985 for a crosslinguistic study). Even Davidson admits that there is some conventional correlation between moods and illocutionary force (1979, pp. 114, 116). What is controversial is whether mood can conventionally indicate that an utterance has a certain force in the same way in which, say, use of the pronoun ‘I’ can conventionally indicate reference to the speaker, or the procedure of writing one’s name on a piece of paper under certain conditions counts as signing a contract. I shall defend this more controversial thesis, namely that there are implementation conventions of this sort in the case of assertion: (A2) There are conventional assertoric force indicators which reliably indicate that an utterance counts as an assertion (as characterized by the correct instance of (A1)). Let us call any account of assertion that combines an instance of (A1) with (A2) a ‘conventional account’ (or ‘C-account’) of assertion, and any notion of assertion that conforms to a conventional account a notion of ‘conventional assertion’, or ‘C-assertion’. The aim of this chapter is to advance some considerations in favour of employing a notion of C-assertion. Strawson (1964) characterizes a notion of essentially conventional speech act that is very close to what I have been trying to characterize. He draws a distinction between two types of speech act. One type is ‘essentially conventional’ speech acts, which ‘could have no existence outside the rule- or convention-governed practices and procedures of which they essentially form parts’ (1964, p. 36). Strawson lists marrying, redoubling, giving out, pronouncing sentence, bringing in a verdict. Speech acts of this kind are ‘standardly intended to further, or affect the course of, the practice in question’, but they can in fact be performed unintentionally (‘A player might let slip the word ‘redouble’ without meaning to redouble; but if the circumstances are appropriate and the play is strict, then he has redoubled’ [1964, p. 36]). He clearly identifies something analogous to (A2) as the mark of these essentially conventional acts: ... the act is identified as the act it is just because it is performed by the utterance of a form of words conventional for the performance of that act. Hence the speaker’s utterance is not only intended to further, or affect the course of, the practice in question in a certain conventional way; in the absence of any breach of the conventional conditions for furthering the procedure in this way, it cannot fail to do so. (1964, p. 36)

Literal Force

115

Thus Strawson’s essentially conventional illocutionary acts are essentially part of a rule- or convention-governed practice (which corresponds to (A1)), and there are also overt procedures which, when performed under the right circumstances, are sufficient for performing the act in question. Strawson does not mention illocutionary acts like assertion. It is the current aim to argue that there is a useful notion of assertion as essentially conventional act, i.e., of C-assertion.10 Before moving on to intentional notions of assertion, I would like briefly to mention Dummett’s views on assertion, which, I believe, are a good example of a C-account. In the chapter on assertion of his Frege: Philosophy of Language, Dummett distinguishes two approaches to assertion: We ... tried to specify which [utterances] constituted assertions by reference to the intention of the speaker. Rather, the correct approach is to consider utterances as conventionally demarcated into types, by means of the form of linguistic expressions employed, and then to enquire into the conventions governing the use of the various types of utterance. (p. 302) According to the first approach, the act of asserting is defined in terms of certain speaker intentions. The same sentence with the same conventional content can be uttered with various different concomitant intentions. Whether an utterance is an assertion depends on the nature of these concomitant intentions. For example, making an assertion might, on one version of this approach, require that the speaker intends to say something true. Or on a different version, it might require that the speaker intends the audience to believe the speaker to intend to be saying something true. By contrast, according to the second approach, ‘it is irrelevant what intentions the person speaking may have had’ (p. 302). Dummett defends the latter, conventional approach.

3

Assertion as a matter of speaker-intention

C-assertion is an essentially conventional or social act which could be performed inadvertently, by virtue of being a participant in a conversation and exhibiting the kind of conduct that conventionally counts as C-asserting. Many theorists, especially Grice and many influenced by him, have a very different conception of assertion and illocutionary force.11 These theorists prefer an intentional account of assertion (or I-assertion). On an I-account, assertion (as well as the other illocutionary acts) is essentially utterances made with certain audience-directed intentions. One kind of I-account says that in order to assert that p one needs to speaker-mean that p. This involves that the asserter have intentions regarding some perlocutionary effect on the audience. Asserting that p, on such a view requires producing some perceptible stimulus with the intention of getting one’s audience to

116

Max Kölbel

believe that (one believes) p.12 Following Grice, the intention to get the audience to believe (that one believes) p must also be accompanied by the intention that this belief arise in part because the audience recognizes the first intention. But the important unifying feature of all I-accounts of illocutionary forces is that each illocutionary act-type requires that the asserter have certain audience-directed communicative intentions. For example, it may require the asserter’s intention that the audience believe what was asserted or that they make true what was requested or that the audience answer what was asked, etc.13 In order to illustrate the difference between C-accounts and I-accounts of assertion, consider the following two utterances of the same sentence: (1) The door is open. (as uttered in response to the question ‘Why is there such a draft?’) (2) The door is open. (as uttered in order to get the audience to shut the door, in the full knowledge that the audience already knows, and knows that the speaker knows, that the door is open.) Utterance (1) will qualify both as an I-assertion and as a C-assertion: the speaker has asserted that the door is open. However, I-account and C-accounts will diverge on whether utterance (2) is an assertion. Utterance (2) is not intended to get the audience to believe (that the speaker believes) that the door is open and is clearly intended by the speaker to get the audience to close the door. A similar effect could be brought about by an utterance of ‘Shut the door!’ This may lead those with an I-account of assertion to argue that utterance (2) is an order or a request rather than an assertion. For the speaker does not have the communicative intentions required for making an I-assertion, but she does have intentions sufficient for a making a request. C-theorists of assertion, by contrast, might hold that utterance (2) is an assertion, because it is the performance of an act type that conventionally counts as an assertion, namely the utterance of a sentence in the assertoric mood. But nevertheless, by asserting that the door is open, the speaker might achieve certain perlocutionary effects, and might do so intentionally. She might intend to get, and succeed in getting, the audience to infer that she wishes the door to be shut and thereby get the audience to shut the door. Thus a C-theorist might hold that (2) is both a C-assertion and an I-command.

4 The argument from indirect speech acts Cases like (2) might be used, by adherents of I-accounts of assertion and command, to argue that there are no conventional force indicators, i.e., no forms of conduct that conventionally count as making assertions, issuing commands etc. For no linguistic convention can dictate what communicative

Literal Force

117

intentions speakers have when they use sentences. Davidson seems to argue in just this way: ... there are many utterances of indicative sentences that are not assertions, ... and assertions may be made by uttering sentences in other moods. ... And similarly for the other moods; we can ask a question with an imperative or indicative (‘tell me who won the third race’, ‘I’d like to know your telephone number’), or issue a command with an indicative (‘In this house we remove our shoes before entering’). (1979, p. 110) However, it is quite clear that this argument only affects the question whether I-assertion is conventionally indicated by the indicative mood. The hypothesis Davidson wants to disprove is the ((A2)-like) hypothesis that: (H) The assertoric mood is a conventional indicator of assertoric force. i.e., the view that by an implementation convention utterances of an assertoric sentence (under certain normal conditions) count as assertions. One of the purported counterexamples is an utterance of the assertoric sentence ‘In this house we remove our shoes before entering’ which is made with the overt intention of getting the audience to remove their shoes, i.e., it is an I-command. The view that this is a counterexample to (H) relies on two assumptions. First, the assumption that an I-command cannot also be an I-assertion. This is not obvious, but let’s grant it for the sake of argument. The second assumption: ‘assertion’ in (H) is to be read as meaning I-assertion, so that certain communicative intentions in the asserter are required for assertion. Even if, on these assumptions, the utterance of ‘In this house ...’ is a counterexample to (H), it is not yet a counterexample to a C-reading of (H). Those who believe that uttering an assertoric sentence counts (under certain normal conditions) as an assertion have in mind C-assertion, not I-assertion. Thus Davidson’s cases are not counterexamples to the hypothesis that an utterance of a sentence in assertoric mood (under certain normal conditions) counts as an assertion. When the convention-theorist claims utterance (2) is an assertion and Davidson says utterance (2) is not an assertion, they do not mean the same with ‘assertion’. One is using an I-notion, the other is using a C-notion of assertion. It might therefore appear that the debate about whether there is a conventional sign for assertoric force is based on a terminological confusion. But the issue is not just terminological. One substantial question is whether a theory of linguistic communication can dispense with one or the other of the two notions of assertion. I believe that both notions are probably indispensable.14 Because the C-notion is the one whose legitimacy I am

118

Max Kölbel

defending here, I shall offer in the next section some reasons for thinking that C-assertion is an indispensable notion. The preceding discussion was a simplification of Davidson’s argument, for Davidson is in fact not a clear exponent of an I-account of assertion (and the other illocutionary types). It will be useful to examine in detail the notion with which Davidson operates. Davidson explicitly denies a Gricean I-account of assertion: The asserter may or may not, in making an assertion, intend to cause his hearer to believe he believes what he says. (1982, p. 268) Nevertheless in some places, Davidson seems to subscribe to some sort of I-account of assertion, for he requires certain speaker intentions that preclude inadvertent assertion: ... in making an assertion, the asserter must intend to make an assertion, and he must intend that this intention be recognized by his audience. (1982, p. 269) And Of course assertion or command must be intentional, as must meaning in the narrow sense. But it is part of the intention that the act should be interpreted as assertive or commanding, ... (1979, p. 114) However, there are other remarks in which he seems to support a C-account: Making an assertion is, then, like playing a game in a respect in which speaking the truth is not: there is a public presumption of purpose. (1982, p. 268) And To assert is, among other things, to represent oneself as believing what one asserts. (1982, p. 270) One might think that it is possible unintentionally to represent oneself as believing something, and inadvertently to make a move that carries a public presumption of a certain purpose. This is why the last two remarks might seem in tension with two preceding remarks. However, there is no such tension if Davidson believes that publicly representing oneself as believing something requires that one have the intention of doing so. If this exegesis is correct it explains why Davidson does not want to allow a conventional

Literal Force

119

sign of assertion: he believes that assertion essentially involves certain communicative intentions, for example the intention to represent oneself as believing and the intention that this be recognized. However, there cannot be a convention that connects ‘what may always be secret ... with what must be public’ (1982, p. 270). So there cannot be a conventional indicator of assertoric force.

5

Literal content and literal force

Before discussing another Davidsonian argument in Section 7, I would like to provide some arguments why a theory should (possibly in addition to any I-notions of illocutionary force) employ a C-notion. Consider the following utterances: (3) You are the salt in my soup. (4) Of course I’ll pay €199 for a hair cut. I can’t think of a better way of spending my money. The distinction between literal and non-literal meaning seems very well suited to explain how these utterances can serve communicative aims. (3) is literally an assertion of the obviously false proposition that the addressee is the salt in the utterer’s soup. The audience knows this and will be able to work out the content the utterer intends to communicate, for example that the addressee is in some sense indispensable for the utterer. In whatever way the audience computes the non-literal content: they need the literal content of (3) (or at least the literal content of the components of (3)15) as initial input. It seems impossible to explain how communication could succeed otherwise in such cases. Similarly, the two propositions literally asserted in (4) are needed by the audience to work out the non-literal content. The view that in such cases communication relies on literal, semantically encoded meaning as an essential step in interpretation seems compelling.16 If we accept the theoretical justification of a distinction between linguistically encoded and actually communicated contents (or sub-contents), then by analogy we should explain why (2) is an I-request on the basis of the fact that (2) is a C-assertion. Here is how this might go: (2) is a C-assertion of a proposition that is already known by all participants and known to be known by all of them. For this reason the point of (2) cannot be to I-assert that proposition. Suppose, for example, that it is part of the conversational score or common ground (accepted by all participants) that it is desirable that the door be shut if open. Under those conditions, (2) can work as an I-request that the door be closed, for the speaker has brought it about that it is now part of the common ground of the conversation that the door is open. Adding this information to the common ground will make it part of the common ground that the door ought to be shut by

120 Max Kölbel

someone now. This will provide a motivation to shut the door, or, if perhaps it is also part of the common ground that one particular member of the audience is responsible for shutting the door, then that member will be motivated. This is just one possible account of the process that might allow a hearer to interpret (2) as an I-command. But any adequate account will need to make reference to two elements: (i) the sentence’s conventional assertoric force (it is not a coincidence that the sentence uttered is assertoric) and (ii) some contextual information that encourages the interpretation of (2) as an I-command. Element (i) might be disputed in the following way. All an adequate account needs to say is that (2) expresses the proposition that the door is open, thus making the obvious fact that the door is open salient and thereby triggering the background imperative that the door be closed if open. However, this account cannot explain why it is not equally appropriate to utter an interrogative sentence with the same content. If, in this situation, a literal question is to function as an I-command to close the door, then it would be the I-question whether the door is closed, not the I-question whether the door is open. If you have doubts about this consider a parent who wants to get a child to do his homework. The parent might say: ‘You have not yet done your homework.’ or ‘Have you done your homework?’ but surely not ‘Have you not done your homework?’ It is also possible to provide other examples where literal force undeniably plays a role. Consider the following utterance: (5) Do fish swim? (uttered in response to the question whether politician X is corrupt) Clearly, the utterer of (5) is not I-asking whether fish swim. Nevertheless, the information that (5) is a question in some sense (i.e., a C-question) is clearly needed to interpret the utterance as suggesting that the question asked has an obvious answer. In sum: if cases like (3) and (4) provide a motivation for admitting literal contents (be it of entire sentences or their subsentential constituents), then cases like (2) and (5) surely provide motivation for admitting literal forces. It seems to be generally accepted that semantically encoded literal contents (or sub-contents) play a key role in non-literal communication. However, the case is just as good for semantically encoded literal illocutionary forces. Any pragmatic account of how an audience can arrive at an interpretation of (2) as an I-command will start with some literal meaning as input. It seems to me that the case for literal forces as input to this pragmatic process is just as compelling as the case for literal contents (sub-contents) as a starting point in cases (3) and (4). This leads me to a related consideration in favour of C-assertion and C-forces in general. It seems quite clear that moods as they actually occur

Literal Force

121

in natural languages (i) have some conventional significance and (ii) this conventional significance plays a role in the defensible correlation often observed between moods and illocutionary forces in the I-sense. (This is observed, for example, by Davidson 1979, p. 115–16.) We ought to explain the conventional significance mentioned in (i), and this should provide a basis for explaining (ii). Moods that conventionally indicate C-forces would be a way to do both these things.17 Let me flesh out how this might go. Suppose we have a characterization of the normative role of C-assertions (some instance of (A1)). If this characterization is adequate, it will explain why, very often, speakers make C-assertions in order to make I-assertions. Thus, for example, Williamson might attempt to explain why we very often perform an act that is governed by his knowledge rule (‘Do not assert what you do not know!’) in order to I-assert. For example, someone might C-assert that Kripke works at CUNY because they want their audience to believe that Kripke works at CUNY (i.e., they are I-asserting that Kripke works at CUNY). This would be an appropriate means to that end because C-asserting that p generally exposes one to criticism should it turn out that one does not know that Kripke works at CUNY, so the audience may assume that one at least has reasonably good grounds for believing that one knows that Kripke works at CUNY now. Someone following a Brandomian line might say that C-assertion involves the obligation to justify what is asserted when asked to do so, and the license of recipients to rely on what has been asserted as a premise. On this view, performing the conventional act of C-asserting that Kripke works at CUNY is again a good means for achieving the aims characteristic of I-assertions that Kripke works at CUNY, namely getting the audience to believe it. A third account (the kind of account I currently favour) would take C-assertion essentially to bring about a change in the conversational score (common ground). For example, if it is C-asserted that Kripke works at CUNY and the assertion is not challenged, then it becomes part of the conversational score that Kripke works at CUNY (and if the score already contains the proposition that Kripke works exclusively at Princeton, then this part of the score will have to be revised). On this view, participants of conversations use the conversation game to gain information and to influence the beliefs of others. Thus, participants may come to believe what others have asserted because these others have exposed themselves to possible criticism should their assertions turn out to be unfounded. Or participants may reasonably expect an audience to come to believe what they have asserted precisely because the audience knows that the asserter has exposed him or herself to criticism should his or her assertion turn out to be unfounded. On all these views of C-assertion we have an explication of (i) the conventional significance of the moods and (ii) an explanation of why the moods are correlated significantly (but imperfectly) with corresponding I-acts.

122 Max Kölbel

6 Meaning and use: escaping the complex-intentions trap Another considerable advantage of a notion of C-assertion is that it helps explain the relation between linguistic meaning and language use. It does so at the cost of giving up the project of reducing linguistic meaning to the mental states of individuals, but in favour of an approach that operates with an unreduced notion of norm. It is attractive and plausible to suppose that the linguistic meaning of utterance types is determined by the use speakers make of these types. Several philosophers have made attempts at explaining this relation of determination, Grice and Lewis among them. Let us briefly consider Grice’s attempt at explaining sentence meaning in terms of speaker meaning. Grice regards speaker meaning (meaningNN, occasion meaning) as basic (e.g., 1989, p. 117), and attempts to elucidate the conventional meaning of utterance types in terms of this basic notion. The rough idea is that an utterance type means that p just if speakers use it to mean that p. In refining this rough idea, Grice faces a number of problems. The first problem is that of constructing the right notion of speaker meaning. If the idea is that roughly an utterance type means something just if speakers use it to mean that thing, then the notion of speaker meaning must somehow capture (at least) those uses that are meaning constitutive, i.e., ‘central’ uses. Grice himself operates with an intuitive constraint on speaker meaning. But it is clear that if his attempt to define sentence meaning in terms of speaker meaning is to succeed, the notion must also be such that central (or meaning constitutive) uses are cases where speaker meaning and sentence meaning coincide. Now, the original (1957) account defined speaker meaning along these lines: (SM1) An utterer U means that p by uttering x just if there is an audience A such that U intends A to come to believe p (at least partly on the basis of recognizing that intention). But there are counterexamples, such as the following utterance: (6) The Battle of Hastings was fought in 1066. (uttered by a pupil in an exam.) Clearly, (6) is among the many uses of ‘The Battle of Hastings was fought in 1066.’ in virtue of which that utterance type conventionally means what it does. It is a central use. In Gricean terms, (6) is intuitively a case of speaker meaning. But it is not a case of speaker meaning as defined by (SM1) because the pupil does not intend the examiner to come to believe that the battle was fought in that year (for the pupil believes that the teacher already knows this). Cases like (6) forced Grice to revise (SM1). Grice’s revision consisted

Literal Force

123

in requiring only an intention that the audience come to believe that the speaker believes what was said: (SM2) An utterer U means that p by uttering x just if there is an audience A such that U intends A to come to believe that U believes that p (at least partly on the basis of recognizing that intention). But (SM2) faces new counterexamples. Consider: (7) I didn’t do it. (as uttered by a prisoner in an interrogation when he knows that he cannot bring it about that the interrogators believe that he believes what he has asserted.) Again, (7) is intuitively a case of the speaker meaning that he / she didn’t do it. Or in our terms: (7) would seem to be one of the central uses of the utterance type ‘I didn’t do it.’ in which speaker meaning and sentence meaning coincide, i.e., what is meant is that the speaker didn’t do it. But according to (SM2), (7) does not speaker-mean that. One could at this point introduce further revisions of (SM2). However, it is not clear that there will not be further counterexamples – even though counterexamples will be increasingly complex. Further revisions along the same lines will make the account of speaker meaning so complex, that the question of the theoretical status of these Gricean reconstructions becomes more urgent. At the very least, there seems to be no principled way to decide at which level of complexity to stop revising. Grice himself admits that ‘a very much more complicated definition’ (p. 124) would be required to meet all counterexamples. The second problem Grice has to solve is that of specifying the exact connection between speaker meaning and sentence meaning. His ‘first shot’ is to say that a sentence means that p in a speaker’s idiolect just if that speaker has a ‘policy (practice, habit)’ (p. 125) to speaker-mean that p with it. But the ‘just if’ here is problematic.18 For another sentence may have the same meaning or the same sentence may have another, additional meaning (i.e., be ambiguous in a wide sense). Grice therefore introduces the notion of ‘having a procedure in one’s repertoire’ (p. 126). Thus, he proposes the following definition: D2: ‘For U utterance-type X means (has as one of its meanings) ‘*ψ p’’ = df. ‘U has in his repertoire the following procedure: to utter a token of X if U intends (wants) A to believe that U ψ-s that p. (1989, p. 126, with slight alteration for better readability) Grice then goes on also to define a notion of utterance type meaning in a group, as opposed to the idiolect of one speaker. His idea is that utterance

124 Max Kölbel

type meaning in a group is constituted by ‘some (many) members’ of the group having the relevant procedure in their repertoires and retain the procedure in their repertoires on the assumption that at least some other group members do likewise. While I believe that Grice is on the right track with his notion of a group having a procedure in their repertoire (my proposal has some similarities with this idea), I do believe that the elucidation of sentence meaning in terms of language use can be substantially facilitated by operating not with a purely intentional notion of speaker-meaning (and in effect of the various illocutionary acts), but rather with conventional notions. The main gain is that (potentially endless) revisions of (SM1) in the light of cases with unusual communicative intentions are rendered unnecessary. If we seek a definition of C-assertion along the lines proposed in Section 3, all we need to specify is the normative role C-assertions play within the conversational game. We can then say that a sentence s is conventionally assertoric of p (its use counts as a C-assertion that p) in a group, just if that group treats utterances of s as C-assertions, i.e., if the relevant norms are in force in that group. A group treating an utterance as a C-assertion, and the norms being in force in the group, will ultimately be a matter of the actions and intentions of individual speakers. But we need not provide a reductive account of norms in order to draw explanatory benefit from these claims about norms. It is clear that such an approach can no longer be treated as a reduction of sentence meaning to what individual speakers mean by uses of sentences. But it is equally doubtful whether Grice’s final proposal in terms of ‘having a procedure in their repertoires’ can be so treated, or, at any rate, if this attempt would be a successful reduction. A similar story can be told about David Lewis’s attempt to relate sentence meaning to language use (i.e., his attempt to explicate the actual language relation (Lewis 1975). Lewis’s account is a descendant of Grice’s. However, Lewis employs a powerful game-theoretical notion of convention which is explicable in terms of the beliefs and desires of the members of the groups in which the convention prevails. Let me transpose freely. Lewis believes that a group uses a sentence as assertoric of p just if they have a convention (i) to try to use the sentence only if they believe that p and (ii) to come to believe that p upon hearing an utterance of the sentence. Now, on Lewis’s definition of a convention each convention is a regularity that is uniformly conformed to ‘with at most a few exceptions’ (1975, p. 165). This account faces objections that are analogous to the counterexamples to Grice’s account. Of course there are more than a few exceptions to the rule that speakers try to utter a sentence that is assertoric of p only when they believe p. Again, the account can be modified, but it is not clear that this will solve the problem short of major complication.19 A conventional account of C-assertion, C-question and C-command can avoid these complications (at the price of abandoning any reductive project): the sentence is assertoric (interrogative,

Literal Force

125

imperative) of p just if utterances of the sentence are governed by the norms characteristic of assertion that (question whether, command that) p.

7 Jokes, stories, and theatre In addition to considerations to do with indirect speech acts (such as (2)), Davidson (1979 and 1982) has also objected to conventional force indicators on principled grounds. I shall now discuss this objection. Davidson argues that there could not be a sentence-feature that conventionally marks assertoric force. For, he says, ‘every joker, storyteller and actor will immediately take advantage’ and simulate assertion (1979, p. 113). His idea is that jokers, storytellers, and actors do not make assertions and are therefore living counterexamples to the view that there are illocutionary force indicators. At first sight Davidson’s point seems compelling: jokers, storytellers and actors don’t seem to be asserters, yet they use assertoric sentences, and they would use any other conventional force-indicator, if there were any. Some reflection, however, shows that it is not at all clear that jokers do not make genuine assertions. In the fictional case it is more plausible to say that these are instances of assertoric utterances without assertoric force, but still far from mandatory. Let me consider some ways in which a C-account can deal with the various cases. By way of preparation, let us consider an analogous argument – if Davidson’s argument works then this analogous argument should work as well. It’s an argument for the view that there cannot be a procedure that conventionally counts as signing a contract. The argument goes like this: suppose there were such a procedure. Then it would immediately be exploited by jokers, frauds and actors. They would use the procedure even though they aren’t signing a contract. They are just pretending to sign a contract. Since we know that there are procedures that conventionally count as signing a contract, we have no difficulty in seeing what is wrong with this argument. We all know what happens if someone writes his or her name under a contract just for fun or as a joke: if the procedure was correct (this may include that the person signing must be of a sound mind, fully conscious, etc), the joker still counts as having signed the contract and thereby as having undertaken the contractual commitments. It doesn’t help the joker that he only wanted to make a joke. If the relevant people have a sense of humour, they might let the joker off the hook, but that doesn’t detract from the fact that the joker does count as having signed. What about frauds? There could be several kinds of fraud. One kind only pretends to be going through the correct contract-signing procedure, but in fact he or she does not. This kind of fraud represents no counterexample, as the correct procedure was never carried out. Another kind of fraud carries out the correct procedure, but has no intentions of complying with the obligations he or

126

Max Kölbel

she thereby creates. If this second fraud is good at being a fraud, he or she will manage to escape the sanctions for non-compliance. This type of fraud is not a counterexample either, because while the correct procedure was carried out, this did constitute the act of signing the contract with all the normative implications. Finally, let’s consider actors: I believe actors on stage do not usually go through the correct procedure when the fictional characters they represent are signing contracts. Let’s suppose that they really do write their character’s name under a contractual document. This would still not be the correct procedure, because no-one signed their own name. But if we suppose that the actor is signing a contract in his or her own name, then (again if the procedure was correct) he or she may well count as having signed the contract. With this in mind, let us think again about Davidson’s jokers, storytellers and actors who utter assertoric sentences. Begin with the joker: Davidson seems to think that a joker can utter an assertoric sentence, and his utterance will not count as an assertion because it was intended as a joke. But life isn’t quite as easy for a joker. Imagine a very clumsy joker, who utters the words ‘fuck off’, addressing a dangerous-looking bouncer outside a night club. A moment later, he says: ‘only kidding’. Has this joker, or hasn’t he, told the bouncer to f*** off? I believe the only credible thing to say is that he has. Having meant it as a joke doesn’t prevent it from being an insult and from creating the normative conversational facts that insulting someone brings about. The bouncer may have a sense of humour and let him off the hook, but that doesn’t mean that the joker never insulted him. Here is another example, more favourable to Davidson’s idea of a joker: A and B, who are work colleagues, are conversing during their lunch break in a café near their workplace. Their boss, C, is mentioned. C is notoriously stingy and obsessed with his employees getting back on time from their lunch breaks. A utters: ‘You know, C has decided to hand out lunchpacks from now on and to force us to have lunch in the office. The cost will be deducted from our salaries.’ The joke will work particularly well if B takes it seriously for a moment. But should we say that A’s utterance was not an assertion that C has decided to hand out lunch packs etc? I believe that would be counterintuitive. Now, Davidson might insist that it is not an assertion, because it is a joke. But how do we explain that the joke works particularly well if B initially takes it seriously? I believe the best explanation is that A’s utterance was an assertion (at least a C-assertion), and that as a result of a pragmatic process, it does not, ultimately count as a vicious attempt to deceive B, but as a prank or joke. There are other cases that might be described as joke-assertions where it is not essential that the audience take the utterance seriously. Thus, for example, B may answer A in the preceding example by saying: ‘Yeah, and he is going to chain us to the desk and hire a slave-driver with a big whip.’ Cases like these should probably be treated as instances of story-telling.

Literal Force

127

But what should the adherent of conventional assertion say about story-telling? Several accounts are possible. On one view, conventional force-indicators involve a number of normality conditions. Thus, the procedure conventionally sufficient for asserting that p does not only consist in uttering a sentence assertoric of p to an audience, but the utterance must also be made as part of a serious conversation, i.e., not in the course of telling a story or developing some other fiction. On this view, the storyteller’s utterances do indeed fail to be C-assertions, but they also fail to be instances of the procedure performance of which is claimed to be sufficient for C-assertion. Thus story-tellers aren’t counterexamples to the view.20 If this is the best solution then the case of stage-assertion, while it doesn’t prove the impossibility of force indicators, does constrain the sort of features that can count as force-indicators. It would show that moods are like indexical referring expressions: they need contextual completion to do their job of indicating force. This would not show that forces cannot be conventionally encoded – indexical reference is also conventionally encoded (the meaning of ‘you’ together with the context of utterance conventionally determines the reference of utterances of ‘you’). But another view of story-telling is available. On that view, the storyteller’s, novelist’s, or narrator’s utterances in the fictional context do count as C-assertions, but there are extra-linguistic rules that suspend or modify the usual normative consequences of making C-assertions in the context of telling the story. Even though, in my experience, it is very difficult to persuade anyone of this view, it is in my view superior and highlights a further advantage of employing the notion of C-assertion. So let me illustrate the view a little. When narrators (e.g., novelists) make utterances, they use the full range of utterance types that are also used in non-fictional discourse. In particular, they will use assertoric, interrogative and imperative sentences. Apart from some stylistic differences, there are no surface differences between fictional and non-fictional discourse. Moreover, linguistic context has the same role in fiction and non-fiction: the fact that a dagger has been mentioned in previous discourse can make that dagger salient, so that it becomes the referent of an indexical or of an anaphoric pronoun. Ambiguous expressions can be disambiguated using linguistic context both in fiction and in non-fiction. Even more significantly, the difference between literal and non-literal meaning appears to apply to fiction just as it applies to non-fiction: a novelist can use irony, metaphor and implicatures and can convey information by using an interrogative sentence. Similarly, presupposition phenomena in fiction are no different from presupposition phenomena in non-fiction. The novelist can convey that the hero has a sister either by presupposing it or by using an assertoric sentence whose content is that the hero has a sister. Thus, there is a prima facie case that linguistically fictional and non-fictional discourse are uniform.

128 Max Kölbel

Some notions of C-assertion (but not all: consider Williamson’s notion) can help us exploit the uniformity between fiction and non-fiction to the full. Consider a C-account that views the normative role of C-assertion, C-question etc in terms of its effect on conversational score. It will treat the use of assertoric sentences in fiction as cases of C-assertion, the use in fiction of sentences that require a presupposition as C-presupposing, the use of interrogative questions in fiction as C-questions etc. Purely linguistic conversational rules are exactly the same in fiction and non-fiction. What is different in fiction is nothing linguistic. Rather the difference lies in the different non-linguistic aims we pursue in making fictional utterances. In fiction as in non-fiction it is inappropriate to presuppose something controversial, to use an anaphor without antecedent, to use ungrammatical sentences etc. But in fiction it is usually ok to describe events and people that never existed, while this is not ok in most non-fictional contexts. On some people’s view this means that we don’t assert in fiction. In my view this shows that the norms that prohibit describing events and people that never existed cannot be purely linguistic norms, and in particular they cannot be norms of assertion. As I have explained, the controversial account of fictional assertion just outlined is not the only possible response to Davidson’s challenge from storytelling. Thus, acceptance of this account is not essential to the overall aim of the chapter. Readers not prepared to grant that story-tellers make genuine assertions are therefore invited to join those who defend the first of the views outlined. The most complex of Davidson’s three purported counterexamples is that of stage-assertion. A special difficulty here is that the actors performing the utterances are not themselves the originators of the phrases they utter. Rather, they are following a script. Clearly, we cannot treat the actors’ utterances on stage as C-assertions that are subject to the rules of conversation. It would be absurd to say that the actor has violated a linguistic rule when he uttered a sentence that presupposed something not already in the conversation’s background. If anyone has violated that rule, then it is the character portrayed by the actor. The actor himself is only subject to the norms of proper acting. Thus, actors on stage are most plausibly treated as not fulfilling the basic condition of being part of a conversation. That’s why they do not count as having C-asserted: while they do utter assertoric sentences, they do not count as participants in a conversation. In other words, they do not perform the procedure conventionally sufficient for C-asserting. However, performance on stage can and should be treated as a representation of a conversation. The conversation represented does involve C-assertions, inappropriate presuppositions etc. The second of the above-described accounts of story-telling can say even more about stage-performances. In some sense, a theatre-performance is a special case of fiction; a special case of telling a story. However, the way in

Literal Force

129

which the story is told is more elaborate. When in a novel the author writes: ‘And then Smith said: ‘You have betrayed me.’’, then a stage-representation of the same story will consist in the actor playing Smith uttering, at the right moment during the play, ‘You have betrayed me.’ Thus the actors aren’t the ones who are telling the story, or at least they are not doing it alone. Rather, the playwright, in collaboration with the crew of actors, director, stage setters, etc, is C-asserting that such-and-such people acted and spoke in such and such a way. I conclude that neither joking, nor story-telling nor stage-acting constitutes a counterexample to the thesis that there are conventional indicators of assertoric force.

8 Non-sentential assertion? The sorts of C-accounts of assertion I have outlined in Section 3 involve the claim that there are force indicators that conventionally indicate assertion. C-accounts are therefore committed to the view that some type of procedure is sufficient for C-assertion. Some C-theorists might wish to claim in addition that the procedure in question is necessary for C-assertion. However, there is a prima facie case against this view, at least if the procedure is to include utterance of an assertoric sentence. For there appear to be assertions that are effected by uttering strings that do not constitute sentences, let alone assertoric sentences. If, for example, I utter ‘Not yet.’ when my ping pong partner is signalling that she is about to serve and asking ‘Ready?’, I am arguably thereby asserting that I am not yet ready (just as she has asked whether I am ready). I have not, however, uttered an assertoric sentence. Thus uttering an assertoric sentence cannot be necessary for asserting something. The obvious move to make here, if one wants to defend the thesis that utterance of an assertoric sentence is necessary, is to claim that the utterance of the mere fragment ‘not yet’ was elliptical for a full assertoric sentence. This move, however, faces a number of difficulties, pointed out in a series of papers by Stainton (e.g., 1995, 1997, 1998, forthcoming a, forthcoming b). One of the main difficulties is that (so it is claimed) ellipsis cannot occur in a position that is ‘linguistically discourse initial’ coupled with the (apparent) fact that one can effect an assertion with a non-sentence even in such a position. This particular difficulty has been ably addressed by Stanley 2000 and especially Merchant 2004. I have nothing to add to this (but see Stainton forthcoming a and b for his responses). However, in one article (Stainton 1997), Stainton uses a different argument to show that apparently non-sentential utterances cannot be analysed as elliptical. In this article, he also argues that such utterances show Dummett’s conventional view of assertion to be wrong. To my knowledge, this argument has not been answered. So in this section I shall try to provide an answer.

130 Max Kölbel

Stainton’s argument is directed against conventional force-indicators of the stronger kind, i.e., force indicators that are not only sufficient but also necessary for assertion. The precise thesis he attacks (and tentatively ascribes to Dummett 1973) is the thesis (RS): (RS) Convention Based Analysis of Assertion: A speaker S asserts that P iff: a. S utters an assertoric sentence whose sense is P b. The set C of conventionally specified conditions for assertion obtain. (Stainton 1997, p. 59) Stainton argues that uttering an assertoric sentence cannot be necessary for assertion because assertions can be effected without uttering a sentence, namely by uttering non-sentential phrases or words that are not sentences, such as: (8) Very fast. (said as a boat is speeding by) As I have already said, a natural move for the C-theorist would be to claim that (8) is an elliptical utterance. For example, (8) might be elliptical for: (8e) That boat is going very fast. However, according to Stainton (1997), this move is not open to the C-theorist because if (8) were elliptical for (8e), then the audience should be able to respond by uttering: (9) That car is too. (in response to (8)). He maintains that utterance (9) would be ungrammatical, and that this suggests that (8) was not elliptical in the proposed way after all.21 I want to show that Stainton’s case provides no evidence for his conclusion. But it takes a little effort to unravel the knot he has tied. To begin with, I have some doubts about the claim that utterance (9) would be ungrammatical. However, let’s accept for the sake of argument that it is (I shall come back to this). Accepting that (9) is ungrammatical does not yet amount to a concession that (8) is not elliptical. At most it amounts to a concession that (8) is not elliptical for (8e). (8) might be elliptical for a different sentence, one which does not license the response (9). Suppose, for example, that (8) is elliptical for: (8ee) That boat can go very fast. In that case we would expect (9) to be ungrammatical, for a partial VP deletion, such as (9), does need to employ an auxiliary verb that matches with

Literal Force

131

the antecedent. Thus the (supposed) ungrammaticality of (9) would be consistent with, and explained by, (8) being elliptical for the assertoric sentence (8ee). However, there are reasons to believe that (8) is not elliptical for (8ee) either, because if it was, then the response ‘That one can too’ ought to be perfectly felicitous. However, it sounds odd, at the very least. What can we make of this? Let’s re-consider what is wrong with (9). I believe there is a straightforward explanation why (9) at least sounds a little odd. Consider the following two exchanges, each with three possible responses: (A1) (R1a) (R1b) (R1c)

John: That dress looks very elegant. Mary: This skirt does too. Mary: *This skirt is too. Mary: This skirt too.

(A2) (R2a) (R2b) (R2c)

John: That dress is very elegant. Mary: *This skirt does too. Mary: This skirt is too Mary: This skirt too.

(R1b) and (R2a) are ungrammatical, because in each case the auxiliary does not match the antecedent. The other two responses are possible (even though the c-version is the most economical). Now consider another exchange, this time with an elliptical opening: (A3) Very elegant. (said by John addressing Mary parading a dress she is trying on) (R3a) This skirt does too. (said by Mary a minute later, now sporting a skirt) (R3b) This skirt is too. (said by Mary a minute later, now sporting a skirt) (R3c) This skirt too. (said by Mary a minute later, now sporting a skirt) I believe that (R3a) and (R3b) are a little odd, but probably not ungrammatical. (R3c), however, is perfectly natural. The obvious explanation is that since (A3) is elliptical, there is some indeterminacy, or at least uncertainty, about the correct completion of (A3). However, only (R3c) is sufficiently neutral to fit with most plausible completions.22 That’s why (R3a) and (R3b), while probably not ungrammatical, sound odd. Each of them would commit Mary to an unnecessary level of specificity in her interpretation of (A3). Now we can explain Stainton’s case. There is a perfectly natural response to (8) which involves VP deletion, and this is ‘That car too.’ This response is itself elliptical in an indeterminate way, to match the indeterminacy in the

132 Max Kölbel

ellipsis that occurs in (8). Other responses are odd (even though not quite ungrammatical) because they are too specific about the correct completion of (8): (8) (Ra) (Rb) (Rc)

Very fast. (said as a boat is speeding by) That car does too That car is too. That car too.

I believe that this view is more plausible than Stainton’s view that (9), i.e., (Rb), is ungrammatical. However, as I said, even if it was ungrammatical, that wouldn’t show that (8) is not elliptical for an assertoric sentence because it might be elliptical for a sentence to which (9) is an ungrammatical response. Let’s briefly take stock: I have shown that Stainton has provided no evidence that (8) is not elliptical. I now want to consider whether his example provides evidence that (8) is elliptical for an assertoric sentence, i.e., whether his example can be turned against its author. Stainton uses the following grammatical principle: Across human languages, VP Deletion – leaving a verb phrase unpronounced – is grammatically possible only if a licensing sentential syntactic structure is present in prior discourse.’ (1997a, p. 65) Now, (Rc) quite clearly is grammatically possible as a response to (8). (Rc) seems to involve VP deletion. If so, then (8) must be sentential, that’s what Stainton’s principle tells us. The problem with this argument is that it relies on the assumption that (Rc) involves VP deletion. Stainton will (correctly) complain that this assumption begs the question against him, for in his view (8) is genuinely non-sentential. So this argument is inconclusive. (However, his own objection to the C-account, in the form of (RS), has been refuted.) There is also a different way to resist alleged counterexamples that involve nonsentential assertion, one that does not depend on the success of the ellipsis account. Elliptical sentences involve ‘unpronounced’ syntactic elements. Now suppose that (8) and (A3) and similar examples are indeed elliptical for complete assertoric sentences. Then the question arises whether the mood, or other syntactic force indicator, of these sentences is included in the ‘unpronounced’ part of the sentence, or whether it is already explicitly there. In the cases here discussed, mood is arguably explicit. The utterance (8) involves an inflection (intonation) typical of assertion, and this is marked (in my own and in Stainton’s examples) with a full stop. An utterance of the same phrase in the same situation with an interrogative inflection (raising voice towards the end) would be elliptical

Literal Force

133

for a complete interrogative sentence. However, if the force-element is not unpronounced, then a C-theorist could take the view that whether or not the utterances in question are utterances of sentences, they are utterances of assertoric (interrogative, etc.) types. This line of resistance to the nonsentential assertion objection does, however, require a change in Stainton’s target analysis (RS): while (RS) makes utterance of an assertoric sentence necessary for assertion, a modified account requires merely utterance of a type involving an assertoric force indicator. On this view, even nonsentences can exhibit force indicators.

Acknowledgements I would like to thank Manuel García-Carpintero, Rob Stainton, and Peter Pagin for discussion and comments.

Notes 1. One version of such an approach was defended by Dummett (1973) and criticized by Davidson (1979 and 1982) and Stainton (1995). More recently, Alston (2000) has revived this kind of view of assertion. Barker’s (2004) notion of Proto-assertion is another recent, but more distant, descendant. However, the account I advocate differs from Dummett’s in the treatment of fictional assertion, from that of Alston in that I do not oppose simultaneously using an intentional notion of assertion, and finally it differs from Barker’s in that it does not constitute as radical a departure from standard semantics as his. My account can be seen as complementing Segal’s (1990) and García-Carpintero’s (2004) view that force-indicators should be treated within semantics, and García-Carpintero’s (2001 and 2004) ‘Neo-Gricean’ view that languages are systems of conventional rules which we exploit for our communicative purposes. 2. There is a controversy about where exactly to draw the line between the literal and the non-literal, and there are debates about the relative location of the dividing line between what is said and what is otherwise conveyed. However, most participants of these debates accept some division between the literal, i.e., linguistically encoded, meaning of utterance types and the non-literal meaning tokens of that type can have on particular occasions of use. For some recent discussion, see, e.g., Bach (2001), García-Carpintero (2001), Predelli (2005), Recanati (2004), or Wilson and Sperber (2002). 3. Compare Searle (1969, pp. 33–42). Austin (1962) introduced the idea of language as conventional action into the philosophical discussion. Strawson (1964) claims that some speech acts are constitutively rule-governed, reserving for other speech acts a Gricean treatment in terms of communicative intentions. See also Millar (2004) for a similar account of essentially rule-governed activities (practices). 4. When calling N, the action that constitutes C ‘non-conventional’, I mean that N is independent of the conventions essential for C. It is of course possible (and frequent) that N is itself a conventional action dependent on other conventions (e.g., handing over money to a policeman when he stops you counts as an attempt to bribe him – handing over money is itself an action type which constitutively depends on conventions).

134 Max Kölbel 5. Some assertions about McDonalds by London Campaigners Dave Morris and Helen Steel led to the longest trial in English legal history, the ‘McLibel’ case in 1994–6. See www.mcspotlight.org or the recent documentary film for more information. 6. Though the norms and conventions of assertion may conversely play a role in the constitutive norms of libel: to commit libel it may be sufficient to make assertions with a certain type of content. 7. More precisely, Williamson argues that the above-mentioned norm is the only non-derivative norm in the form of a conditional prohibition which governs all assertions and does not govern all instances of any other illocutionary act type. In my view, saying merely that asserting something unknown is wrong in some unspecified sense (2000, p. 140), is not to say much about the normative significance of making assertions. Full understanding of the activity of asserting would require knowledge of a host of further normative facts, no doubt some of them not specific to assertion and some of them not purely linguistic. 8. Pagin forthcoming articulates a recipe for generating counterexamples to any account that makes a certain social normative fact constitutive of asserting that p. The recipe is to construct a complex performative that brings about the same social normative fact, but which, arguably, cannot be used to assert that p. I do not here have the space to do justice to this objection. At this point let me just briefly indicate two possible lines of reply: First, the conventionalist could insist that if Pagin’s complex performative does indeed bring about the same normative fact, then it constitutes an alternative way of making assertions. A second, perhaps better, line of response would make it constitutive of assertion that the relevant normative fact be brought about by a simple force-indicating device specifically designed to bring about such facts. 9. The necessity operator in (A1) reflects the fact that the norms schematically mentioned on the right hand side of the conditional are essential to assertion. There may be a problem about requiring that the constitutive norms of assertion be expressed in this form, for some theorists (e.g., Williamson 2000, GarcíaCarpintero 2004) use an obligation operator to express the norms governing assertion, e.g., ‘One must: assert p only if one knows p’, or equivalently: ‘it is prohibited to assert p while one does not know p’. But it is not obvious that these forms are equivalent to ‘Necessarily, if one asserts p, then if one does not know p one is violating the norms of assertion’. But I believe that this subtlety can be passed over in the current discussion. Thanks to Krister Bykvist for discussion. See Broome (1999) for a general discussion of the differences between wide-scope and narrow-scope norms. 10. The other type of illocutionary act Strawson speaks about includes acts that are not essentially conventional, but which require uptake by the audience for their occurrence. The example he gives is Grice’s meaningNN, which, he believes, captures a pre-theoretical notion of ‘attempting to communicate’. There are no other examples; assertion, question, command are not mentioned. 11. A good example is Grice 1989, p. 123: D1: ‘By (when) uttering x U meant that *ψ p’ = df. ‘(∃A) (U uttered x M-intending [i] that A should think U to ψ that p and [in some cases only, depending on the identification of *ψ p] (ii) that A should, via the fulfilment of [i], himself ψ that p).’ In Grice’s theory, the analogues of asserting that p, asking whether p, and ordering that p are: meaning that p, meaning that ?p and meaning that !p, each of

Literal Force

12. 13.

14. 15. 16.

17. 18.

19. 20. 21.

22.

135

them associated with some ‘corresponding propositional attitude’ (123), such as believing or intending. The modification added in brackets was added by Grice in response to some difficulties. I discuss this modification of the Gricean intentions in Section 6. Lewis (1975, p. 172) has a slightly divergent intentional account of assertion. He proposes in effect that assertoric sentences of a language are the ones which speakers try to use only when their contents are true and whose contents audiences tend to believe (with a few exceptions: p. 165). Segal (1990, section IV) seems to agree: he distinguishes the semantic (literal) force of an utterance from its pragmatic force. See, e.g., Wilson and Sperber (2002). Even Davidson (1986) is forced to admit that we need ‘temporary theories’ as an interpretative starting point, and even Sperber and Wilson (1995) assign a clear role to semantically encoded literal meanings, though they deny that the literal content of (3) as a whole needs to figure in the interpretation process (Wilson and Sperber 2002). Something along these lines is Segal’s argument for his semantics of s-forces, see his 1990, p. 103–4. In fact, Grice’s first shot involves only an ‘if’. But his subsequent discussion is conducted in such a way as to suggest that he means ‘if and only if’ (see 1989, p. 126). See Kölbel (1998) for a detailed exposition of, and attempt to solve, this problem. See also Hawthorne (1990 and 1993) and Lewis (1992). Dummett’s seems to go down this route in his 1973. García-Carpintero endorses a similar account of moods in his 2004. Stainton bases this conclusion on the following claim about grammar: ‘Across human languages, VP Deletion ... is grammatically possible only if a licensing sentential syntactic structure is present in prior discourse.’ (1997a, p. 65, my emphasis). Presumably he also assumes that the presence of a licensing sentential syntactic structure in prior discourse is sufficient for the grammaticality of verb deletion, for otherwise the argument would not even get off the ground. Why not all? One plausible completion would be: ‘In that dress you look very elegant.’. Neither of the three responses would be grammatical as a response to that antecedent. We need something like ‘In this skirt I do too.’

Bibliography Alston, W. (2000) Illocutionary Acts and Sentence-Meaning (Ithaca, N.Y.: Cornell University Press). Austin, J.L. (1962) How To Do Things with Words (Oxford: Clarendon Press). Bach, K. (2001) ‘You Don’t Say’, Synthese, 128, 15–44. Barker, S. (2004) Renewing Meaning: A Speech-Act Theoretic Approach (Oxford: Oxford University Press). Bird, G. (1994) ‘Relevance Theory and Speech Acts’, in S. Tsohatzidis (ed.) Foundations of Speech Act Theory (London: Routledge). Brandom, R. (1983) ‘Assertion’, Nous, 17, 637–50. Broome, J. (1999) ‘Normative Requirements’, Ratio, 12, 398–419. Davidson, D. (1979) ‘Moods and Performancesí, in A. Margalit (ed.) Meaning and Use (Dordrecht: Reidel). Reprinted in his Inquiries into Truth and Interpretation (Oxford: Clarendon Press, 1984).

136 Max Kölbel —— (1982) ‘Communication and Conventioní, in D. Davidson, Inquiries into Truth and Interpretation (Oxford: Clarendon Press, 1984). —— (1986) ‘A Nice Derangement of Epitaphsí, in E. Lepore (ed.) Truth and Interpretation (Oxford: Blackwell). Dummett, M. (1973) Frege: Philosophy of Language (London: Duckworth). —— (1993) ‘Mood, Force, and Convention’, in B. Vermazen and M Hintikka (eds) Essays on Davidson (Oxford: Oxford University Press). Reprinted in his The Seas of Language (Oxford: Oxford University Press, 1992). García-Carpintero, M. (2001) ‘Gricean Reconstructions and the Semantics/Pragmatics Distinction’, Synthese, 128, 93–131. —— (2004) ‘Assertion and the Semantics of Force-Markers’, in C. Bianchi (ed.) The Semantics/Pragmatics Distinction (Stanford: CSLI Publications). Harnish, R.M. (1994) ‘Mood, Meaning and Speech Acts’, in S. Tsohatzidis (ed.) Foundations of Speech Act Theory (London: Routledge). Hawthorne, J. (1990) ‘A Note on “Languages and Language” ’, Australasian Journal of Philosophy, 68, 116–18. —— (1993) ‘Meaning and Evidence: A Reply to Lewis’, Australasian Journal of Philosophy, 71, 206–11. Kölbel, M. (1998) ‘Lewis, Language, Lust and Lies’, Inquiry, 41, 301–15. Laurence, S. (1996) ‘A Chomskian Alternative to Convention-Based Semantics’, Mind, 105, 269–301. Lewis, D. (1975) ‘Languages and Language’, in K. Gunderson (ed.) Minnesota Studies in the Philosophy of Science 7 (Minnesota: University of Minnesota Press). Reprinted in his (1983). Page references to reprinted version. —— (1979) ‘Scorekeeping in a Language Game’, Journal of Philosophical Logic, 8, 339–59. Reprinted in his (1983). Page references to reprinted version. —— (1983) Philosophical Papers (Oxford: Oxford University Press). —— (1992) ‘Meaning Without Use: Reply to Hawthorne’, Australasian Journal of Philosophy, 70, 106–10. Merchant, J. (2004) ‘Fragments and Ellipsis’, Linguistics and Philosophy, 27, 661–738. Millar, A. (2004) Understanding People: Normativity and Rationalizing Explanation (Oxford: Oxford University Press). Pagin, P. (forthcoming) ‘Is Assertion Social?’, Journal of Pragmatics. Predelli, S. (2005) Contexts (Oxford: Oxford University Press). Recanati, F. (2004) Literal Meaning (Cambridge: Cambridge University Press). Rumfitt, I. (1995) ‘Truth Conditions and Communication’, Mind, 104, 827–62. Sadock, J.M. and Zwicky, A. (1985) ‘Speech Act Distinctions in Syntax’, in T. Shopen (ed.) Language Typology and Syntactic Description. Volume 1: Clause Structure (Cambridge: Cambridge University Press). Searle, J. (1969) Speech Acts (Cambridge: Cambridge University Press). Segal, G. (1990) ‘In the Mood for a Semantic Theory’, Proceedings of the Aristotelian Society, 41, 103–18. Sperber, D. and Wilson, D. (1995) Relevance: Communication and Cognition, 2nd edn (Oxford: Blackwell). Stainton, R. (1995) ‘Non-Sentential Assertions and Semantic Ellipsis’, Linguistics and Philosophy, 18, 281–96. —— (1997) ‘What Assertion Is Not’, Philosophical Studies, 85, 57–73. —— (1998) ‘Quantifier Phrases, Meaningfulness “In Isolation”, and Ellipsis’, Linguistics and Philosophy, 21, 311–40. —— (forthcoming a) ‘Neither Fragments nor Ellipsis’, in L. Progovac et al. (eds) The Syntax of Nonsententials (Philadelphia: John Benjamins).

Literal Force

137

—— (forthcoming b) Words and Thoughts. Stalnaker, R. (2002) ‘Common Ground’, Linguistics and Philosophy, 25, 701–21. Stanley, J. (2000) ‘Context and Logical Form’, Linguistics and Philosophy, 23, 391– 434. —— (2002) ‘Making it Articulated’, Mind & Language, 17, 149–68. Strawson, P.F. (1964) ‘Intention and Convention in Speech Acts’, Philosophical Review 73, 439–60. Reprinted in J.R. Searle (ed.) Philosophy of Language (Oxford: Oxford University Press, 1971). Williamson, T. (1996) ‘Knowing and Asserting’, Philosophical Review, 105, 489–523. This article reappears in revised form as chapter 11 of his (2000). —— (2000) Knowledge and Its Limits (Oxford: Oxford University Press). Wilson, D. and Sperber, D. (2002) ‘Truthfulness and Relevance’, Mind, 111, 583–632.

7 A Plea for Understanding Guy Longworth

1

Introduction

The Philosophy of Language seeks ultimately to address two questions: What are the linguistic facts, and in particular the facts that determine what may be said on occasion by the use of language? And how are speakers in a position to exploit those facts? In a slogan, the first question requests a theory of meaning, the second a theory of understanding. As Michael Dummett has long advocated, a theory of meaning must engage properly with a theory of understanding so that the two questions ultimately receive an integrated answer.1 However, historically, the second question has been recessive. My brief in this chapter is to make a mild plea in its favour, through a preliminary exploration of some issues that arise when it is brought to prominence. In the background are some large questions about the proper division of labour between the theory of meaning and the theory of understanding. Should we view linguistic understanding as a form of propositional knowledge of independently discernible meaning facts? Or is meaning more intimately related to understanding than that picture would require, so that facts about meaning are partly absorbed into, and so only accessible through, a theory of understanding? Indeed, is there even a separable question for the theory of meaning to address, or is understanding an achievement that makes no independently specifiable demands upon the facts?2 These are large questions, and I shall not attempt to foreground them here. But it is important to recognize that a satisfactory answer to our opening questions must engage with them. I shall begin in Section 2 by distinguishing some varieties of understanding. Section 3 sketches an argument against the view that propositional knowledge suffices for understanding. The aim is not simply to support that conclusion, but also to suggest that understanding is distinguished from other forms of epistemic standing by its dependence upon a specific form of integration of other types of epistemic standing. Section 4 marks a further distinction, between what I shall call intake and uptake. Section 5 discusses 138

A Plea for Understanding

139

the prospects of two accounts of uptake, and suggests some desiderata on a unified account of intake and uptake. It is here that the large issues mentioned loom closest to the surface. To reiterate, my aim here is not to provide detailed arguments for or against specific positions or desiderata, but rather to outline some central issues and to indicate some topics that I think are worthy of further reflection.

2 Some varieties of understanding Dummett distinguishes two senses of the verb ‘to understand’: ... that in which someone is said to understand a word, phrase or sentence, considered as a type, and that in which he may be said to understand a particular utterance. We may call these the ‘dispositional’ and the ‘occurrent’ senses of ‘to understand’. (Dummett 1993, p. 58) Why is there a need to mark (or to keep track of) this distinction? Dummett offers two sorts of reasons. The first adverts to a distinction putatively amongst the objects of understanding. The second adverts to a distinction amongst modes of understanding. We need an occurrent sense of ‘understand’ for two reasons: indexicality and ambiguity ... . If, for example, I hear someone say ‘There is a sinister smell here’, how much do I need to know about where he is to know what statement he was making or what thought he was expressing, in that sense under which, if true, it is true absolutely? If someone utters an ambiguous sentence, his hearers may understand it in a particular way, whether as he intended or not; we may speak also, not only of how the speaker meant it, but of how he was understanding it. (Dummett 1993, p. 60) This first reason resides with a distinction between what a type of expression means – what the standing meaning of the expression type is – and what is said by the use of that expression – what thought is expressed by the use of the expression on an occasion. The second reason resides with a distinction between possession of a capacity to understand expression types or utterances and proper exercise of that capacity on particular occasions. Dummett focuses upon the latter reason in response to what he takes to be Wittgenstein’s refusal to acknowledge occurrent understanding: ... it is difficult to see how it can be maintained that no occurrent notion of understanding is required: for it is possible to be perplexed by a sentence on first hearing, through a failure to take in its structure, and to attain an understanding of it on reflection. (Dummett 1993, p. 103)3

140

Guy Longworth

When the two types of reason are distinguished, the need for an at least four-way distinction emerges: (i) dispositional understanding of standing meaning; (ii) occurrent understanding of standing meaning; (iii) dispositional understanding of what is said in particular utterances; and (iv) occurrent understanding of what is said. To see the need for (i)–(iv), consider attending to an utterance of the sentence type in (1): (1) He is too intelligent to expect us to beat. A typical response to an utterance of (1) would be blank incomprehension: absence of occurrent understanding even of the standing meaning of the sentence type employed. On reflection, however, one is able to ‘take in its structure, and to attain understanding’. To a good first approximation, the standing meaning of (1) is given in (2): (2) A contextually determined male is too intelligent for one to expect a contextually specified group including the speaker to beat the contextually determined male. Since this feature of (1) is accessible on reflection, it is plausible that one anyway had dispositional understanding of the sentence type. Clearly, occurrent understanding of the sentence type does not suffice for occurrent understanding of the utterance: for that, one needs to grasp what was said in the utterance, in the sense in which what was said determines truthconditions.4 And for that, one would need to be in a position to ascertain (at least) the referents of ‘he’ and ‘us’.5 In at least a thin sense, the fact that one would ordinarily be able to do this supports the appropriateness of attributing dispositional understanding of the utterance – i.e., a capacity to figure out what was said in the utterance. But one might hedge here if the transition from occurrent grasp of meaning to occurrent grasp of what is said required the acquisition of specific cognitive capacities – for instance, if, in order to grasp occurrently what the speaker said, one had to acquire a capacity to think about him through becoming perceptually acquainted with the male demonstrated by the speaker.6 A further distinction can be drawn at this point between the state that is the upshot of one’s coming to understand a particular utterance, and the episodic achievement through which one enters that state. And ‘occurrent understanding’ is not entirely appropriate for use in application to the state rather than its onset, as witness the oddity of combination of ‘occurrent’ with other verbs for states, as in ‘occurrent knowledge’, ‘occurrent belief’, etc.7 However, emphasis on the distinction between states and episodes can be suppressed in the present context. What matters here is the distinction between a mere capacity to understand on occasion and the upshots of

A Plea for Understanding

141

proper exercise of that capacity on an occasion, whatever the specific temporal profile of the latter.8 I propose in the remainder to focus upon occurrent understanding, or grasp, of what is said. And I shall prescind, as far as is practicable, from consideration of constitutive connections amongst the four categories, except those that arise due to occurrent understanding involving the exercise of dispositional understanding. We will thus be attending to phenomena at the interface between understanding language and understanding the users of language whilst attempting to ignore issues that arise on either side of the interface.

3 Understanding and knowledge Understanding what someone has said, in the occurrent sense, ordinarily helps to put one in a position to know that someone has produced a particular utterance and thereby said, with one or another type of force, that such-and-such. For instance, understanding what someone, say Florence, has said by use of (1), where the speaker used ‘he’ to refer to the World Chess Champion for 2008, might help to put one in a position to know that (3) or that (4): (3) Florence produced an utterance of (1) and thereby asserted that Viswanathan is too intelligent for one to expect us to beat him (Viswanathan). (4) Florence produced an utterance of (1) and thereby asked whether Viswanathan is too intelligent for one to expect us to beat him (Viswanathan). The thin use of saying involved here can be understood as a sort of determinable of each of the determinate forces with which an utterance can be produced. It corresponds with what one might know if one knew that Florence had expressed the thought that Viswanathan is disqualified but did not know whether she had asserted that Viswanathan is disqualified, asked whether he is, ordered, or optated that he be. I shall ignore issues arising from our grasp of the forces of utterances and focus on our engagement with facts like that stated in (5), with ‘said’ understood in the thin way. (5) Florence produced an utterance of (1) and thereby said that Viswanathan is too intelligent for one to expect us to beat him (Viswanathan). In what, then, does one’s understanding of Florence’s utterance consist? An immediate hypothesis is that one’s understanding is one’s knowing that (5), or something similar. The immediate hypothesis has the advantage that it

142

Guy Longworth

involves only minimal departure from what would ordinarily be the case were one to understand Florence’s utterance. But it suffers from numerous disadvantages. The hypothesis can be understood in at least three ways. First, it can be understood as embodying the claim that one’s understanding of an utterance is a simple matter of one’s knowing what was said in that utterance and so is neutral with respect to any more specific account of how one knows what was said. Second, it can be understood as embodying the claim that one’s understanding of an utterance is a brutal matter of one’s knowing what was said in that utterance and so requires that there is no more specific account of how one knows what was said. Third, it can be understood as embodying the claim that one’s understanding of an utterance is an unspecified matter of one’s knowing in a particular way what was said. The unspecified understanding of the hypothesis, with its uncomfortable conjunction of acceptance of the possibility of further specification with refusal to supply it, can be rejected immediately in the present context. Reasons for rejecting the simple and brutal understandings are a little less straightforward. To a first approximation, both should be rejected due to their respective failures appropriately to distinguish understanding from other forms or ways of knowing, either through failing to mark understanding off from other forms or ways of knowing, or through marking understanding off in a way that renders its standing mysterious. I shall begin to explain those failings by considering the hypothesis that understanding is a simple matter of knowing what was said. Seeing things can put one in a position to have propositional knowledge about those things. And we think of seeing things as a specific way in which one can be put in a position to know about those things, a way distinct from being put in a position to know through understanding. The point is not (yet) that an account of understanding must underwrite special treatment of understanding; but rather, that the account of understanding must not disrupt special treatment of other cases. Yet the simple understanding of the hypothesis is consistent with cases of knowledge through sensory perception – for instance, seeing that the game has begun or hearing that the clock has stopped – also being cases of understanding. Hence, the simple understanding of the hypothesis fails appropriately to distinguish understanding from other ways we have of being (or coming to be) in a position to know. One response at this point would be to reject the complaint as premised on a failure to exploit all of the resources available to the defender of the simple hypothesis. The hypothesis is, not simply that understanding of an utterance may be any form of propositional knowledge about it, but also, more specifically, that it is a matter of being in a position to have propositional knowledge of what was said in the utterance. And, the response continues, the special nature of the objects of understanding, that they are facts concerning what was said, forces the required distinction amongst our ways

A Plea for Understanding

143

of being put in a position to know. For the involvement of what is not sense perceptible – i.e., content – in what was said means that one cannot in other specific ways – i.e., by seeing, hearing, etc. – come by knowledge concerning what was said. As Tyler Burge puts it: We do not perceive the contents of attitudes that are conveyed to us; we understand them. We perceive and have perceptual beliefs about word occurrences. We may perceive them as having a certain content and subject matter, but the content is understood, not perceived. (Burge 1993, p. 478) There is room for discussion at this point concerning the understanding of the bounds of the (sense-) perceptible required by the envisaged defence of the simple understanding of the hypothesis. In particular, those who wish to employ the defence and who also wish to endorse a view of sense perceptions as bearers of propositional content will have work to do in ensuring a difference in the modes of engagement with content involved in perception and understanding that would make appropriate Burge’s differential attitude. And it is not obvious that making out the required distinction would not require going beyond the simple hypothesis. But a more immediate problem with the envisaged defence of the simple hypothesis is that, rather than evading the need further to specify the nature of understanding, it positively invites further specification. For the distinction with sense perception relies, not upon the impossibility of sensory engagement with content, but rather upon the impossibility of engagement through understanding with the proper objects of sense. And making out that impossibility would seem to require a substantive account of understanding. The brutal understanding of the hypothesis has the resources to distinguish understanding from knowledge got through sense perception. On its brutal understanding, the hypothesis provides a negative specification of the way of coming to know characteristic of understanding, according to which there is no more specific characterisation of the way of coming to know that constitutes understanding. On this view, what distinguishes being in a position to know on the basis of seeing, hearing, etc., from being in a position to know as a matter of understanding is that in the former cases, by contrast with the latter, it is possible further to specify the way in which one knows. By contrast with cases of knowledge got through seeing, hearing, etc., a complete answer to the question, how one knows what was said in that utterance – where the question presupposes that we have an exhaustive answer to all sub-questions pertaining to perceptual sources, for instance the question of how you were in a position to know about that utterance rather than this one – might be: ‘One just does’. Although it is plausible that the brutal hypothesis can underwrite a distinction between understanding and sense-perceptual sources of epistemic

144

Guy Longworth

position, it appears unable to register a distinction between understanding and other sources. At least that is so on the plausible view that with respect to at least certain basic pieces of one’s a priori knowledge, there is no articulate answer to the question of how one knows them.9 On that view of basic a priori knowledge, the brutal hypothesis fails to acknowledge a genuine distinction, by classifying understanding together with any way of having non-perceptual knowledge – or, at least, with all ways of having such knowledge for which no articulate characterisation is available. And the immediate corollary of that failure is apt to appear even more pressing. By classifying understanding together with basic a priori knowledge, it treats understanding as a way of having such knowledge. And that will seem to many to count decisively against the hypothesis. The consequence will seem decisive against the brutal hypothesis to many theorists, though not to all. For one prominent example, Burge is rendered immune to the present charge because he anyway holds that, in at least some cases, our knowledge of what was said is a priori. More carefully, Burge holds that, if we hive off the distinctive contribution of sense-perception in coming to know what was said, by viewing it as a mere trigger to the operations of the intellectual faculty involved in appropriately entertaining the content of what was said, then we can view the source of our knowledge as the proper operation of the intellectual faculty itself. And it is reasonable in that case to group knowledge gained via understanding together with other cases of knowledge got by intellection, and so to view understanding as a source of, or way of having, a priori knowledge.10 I’ve argued elsewhere that Burge’s view of understanding should not be accepted, at least in full generality, and that it should be rejected, in particular, for a range of core cases in which understanding puts one in a position to acquire knowledge from an interlocutor. The basic difficulty it faces is that, in the core cases, understanding is implicated in making available, not only an expressed content, but also its having been expressed through an episode of the production of speech by a particular agent. It is this function of understanding that puts one in a position to know who said what and to know vicariously on the basis of testimony obtained from particular sources. That function, I’ve argued, requires the integrated exercise of intellectual faculties and sense perceptual faculties so that the form of understanding involved does not in any straightforward way give rise to a priori knowledge of what is said.11 However, even if we accept Burge’s account, there is a more immediate difficulty. Although Burge’s account is able to underwrite the possibility of a priori knowledge got through understanding, it is unable – at least in the context of brutalism about basic a priori knowledge – to rule out the possibility that every piece of basic a priori knowledge is got through understanding. It might be thought that, if the account of understanding that I proposed in place of Burge’s is accepted, it might supply resources to distinguish

A Plea for Understanding

145

understanding from other forms or ways of knowing. According to the alternative view, core cases of understanding are the upshot of, or participate in the upshot of, integrated operations of sense-perceptual and intellectual faculties. On the basis of that account, it might be suggested that what is special about understanding is just that it is the upshot of both sense perception and intellection. Although such an account would go beyond the simple identification of understanding with knowing what was said, it would do so by drawing only on specifications anyway implicated in characterizing its component achievements. Second thoughts serve to scotch the suggestion. Although the proposed account is able to distinguish understanding from the upshots of sense perception or intellection, it is unable, without supplementation, to distinguish it from other cases of knowledge got through both. For one example, it appears unable, absent supplementation, to distinguish understanding from ordinary cases of knowing on the basis of sense perception, where this involves the application of concepts – so intellection – to the deliverances of the senses. For another example, it appears unable, without further ado, to distinguish understanding from a posteriori knowledge of necessities, where such knowledge is the upshot of integrated operations – in typical cases, inferentially integrated operations – of intellection and perception. The discussion to this point has been premised on the assumption that basic a priori knowledge is not subject to further specification. Perhaps that assumption is the villain. Let’s suppose, then, that basic a priori knowledge is subject to further specification – perhaps, for example, as knowledge got through intellection or reflection. Obviously, the supposition is subject to its own explanatory demands, in particular the demand for an account of the powers of intellection and reflection. But even supposing those demands discharged, the consequent account of a priori knowledge would be unable to save the brutal hypothesis. Understanding would be not only special, by virtue of its simple distinction from other forms or ways of knowing, but also an oddity, by virtue of the manner of that distinction. It would be the only form or way of knowing not subject to further specification. And the mystery occasioned by that special standing is deepened by the observation that it would be a brutal way of knowing time-bound contingencies, e.g., concerning who said what, and when. By far the most natural view at this point is that no obvious version of the hypothesis that understanding of an utterance is knowledge of what was said in that utterance should be accepted. Rather, if understanding is a form or way of knowing, then it is a specific form or way of knowing. The conclusion to this point is supported by an observation of Christopher Peacocke’s. Peacocke observes that it is possible to know what someone has said in a particular utterance without understanding their utterance. For instance, someone in the know might tell you that, in producing a particular utterance, Florence said that Viswanathan is disqualified. Supposing

146

Guy Longworth

appropriate conditions are met, it is possible to acquire knowledge from such testimony. Hence, one might acquire knowledge that Florence said that Viswanathan is disqualified on the basis of testimony, so independently of one’s understanding Florence’s utterance to that effect.12 A similar result can be achieved by considering the difference between merely coming to know, through being told, that an utterance of (1) says that (2) and coming, perhaps through engagement with (2), simply to understand an utterance of (1). In addition to supplying immediate support to the conclusion that understanding is not simply knowledge of what was said, Peacocke’s observation bolsters an intervening step in the argument. For it indicates that understanding is not distinguished from other forms of knowledge simply by virtue of being knowledge specifically of what was said. And it suggests that if understanding is a form or way of knowing what was said, then it is a peculiarly immediate form or way.

4

Intake and uptake

A natural hypothesis at this point would be that understanding is a specific form of propositional knowledge, or a determinate way of having the determinable, propositional knowledge, akin to other specific forms, or ways of having, propositional knowledge, e.g., seeing that such-and-such and remembering that such-and-such. However, pursuit of the comparison with other specific forms, or ways, of knowing suggests an alternative hypothesis. Consider seeing that the game has started. If we accept that this is a specific form, or way, of knowing that the game has started – in particular, that it is a different form, or way, of knowing from hearing that, or remembering that, the game has started – then it is pressing to say in what its specificity lies. And an obvious answer would be that seeing that the game has started is a matter of knowing, by seeing, that the game has started.13 By parity, then, one would naturally predict that, if understanding is a specific form of knowing, then it is a matter of knowing by understanding. And if one understood the characterisation of the specific forms of knowing involved here as going via appeal to an episodic basis for the acquisition of knowledge – for example, an episode of seeing the game start, or of understanding Florence’s utterance – then it might appear natural to view the understanding of an utterance, not as the output of an epistemic achievement, but rather as its episodic input. The suggested understanding of the structure of specific forms of knowledge is not immediately forced. For one might view the ‘know by φ−ing’ specification as indicating involvement of a specific epistemic capacity, rather than specific input to a general epistemic capacity. For present purposes, we needn’t attempt to decide the issue between the two understandings, for there is reason to think that, whatever its precise role in determining

A Plea for Understanding

147

epistemic standing, exercises of dispositional understanding can leave a mark in consciousness independent of the achievement of epistemic standing. And it would then be natural to view those episodes as cases of understanding. A reason to think that exercises of dispositional understanding can make a psychological difference independent of the achievement of epistemic standing is provided by the possibility of rational withholding of belief about what is said consistent with exercise of understanding otherwise suitable to underwrite knowledge of what is said.14 The structure of the case is similar to an analogous case for seeing. In that case, we begin with a situation in which one sees that such-and-such, for instance a situation in which one sees that a chess piece is black. In that situation, one knows by seeing that the chess piece is black. Since knowing that the chess piece is black entails believing that the chess piece is black – at least modulo the subject’s rationality and their ability to believe that the chess piece is black – one believes that the chess piece is black. But one might be in almost precisely the same position with respect to one’s perceptual and epistemic standing towards the chess piece whilst withholding belief that the chess piece is black. If one has apparently good reasons for withholding belief – perhaps one has, or appears to have, good reason for thinking that one is undergoing a brain manipulation that would make only red chess pieces appear black to one – then one’s withholding belief might be rationally permissible. Hence, one might rationally withhold belief and, so, fail to know that the chess piece is black. Plausibly, one might nonetheless see the chess piece, and the chess piece’s colour. More generally, the upshot of exercise of one’s seeing capacity might have the same potential to determine one’s epistemic standing as such an upshot would have in a healthy doxastic environment. For instance, if the apparent reason for withholding belief were extinguished, then – ceteris paribus – it is plausible that one would be in a position to know that the chess piece is black. In the case of understanding, we can begin with a situation in which one knows on the basis of understanding that Florence has said that the game has started. In this case, it appears plausible that one might in almost precisely the same way undergo an exercise of one’s capacity to understand in a context in which one rationally withholds belief from the proposition that Florence has said that the game has started. One might still take in Florence’s saying that the game has started even if apparently reasonable doubts about brain manipulation prevented one from making epistemic use of what one took in. More generally, one might be in a position such that, if one’s apparent reasons for withholding belief were extinguished, then – ceteris paribus – one would be in a position to know that Florence had said that the game has started.15 I suggested that, given the distinction between the non-epistemic upshot of exercise of a capacity to understand and knowledge attained on the basis

148

Guy Longworth

of that exercise, it would be natural to identify understanding with the upshot rather than the knowledge. But the distinction between upshot and knowledge does not dictate adoption of that position on the location of understanding. Given the distinction, a question arises as to the function of knowledge – and, in particular, the belief requirement on knowledge – given that apparently one can take in elements of one’s environment in its absence. And a plausible answer to that question – an answer given credence by reflection on what is lacking in a subject who mistakenly withholds belief in what they take in – is that knowledge of a fact is what allows one to exploit that fact – to have that fact serve as a reason for one – in one’s theoretical and practical reasoning. On that view, what is missing, when one withholds what would otherwise be reasonable belief about what one takes in, is a capacity to have one’s practical and theoretical position controlled by how things are.16 If that is right, then exercises of a capacity to understand that do not result in knowledge, like exercises of a capacity to see a black chess piece’s colour that do not result in seeing that the chess piece is black, might be thought to involve a form of cognitive blindness.17 And we might then rather identify understanding with a specific form of openness to the facts about what has been said, than with the type of input that determines its specific form. Again, we are not required for present purposes to take a stand on the precise location of understanding. We have seen grounds for requiring, of an account of understanding, that it give accounts both of the episodes by which one takes in what is said – what I shall refer to as an account of intake understanding – and also of the form of openness to what one thereby takes in that enables one to exploit facts about what is said in one’s theoretical and practical reasoning – what I shall refer to as an account of uptake understanding. I shall turn, in the next section, to the question of the form of uptake understanding, in the hope that addressing that question might provide clues as to the further specification of intake.

5 Uptake and knowledge of truth-conditions If openness to what one takes in through understanding is a matter of understanding that such-and-such, in the way that openness to what one takes in through seeing is a matter of seeing that such-and-such, then it is too amorphous a subject matter to warrant focussed attention. For one can see that a chess piece is black without seeing the chess piece’s colour, for instance by seeing the colours of the other chess pieces in a set. And one can understand that Florence said, in a demonstrated utterance, that the game has started without understanding an utterance of Florence’s to the effect that the game has started, for instance by understanding testimony from someone other than Florence. What is required is a restriction to the epistemic positions one can occupy just through exercise of one’s capacity to understand. Put another way, what is wanted is an account of

A Plea for Understanding

149

epistemic standing that would be (with appropriate modal qualification) both necessary and sufficient for uptake. The requirement for an account of epistemic standing that would be both necessary and sufficient for uptake has obvious affinities with a requirement of Donald Davidson’s, according to which a theory of meaning for a particular language should ‘explicitly state something knowledge of which would suffice for interpreting utterances of speakers of the language to which it applies’ (Davidson 1976, p. 171). There are two relevant differences between Davidson’s formulation and ours. First, Davidson seeks an account of facts, or truths, that are exploited by those who understand utterances, rather than an account of the way(s) in which those facts, or truths, are exploited. That provides sufficient grounds for Davidson’s decision not to join us in requiring an account of knowledge that is necessary for openness to what one understands, grounds that Davidson seeks to bolster through general scepticism concerning the probity of the more demanding aim.18 Second, and related, Davidson seeks to provide sufficient conditions for being in a position to interpret (his version of uptake) through a statement of the facts, or truths, knowledge of which would sustain ability to interpret. By contrast, our formulation leaves open whether Davidson’s aim is, in full generality, sustainable, by allowing that understanding might depend upon something other than propositional (i.e., stateable) knowledge.19 Davidson’s proposal for meeting his requirement is that knowledge of the output theorem of an interpretative truth theory that applies to an utterance, together with knowledge that it was a theorem of such a truth theory, would suffice for uptake of that utterance.20 Let’s begin by considering whether Davidson’s proposal can be transposed into an answer to our question about uptake. So understood, it becomes the hypothesis that the knowledge involved in Davidson’s proposal is both necessary and sufficient for uptake. For example, uptake of Florence’s utterance of (6) would be a matter of knowing that (7), as the theorematic element of an interpretative truth theory covering Florence’s utterance, u. (6) Viswanathan will win. (7) u is true iff Viswanathan will win. From the perspective of our requirement, the hypothesis has two main benefits but suffers from at least one critical flaw. The first main benefit is that the hypothesis sustains an attractive requirement that I shall call the transparency of understanding. In discussing the logical status of supposition, Dummett notes that it is not logically an imperative for, he observes, I could, having said, ‘Think of a number’, ask ‘Have you done so yet?’, but it would be a joke if I asked that question having said, ‘Suppose the witness is telling the truth’. (Dummett, 1973, p. 309)21

150 Guy Longworth

As is appropriate, Dummett doesn’t explain the joke. One obvious source of humour is the inappropriate form of the verb. In order for supposition to be something one could do, so something one could be ordered do to, it would need to be able to stand for a specified process or the specified outcome of a process. And that would require the verb ‘suppose’ to take (distinctive22) progressive form, contrary to fact.23 But a second potential source is brought out more clearly by the imperative, ‘Entertain the thought that the witness is telling the truth’.24 The second source is that one can’t understand the order to entertain the thought, or idly to suppose, without complying thereby with its demand. The transparency of understanding is responsible for the latter effect, whereby one cannot in general understand an utterance without entertaining a thought that it is used to express. More carefully, the transparency of understanding is the requirement that one who has uptake of the expression of a particular thought is thereby in a position to treat reasons for or against accepting the thought itself as reasons for or against endorsing the thought registered through understanding. For instance, according to transparency, one who has uptake of the expression of the thought that Viswanathan will win is in a position to treat reasons for denying that Viswanathan will win as reasons for denying what they understand as having been expressed. In that way, what is immediately before one’s mind, by virtue of one’s understanding of an utterance, is the (putative) subject matter determined by the thought, rather than, for example, the thought itself being the immediate object of one’s thinking. The transparency requirement is closely connected with John McDowell’s observation that our most basic engagements with what speakers say are ways of acquiring information about the subject matter of what they say rather than about what the speakers have said or believe.25 It is weaker than the requirement that all thoughts be transparently communicable, so that for any thought of any thinker, it is possible for another thinker to entertain that thought. The latter requirement faces obvious difficulties arising from the special natures of certain indexical thoughts.26 And it may be that similar difficulties afflict the less demanding requirement for transparency of understanding. In that case, or on other grounds, one might consider a further weakening to the requirement that understanding the expression of a thought with subject matter P involves entertaining a thought with subject matter P.27 Further wrinkles might then involve specification of additional conditions on the relation between expression and entertaining. The requirement obviously warrants further articulation and defence, but its provisional endorsement will be harmless in the present, exploratory context. The hypothesis derived from Davidson’s proposal sustains transparency in the following way. We have it that Florence said that Viswanathan will win and so gave expression to the thought that Viswanathan will win. Hence, according to transparency, understanding Florence’s utterance must involve entertaining the thought that Viswanathan will win. The required

A Plea for Understanding

151

result is secured by the hypothesis because one cannot know that (7) without entertaining (7), and so without entertaining a thought of its right hand side, to the effect that Viswanathan will win.28 A first obvious alternative to the present hypothesis is that uptake is a matter of knowing what was said in an utterance – for instance, that uptake of what Florence said in u is a matter of knowing that what Florence said in u was that Viswanathan will win. Knowing that is a matter of knowing which thought Florence expressed rather than a matter of entertaining the thought she expressed. To see the difference, and also an aspect of its importance, consider that it is consistent with knowing that Florence has said that Viswanathan will win that one should fail to think in accord with the truth conditions of what Florence said. For instance, it is consistent with an inability to take reasons for thinking that Viswanathan did not win as reasons for thinking that what one took Florence to have said is false. Indeed, it is consistent with apparently rational refusal to treat what Florence said as having truth-conditions, for one might have apparent grounds for endorsing a view of expressed thoughts according to which they do not.29 Notice that, if the possibility of this type of block on the appreciation of reasons marks off knowledge of what was said from understanding proper, then it would provide an at least partial explanation for Peacocke’s observation that one can know what was said without understanding. The partial explanation would be that mere knowledge, unlike understanding, fails the transparency requirement. Anyway, if we assume the transparency requirement, then the present hypothesis has a key advantage over the first obvious alternative. A second, related, benefit of the hypothesis is that it ensures what I shall call appreciation of the expressive connection between an utterance and the thought one understands it to express. Appreciation of expressive connection is a matter of appreciating that reasons for or against what one understands to have been expressed are reasons for or against the acceptability of the utterance in which it was expressed. Appreciation is sustained by the hypothesis because if one derives knowledge of (7) from knowledge of an interpretative truth-theory covering u, then one is in a position to know that the truth-value of u is coeval with the truth-value of the thought that one entertains on the basis of understanding. One is therefore in a position to derive reasons to take u to be true (/false) from reasons to believe (/deny) that Viswanathan will win and vice versa.30 In this case, there is room for a gap to open up, on the hypothesis, between uptake and its derivational exploitation, for someone might know (7) as a theorem of an interpretative truth-theory and yet have apparent grounds for rejecting the derivability of coeval status for its right and left hand sides, due perhaps to an idiosyncratic understanding of the bi-conditional.31 Following the earlier suggestion about Peacocke’s observation, the gap here might help to explain why testimonial knowledge of truth-conditions appears not to suffice for understanding. But

152 Guy Longworth

perhaps it could be argued that someone with an idiosyncratic conception of the bi-conditional would nonetheless be in a position to appreciate an expressive connection between utterance and condition, despite their not being in a position properly to exploit what they appreciate. On the assumption that transparency is to be respected, a second obvious alternative to the present hypothesis is a view according to which uptake of what is said in an utterance is a matter simply of entertaining an expressed thought, without entertaining the thought on the basis of knowing a truththeorem for the utterance. Such a view would obviously preserve transparency, but at a cost. For someone might entertain the thought expressed in an utterance without, so to speak, appreciating it as having been expressed in the utterance. In particular, one who entertained the thought expressed in an utterance might fail to occupy an epistemic position in which reasons to reject the expressed thought are available to them as reasons to reject the utterance.32 Hence, the hypothesis has a key advantage over the second obvious alternative. Although the hypothesis has important advantages over the two obvious alternatives that we have considered, it also suffers from a critical flaw.33 The flaw can be presented as a dilemma. The first horn of the dilemma arises from the demands that the hypothesis imposes upon uptake. In the first place, the hypothesis secures transparency only by securing more than transparency appears to demand. In order to entertain the content expressed by an utterance, the proposal requires that subjects entertain a richer content, embedding a concept of truth. And it might reasonably be denied that it is a necessary condition on a subject’s having uptake that they have, or are required to exercise in uptake, facility with a concept of truth.34 Moreover, in the second place, the proposal, as presented, is yet more demanding on subjects, since it requires subjects to appreciate that the truth-theorems that they apply to particular utterances are elements in an interpretative truth-theory. Again, it might reasonably be doubted that uptake in general depends upon the sort of reflective assurance provided by knowledge of the interpretative status of a truth-theorem applying to a particular utterance.35 The first horn of the dilemma, then, is the claim that, as presented, the proposal is implausibly demanding of the capacities, and exercises of capacities, involved in uptake. On the second horn of the dilemma is the claim that the features of the hypothesis found problematic on the first horn are not optional. Obviously, the core component of the proposal, that transparency and appreciation of expressive connection involve knowledge of truth-theorems, cannot be shorn of its apparently extraneous element, that it makes uptake depend upon exercise of a concept of truth. And the hypothesis secures appreciation of expressive connection only by requiring, not only that subjects know truth-theorems applying to particular utterances, but also that they know that the theorems are elements in an interpretative truth-theory. Someone

A Plea for Understanding

153

who knew only the bi-conditional theorems, without also knowing their pedigree, would not be in a position to treat reasons for rejecting (/accepting) one side of a theorem as reasons for rejecting (/accepting) the other. For example, someone in that position might view reasons for rejecting either side of the bi-conditional as reasons for rejecting the bi-conditional itself.36 The second horn of the dilemma, then, involves a version of John Foster’s infamous objection to Davidson’s proposal, according to which knowledge of what is in fact an interpretative truth-theorem covering an utterance does not suffice for appreciation of the utterance’s expression of content.37 On the assumption that the hypothesis is too demanding to supply a necessary condition on uptake, the requirements that we have discerned on an adequate account of uptake might appear to be impossible to meet. That is, it might appear impossible to provide an account that is less demanding than the hypothesis, and yet able to ensure both transparency and appreciation of expressive connection. And in that case, one might be inclined either to reject one of the requirements, or to take their conjunction to sustain a sort of transcendental argument to the effect that the demands imposed by the hypothesis are appropriate. However, we signalled at the outset an additional feature of the hypothesis that is not obviously a mandatory component of any account of uptake: the requirement that an account of uptake take the form of an account of the content of propositional knowledge. We considered, and rejected, one alternative hypothesis that involves rejection of the view, the hypothesis that uptake is a matter simply of entertaining an expressed thought. But having made the requirement explicit, it is clear that there are further alternatives to be explored before we accede to one of the hypotheses we have already considered. One important alternative, defended in recent work by Ian Rumfitt, gives up the view in favour of a treatment of uptake as constituting, not propositional knowledge about utterances, but rather, as he puts it, ... a second-order cognitive capacity: [in the case of uptake of assertionlike sayings] one who possesses it is in a capacity to gain new knowledge from old (Rumfitt 2005, p. 444).38 Specifically, My understanding an utterance u as [assertion-like] saying that P puts me in a position (a) to know that P, in the event of my coming to know that u is true; (b) to know that u is true, in the event of my coming to know that P; (c) to know that u is false, in the event of my coming to know that not P; and (d) to know that not P, in the event of my coming to know that u is false.

154

Guy Longworth

Understanding a[n assertion-like] saying, in other words, allows knowledge to spread back and forth between the saying’s content and attributions of truth to it, and between that content’s negation and attributions of falsity to the saying (Rumfitt 2005, p. 443).39 Rumfitt’s proposal has two main advantages over the truth-theoretic proposal. First, Rumfitt’s proposal directly sustains appreciation of expressive connection, rather than running a dogleg through propositional knowledge of interpretative truth theorems. Or, rather, his proposal makes a single requirement on understanding do the work we earlier divided between transparency and appreciation of expressive connection. Second and closely related, the type of appreciation involved in understanding is, on Rumfitt’s account, sufficiently demanding that it apparently removes the need for the sort of reflective assurance of appreciation of expressive connection supplied, on the truth-theoretic proposal, by knowledge of the interpretative status of theorems. Both advantages appear to be consequent upon the transition to a view of understanding as involving second-order capacities together with a very demanding conception of the epistemic standing of those capacities. At this point, a natural question about Rumfitt’s proposal concerns intake. What account of intake would serve the proposal, in particular in underwriting someone’s coming to be in the type of epistemic position characterized through (a)–(d)? Even setting aside the sorts of issues arising from the finitude of our capacities that give rise to a concern with compositionality, it is not plausible that we simply bring to bear on utterances prior knowledge of the sort described in (a)–(d). At best, we might have such knowledge about the sentence types instanced in particular utterances. And that appears to be the model to which Rumfitt wishes to appeal in accounting for the onset of the understanding of particular utterances. Rumfitt describes appreciation of the connections in (a)–(d) as akin to knowledge of derived rules of inference, in this case as capacities derived from capacities with respect to rules governing the contribution of sub-sentential expressions to sentential level rules.40 What is required of intake in such a context is that it should afford knowledge about which expression types are instanced in a particular utterance. In effect, (a)–(d) are to be viewed as derived from a combination of analogues for (a)–(d), or their derivational sources, governing expression types, rather than utterances, together with subjects’ competent exercise of abilities to recognize the instancing of those expression types in particular utterances. It is therefore plausible that the second component – the exercises of abilities to recognize the instancing of sentence types – provides Rumfitt’s account of intake, his account of the onset of understanding of particular utterances. Although Rumfitt does not explicitly endorse the view of intake as recognition of the instancing of sentence types, it is the view apparently most in

A Plea for Understanding

155

accord with the derivational shape of his proposal. And the appearance is sustained by the fact that he provides an explicit account of the recognition of the instancing of sentence types. According to that account, recognition of the instancing of sentence types is a matter of appreciating the intentions with which speakers make some of the noises that they do, where the relevant range of intentions are intentions to be recognized as having produced an utterance instancing one or another specific type of sentence.41 Rumfitt’s proposal deserves more discussion than I can afford it here. Indeed, I think that its core, or something close to that core, can provide an account of uptake that is quite plausible. I shall simply note and set aside one difficulty and then all too briefly press a second. The difficulty that I wish simply to note is that, as stated, the proposal is more or less as demanding of conceptual resources as the truth-theoretic account it is designed to surpass.42 For on the assumption that being in a position to know that an utterance is true or false demands possession of a concept of truth or falsity, understanding will, on Rumfitt’s account, often depend upon possession of those concepts. Moreover, attainment of utterance understanding requires, on Rumfitt’s proposal, the manipulation of rich theoretical machinery – albeit a machinery, not of propositions, but rather of connections amongst propositions. However, what is central to the proposal is that uptake opens a conduit for the transmission of reasons, however those reasons are characterized, and however precisely the conduit is opened. And it is not implausible to suppose that the types of reasons available to a subject will impact upon the type of connections amongst reasons accessible to the subject. If that is right, then it is plausible that more minimal analogues of (a)–(d) might be provided to characterize the understanding of subjects who lack facility with a concept of truth.43 And the concern about the derivational machinery that must be manipulated in order to achieve understanding is readily absorbed into the following difficulty. The difficulty that I wish to press concerns the interaction between Rumfitt’s account of uptake and what appears to be his favoured account of intake. In the first place, notice that neither Rumfitt’s account of intake, nor his account of uptake, sustains transparency. For one might be in the position that Rumfitt describes as understanding and yet have no view at all about any of the reasons that one’s understanding connects. In that case, one might understand u without entertaining a thought that u expresses. As noted, Rumfitt in effect collapses transparency and appreciation of expressive connection into appreciation of (a)–(d). Accordingly, one’s occurrent understanding of an utterance need have no impact upon one’s first-order psychology aside from one’s recognition of the instancing of sentences by utterances. That result anyway appears phenomenologically off-key. But in the present context, it is especially problematic. For despite one’s initial failure to entertain a thought expressed by an utterance, acquiring knowledge

156

Guy Longworth

that the utterance is true immediately puts one in a position to entertain that thought through acquisition of knowledge with the expressed thought as content. It is as if one engages with the thought expressed by an utterance only when, in addition simply to understanding the utterance, one also acquires reason to endorse or to reject it.44 That indicates an oddity of the proposal, though perhaps not yet an objection. An objection comes into view when one reflects further on the embedded account of intake. What is required is an account of what puts one in a position to secure uptake, so what enables one to appreciate the connections specified in (a)–(d). According to the account of intake that fits most comfortably with the derivational shape of Rumfitt’s proposal, what puts one in that position is a combination of (i) one’s recognition that the speaker is trying to direct an instance of a sentence type at one and one’s knowledge of which sentence type the speaker is trying to direct at one45 and (ii) one’s derivation of a sentential analogue of the type of second-order capacity characterized through (a)–(d) appropriate to the intended sentence type. The oddity remarked before turned on the fact that this proposal appears to involve less than is involved in ordinary understanding of utterances. But the proposal also seems to involve more than is involved in ordinary engagement with utterances. For it is not merely off-key, but false, to claim that speakers typically have sufficiently detailed intentions concerning the types of expressions that they use to serve as appropriate input to the type of derivation that Rumfitt envisages. What speakers typically intend is simply to say that such-and-such, and perhaps to be understood in accord with their intentions, without prejudice as to the specific linguistic means by which they achieve those ends. And competent auditors are typically able to understand utterances despite failing to recognize the details either of expression-directed intentions, or the specifics of the expression types with which they are confronted.46 What is, perhaps, the most pressing difficulty for Rumfitt’s proposal arises from the burden imposed on uptake as a consequence of the account of intake as recognition of the instancing of sentence types. The reasonconnections discerned through uptake, between sentence types, or their instancing in utterances, and thoughts, are contingent: the sentence types could have been reason-connected with different thoughts. And it is difficult to see how epistemic standing with respect to such a structure could be acquired except via something akin to induction from cases – in this case, induction over cases in which one was in a position to know both P (/not P) and that u is true (/false). But in that case, it is not clear that being in an epistemic position, so derived, would differ in significant respects from the position of someone who knew, on similar inductive grounds, an appropriate truth theorem. In particular, it is not clear that induction could put one in a position to treat knowledge that P as putting one in a position to know that u is true, rather than as putting one in a position to know that either u is true or there is a counterexample to the inductively based generalization.

A Plea for Understanding

157

Moreover, even if it were possible to achieve the required position on the basis of something akin to induction, it is far from clear that ordinary speakers typically exploit such a means to that end. For ordinary speakers typically lack detailed appreciation of the evidential base required for such induction, and they would typically allow their apparent understanding to trump conclusions drawn from such a base.47 Put another way, if one were in the cognitive position described through (a)–(d), then one would be in a better position than someone who had merely inductive knowledge of a relevant truth-theorem. But it is not clear how one could attain that position on the basis of a combination of the sort of intake that Rumfitt appears to allow and ordinarily ascertained, or ascertainable, facts concerning the contingent correlations amongst the left and right hand sides of (a)–(d). Rumfitt seeks to explain knowledge of the type of reason-connection involved in his proposal by a comparison with knowledge of rules of inference in logic. If I am right, then the comparison limps at a critical juncture. What sustains the special powers of appreciation of reason-connections in logic appears to be a property that those connections have and that the connections involved in Rumfitt’s proposal lack: a priori accessibility.48

6 Conclusion The foregoing constitutes a preliminary exploration of some issues that arise when an attempt is made to develop an account of linguistic understanding. The upshot is not yet an account of understanding, even in sketch form. But I believe that many of the pieces required to develop at least a sketch are now in place. In particular, I believe that an account of uptake close to the core of Rumfitt’s proposal may be correct. What is required is a way of integrating that core with an adequate account of intake. I shan’t attempt to make good on that suggestion now. As I said at the start, my aim here is to make a mild plea for attention to understanding, by advertising some major targets for reflection. And it would be inappropriate, given that aim, to render further attention superfluous.

Acknowledgements Thanks for very helpful discussion to Bill Brewer, Stephen Butterfill, Naomi Eilan, Christoph Hoerl, Hemdat Lerman, Johannes Roessler, Sarah Sawyer, and especially Matthew Soteriou. This chapter is dedicated to the memory of Nadine Sheldon Green.

Notes 1. Dummett rarely frames the claim in the strong form according to which a theory of meaning should be a theory of understanding. His discussions of the claim

158 Guy Longworth

2.

3. 4. 5.

6. 7. 8.

9. 10. 11.

12. 13. 14. 15.

16.

invariably are shaped by the question of proper integration of theories, rather than the identity of their targets, and are highly sensitive to the difficulty of adequately specifying the integration requirement. See, e.g., Dummett (1981 and 1991). For general discussion of the integration requirement, see Smith (1992). See, e.g., Moore (1989). The issues here are close relatives of questions about the integration of knowledge and fact pursued by Peacocke (1999), and questions about their separability pursued by Fine (2005). The question whether Wittgenstein refused to recognize an occurrent sense of ‘understand’ is answered negatively in McDowell (2009). The need for an additional and less demanding notion of what is said is defended in Travis (2006). In addition, one might need to discern, not only the referents of the unspecified subject of ‘expect’ and the anaphoric object of ‘beat’, but also the specific contents of ‘too’, ‘intelligent’, ‘expect’, and ‘beat’. See, e.g., Travis (1997). See Martin (2002) for discussion of related issues. ‘Occasioned’ might be an improvement on ‘occurrent’ in such contexts. Consistently with my understanding of what it takes for something to be a state, I needn’t take issue with Rundle (2001) and Baker and Hacker (2005), who argue that what we have labelled ‘occurrent understanding’ is itself a form of disposition or ability. Although Rundle and Baker and Hacker take themselves to be in dispute with Dummett with regard to this claim about occurrent understanding, it is not clear that Dummett either takes, or needs to take, a stand on the issue in the discussion that they target. And that would be so even on the episodic understanding of ‘occurrent’, for the onset of a disposition or ability is an episode. For a discussion and defence of this view of basic a priori knowledge, see McFetridge (1990). Burge (1993, 1997, 1998, and 1999). Longworth (2008b). One reason for the qualification ‘in any straightforward way’ is that my disagreement with Burge concerns only the status of core cases of human occurrent understanding. And it is consistent with the a posteriori status of occurrent understanding that it can sustain acquisition (or transmission) of a priori knowledge via testimony, so that it is open to me to endorse Burge’s further claim that such acquisition (or transmission) is possible. If it is possible, and if it also possible for beings other than us – e.g., infinite beings – to have a priori knowledge of what is said, then it may be possible for humans to acquire from such beings a priori knowledge of what is said. Peacocke (1976). See also Fricker (2003). See, e.g., Williamson (2000). See Hunter (1998) and Longworth (2008a). It is plausible to view the possibility of rational withholding of belief as marking a line between a kind of receptivity and a kind of spontaneity, where that line marks a boundary around the domain of cognition for which a subject is responsible. Since one has at best limited control over what one takes in through understanding or sense perception, and since the capacities that sub-serve that intake are fallible, allowing intake to encroach on the domain of spontaneity would have the potential to make one responsible for errors without one’s control. For discussion and defence of this view of a function of knowledge in rational psychology, as an enabling condition for the exploitation of factive reasons, see Dancy (2000 and 2008); Hornsby (2008); Hyman (1999 and 2006); McDowell (1982, 1994, 1995); and Williamson (2000).

A Plea for Understanding

159

17. It may be that this condition applies only to creatures with a capacity to have propositional knowledge, or even creatures with a capacity to know of the sort possessed by normal humans. Perhaps, e.g., knowledge would not be required to play this role in creatures with a perception-desire psychology. 18. See, e.g., Davidson (1984, p. 125 and 1986). 19. Indeed, our formulation is consistent with the required epistemic standing being content-less, as it would be, on some views, if it were a form of purely practical knowledge. See especially Moore (1997). For more general discussion of the content of knowledge-how and practical knowledge see Rumfitt (2003); Snowdon (2003); and Stanley and Williamson (2001). 20. Davidson (1984). 21. Thanks to Matthew Soteriou for reminding me about Dummett’s observation and for helpful discussion of it. 22. Some speakers will find acceptable ‘supposing’, ‘believing’, etc., but without according them a construal distinct from ‘supposes’, ‘believes’, etc. 23. The closest relevant process in the case of supposition would be reasoning under the supposition. 24. Here and throughout I use ‘entertaining’ as a generic, or determinable, for all specific, or determinate, ways of engaging with a thought, including thinking, believing, knowing, etc. And I shall assume that entertaining logically complex thoughts entails entertaining their elementary component thoughts, so that, for example, entertaining the thought that if Viswanathan will win, then he will not be disqualified, entails entertaining the thought that Viswanathan will not be disqualified. 25. See McDowell (1980). For related discussion, see Burge (1999); Hornsby (1989); and McDowell (2005). 26. See, e.g., Dummett (1981); Frege (1918/56); Higginbotham (2002); and Peacocke (1981 and 1997). 27. Two other potential grounds: (i) Transparency embodies a controversial stand on the question whether understanding an utterance requires engaging a specific thought expressed by the utterance, rather than a thought with the same reference as the utterance; (ii) It is impossible transparently to understand an utterance involving the expression of expletive concepts, or thick ethical concepts, that one cannot, or will not, think with. 28. Transparency is lost on some formulations of Davidson’s proposal. For instance, one of Davidson’s formulations has it that ‘what somebody needs to know is that some T-theory for L states that ... (and here the dots are to be replaced by a T-theory)’ (Davidson, 1976, p. 174). If that were all somebody knew, then their knowledge would not amount to entertaining the thought expressed by a target utterance. An improved formulation would add the requirement that somebody must know that ... (and here the dots are to be replaced by the appropriate theorem of a T-theory). 29. See, e.g., McFarlane (2003). 30. This is to run together two aspects of the situation that might otherwise be distinguished: (i) appreciation that u is true iff the thought one understands to have been expressed, P, is true and (ii) appreciation, via transparency, that the thought one understands to have been expressed, P, is true iff P. 31. For example, one might have apparently reasonable grounds for rejecting the general validity of modus ponens. See, e.g., McGee (1985), and for related discussion Williamson (2003).

160

Guy Longworth

32. The concern here is akin to that raised earlier about Burge’s account of understanding. A version of the concern is developed in more detail in Longworth (2008b). 33. To reiterate, these are not disadvantages of Davidson’s proposal per se, for that proposal concerns the metaphysics of the expression of content and not our modes of access to the expression of content. 34. A central case for doubt about the necessity even of facility with a concept of truth derives from reflection on the apparent possibility of uptake by small children who plausibly lack facility with such a concept. See Soames (1989) and (2008); and Longworth (2008a). 35. Cf. Higginbotham (1992). 36. The most obvious case would be one in which a subject knew the bi-conditional on the basis of knowing the truth of both sides of the bi-conditional. 37. See Foster (1976). See also Davidson (1976); Higginbotham (1992); Rumfitt (1995); Soames (1989) and (2008); and Wiggins (1992). 38. See also Moore’s proposal that understanding is (in general) ‘knowledge of how to process knowledge’ (Moore, 1997 p. 189). 39. See also Rumfitt (1995 and 2001). 40. Rumfitt (2005, pp. 449–451). 41. Rumfitt (2005, pp. 433–437). 42. Rumfitt is under no illusion about this: (Rumfitt, 2005, p. 445). 43. One suggestion here would be that analogues of (a)–(d) might be framed in terms of a capacity to exploit connections amongst acceptance (/rejection) of P and trust (/distrust) in u. 44. Compare the earlier discussion of the residue of withholding belief in what one takes in through understanding. On Rumfitt’s view, the residue would be exhausted by awareness of the instancing of sentence types in utterances. 45. Rumfitt (2005, p. 435). 46. For instance, it is unlikely that you are now in a position to reconstruct the sentence to which this note is appended, despite (I hope) having understood my use of that sentence. See Burge (1999); and Hornsby (2005). 47. Alternatively, if one views the connections between sentence types and thoughts as non-contingent, the burden will shift to explaining recognition of the instancing of sentence types so construed, that is to the derivation of (a)–(d) on the basis of knowledge of their non-contingent sentential analogues. Notice that the pressure is increased when one removes the simplifying assumption that the reason-connections made available to one through understanding concern utterances, rather than things done by speakers. For it then becomes apparent just how contingent the reason-connections are, and how little evidence that the connections hold is typically available to ordinary auditors. For discussion of issues in this area, see Burge (1999); and McDowell (1994). 48. The trade-off between austerity of intake and extravagance of resources required for uptake bears comparison with a similar situation in the theory of perception. Consider, e.g., the demands imposed on one’s ability to attain propositional knowledge about ordinary objects by a sense-data account of sensory intake.

Bibliography Baker, G.P. and Hacker, P.M.S. (2005) Wittgenstein: Understanding and Meaning, I, 2nd edn (Oxford: Blackwell).

A Plea for Understanding

161

Burge, T. (1993) ‘Content Preservation’, Philosophical Review, 102, 457–88. —— (1997) ‘Interlocution, Perception, and Memory’, Philosophical Studies, 86, 21–47. —— (1998) ‘Computer Proof, Apriori Knowledge, and Other Minds’, Philosophical Perspectives, 12, 1–37. —— (1999) ‘Comprehension and Interpretation’, in Lewis E. Hahn (ed.) The Philosophy of Donald Davidson (Chicago and La Salle, Ill.: Open Court). Dancy, J. (2000) Practical Reality (Oxford: Oxford University Press). —— (2008) ‘On How to Act – Disjunctively’, in Haddock and MacPherson (2008). Davidson, D. (1976) ‘Reply to Foster’, in G. Evans and J. McDowell (eds) Truth and Meaning: Essays in Semantics (Oxford: Clarendon Press). References in the text are to the reprint in Davidson (1984). —— (1984) Inquiries into Truth and Interpretation (Oxford: Clarendon Press). —— (1986) ‘A Nice Derangement of Epitaphs’, in E. Lepore (ed.) Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson (Oxford: Blackwell). Dummett, M. (1973) Frege: Philosophy of Language (London: Duckworth). —— (1981) The Interpretation of Frege’s Philosophy (London: Duckworth). —— (1991) The Logical Basis of Metaphysics (Cambridge, Mass.: Harvard University Press). —— (1993) Origins of Analytical Philosophy (London: Duckworth). Fine, K. (2005) ‘Tense and Reality’, in K. Fine, Modality and Tense: Philosophical Papers (Oxford: Oxford University Press). Foster, J.A. (1976) ‘Meaning and Truth Theory’, in G. Evans and J. McDowell (eds) Truth and Meaning: Essays in Semantics (Oxford: Clarendon Press). Frege, G. (1918/1956) ‘The Thought: A Logical Inquiry’, Mind, 65, 289–311. Fricker, E. (2003) ‘Understanding and Knowledge of What Is Said’, in A. Barber (ed.) Epistemology of Language (Oxford: Oxford University Press). Haddock, A. and MacPherson, F. (eds) (2008) Disjunctivism: Perception, Action, Knowledge (Oxford: Oxford University Press). Higginbotham, J. (1992) ‘Truth and Understanding’, Philosophical Studies, 65, 3–16. —— (2002) ‘Competence with Demonstratives’, Philosophical Perspectives, 16, Language and Mind, 1–16. Hornsby, J. (1989) ‘Semantic Innocence and Psychological Understanding’, Philosophical Perspectives, 3, Philosophy of Mind and Action Theory, 549–74. —— (2005) ‘Speakers’ Knowledge and Practical Knowledge’, Proceedings of the Aristotelian Society, supp. vol. 79, 107–30. —— (2008) ‘A Disjunctive Conception of Acting for Reasons’, in Haddock and MacPherson (2008). Hunter, D. (1998) ‘Belief and Understanding’, Philosophy and Phenomenological Research, 53(3), 559–80. Hyman, J. (1999) ‘How Knowledge Works’, Philosophical Quarterly, 49, 433–51. —— (2006) ‘Knowledge and Evidence’, Mind, 115, 891–916. Longworth, G. (2008a) ‘Linguistic Understanding and Knowledge’, Nous, 42, 50–79. —— (2008b) ‘Comprehending Speech’, Philosophical Perspectives, 22, Philosophy of Language, 297–331. MacFarlane, J. (2003) ‘Future Contingents and Relative Truth’, The Philosophical Quarterly, 53, 321–36. McDowell, J. (1980) ‘Meaning, Communication, and Knowledge’, in Z. van Straaten (ed.) Philosophical Subjects: Essays Presented to P.F. Strawson (Oxford: Oxford University Press). Reprinted in McDowell (1998).

162

Guy Longworth

—— (1982) ‘Criteria, Defeasibility, and Knowledge’, Proceedings of the British Academy, 68, 455–79. Reprinted in McDowell (1998). —— (1994) ‘Knowledge by Hearsay’ in B.K. Matilal and A. Chakrabarti (eds) Knowing from Words: Western and Indian Philosophical Analyses of Understanding and Testimony (Kluwer: Dordrecht). Reprinted in McDowell (1998). —— (1995) ‘Knowledge and the Internal’, Philosophy and Phenomenological Research, 55, 877–93. Reprinted in McDowell (1998). —— (1998) Meaning, Knowledge, and Reality (Cambridge, Mass.: Harvard University Press). —— (2005) ‘Evans’s Frege’, in J.L. Bermúdez (ed.) Thought, Reference, and Experience: Themes from the Philosophy of Gareth Evans (Oxford: Clarendon Press). —— (2009) ‘Are Meaning, Understanding, etc., Definite States’, in J. McDowell, The Engaged Intellect: Philosophical Essays (Cambridge, Mass.: Harvard University Press). McFetridge, I.G. (1990) ‘Explicating ‘X Knows a priori That P’, in J. Haldane and R. Scruton (eds) Logical Necessity and Other Essays (London: Aristotelian Society). McGee, V. (1985) ‘A Counterexample to Modus Ponens’, Journal of Philosophy, 82, 462–71. Martin, M.G.F. (2002) ‘Particular Thoughts & Singular Thought’, in A. O’Hear (ed.) Logic, Thought and Language (Cambridge: Cambridge University Press). Moore, A.W. (1997) Points of View (Oxford: Clarendon Press). Peacocke, C. (1976) ‘Truth Definitions and Actual Languages’, in G. Evans and J. McDowell (eds) Truth and Meaning (Oxford: Clarendon Press). —— (1981) ‘Demonstrative Thought and Psychological Explanation’, Synthese, 49, 187–217. —— (1997) ‘Concepts Without Words’, in R. Heck (ed.) Language, Thought and Logic: Essays in Honour of Michael Dummett (Oxford: Oxford University Press). —— (1999) Being Known (Oxford: Oxford University Press). Rumfitt, I. (1995) ‘Truth-Conditions and Communication’, Mind, 104, 827–62. —— ‘Semantic Theory and Necessary Truth’, Synthese, 126, 283–324. —— (2003) ‘Savoire Faire’, Journal of Philosophy, 100, 158–66. —— (2005) ‘Meaning and Understanding’, in F. Jackson and M. Smith (eds) The Oxford Companion to Contemporary Philosophy (Oxford: Oxford University Press). Rundle, B. (2001) ‘Meaning and Understanding’, in H-J. Glock (ed.) Wittgenstein: A Critical Reader (Oxford: Blackwell). Smith, B.C. (1992) ‘Understanding Language’, Proceedings of the Aristotelian Society, 92, 109–41. Snowdon, P. (2004) ‘Knowing How and Knowing That: A Distinction Reconsidered’, Proceedings of the Aristotelian Society, 104, 1–29. Soames, S. (1989) ‘Semantics and Semantic Competence’, Philosophical Perspectives, 3, 575–96. —— (2008) ‘Truth and Meaning: In Perspective’, Midwest Studies in Philosophy, 32, 1–19. Stanley, J. and Williamson, T. (2001) ‘Knowing How’, Journal of Philosophy, 98, 411–44. Travis, C. (1997) ‘Pragmatics’, in B. Hale and C. Wright (eds) A Companion to the Philosophy of Language (Oxford: Blackwell). —— (2006) ‘Critical Notice of Cappellan and Lepore Insensitive Semantics’, Mind & Language, 21, 39–49.

A Plea for Understanding

163

Wiggins, D. (1992) ‘Meaning, Truth-Conditions, Proposition: Frege’s Doctrine of Sense Retrieved, Resumed and Redeployed in the Light of Certain Recent Criticisms’, Dialectica, 46, 61–90. Williamson, T. (2000) Knowledge and Its Limits (Oxford: Oxford University Press). —— (2003) ‘Understanding and Inference’, Proceedings of the Aristotelian Society, supp. vol. 77, 249–93.

8 From the Expressive to the Derogatory: On the Semantic Role for Non-Truth-Conditional Meaning Stefano Predelli

1

Introduction

Consider the sentences: (1) Alas, Tony will be re-elected. and (2) Hurray, Tony will be re-elected. These sentences share their truth-conditional profile: each of them is true exactly on the condition that Tony will be re-elected, that is, exactly on the condition required for the truth of: (3) Tony will be re-elected. Yet, fairly uncontroversially, (1), (2), and (3) have different meanings, and are appropriately employed in different circumstances. Since these differences are apparently due to the occurrence of the expressions ‘alas’ and ‘hurray’ in (1) and (2), it follows that the contributions offered by the meanings of these expressions may not be reflected at the truth-conditional level. The primary aim of this chapter is that of sketching an apparatus suitable for the analysis of the characteristic aspects of the meaning of expressive prefixes such as ‘alas’ or ‘hurray’. After a general presentation of some of the relevant features of the classic truth-conditional approach to meaning in Section 1, Sections 2 and 3 explain how certain truth-conditionally idle conventional properties of an expression may be rendered in terms of constraints formally definable as restrictions on classes of contexts. Sections four and five extend this explanation to ‘sub-sentential expressive’, and, tentatively, to derogatory terms. 164

From the Expressive to the Derogatory 165

2

Character

Natural language semantics is devoted to the study of the semantically relevant properties of expressions. In the classic approach, semantic values are assigned to the simple expressions within the language, with respect to parameters of an appropriate kind – in particular, in the case of indexical languages, with respect to contexts. Formally, a context c is representable as an n-tuple that supplies the sort of information required by the meaning of the indexicals occurring in the language, such as, for instance, an agent ca (for the interpretation of ‘I’), a time ct (for the interpretation of ‘now’), or a possible world cw (for the interpretation of ‘actually’). The semantic value of a complex expression (with respect to a context c) is then determined systematically, on the basis of its syntactic structure, and of the semantic values (with respect to c) of its component expressions. So, for instance, a simple sentence such as (3), repeated here for readability’s sake: (3) Tony will be re-elected turns out to be true with respect to a context c (truec) iff Tony, the semantic value for ‘Tony’, is re-elected in c’s possible world, at some time posterior to the temporal parameter of c.1 Within this framework, the truth-conditionally relevant aspects of the meaning of an expression may be reflected in terms of what Kaplan calls characters, that is, functions from contexts to semantic values (Kaplan 1977). So, for instance, by virtue of its conventional meaning, the English expression ‘I’ provides a certain contribution to the truth-conditions of sentences in which it occurs, representable as the function which, given any context c, yields the agent of c as that expression’s semantic value. Similarly, the truthconditionally relevant aspects of the meaning of a non-indexical expression such as ‘Tony’ may be represented by the constant character which, for any context c, yields Tony as that name’s semantic value.2 Consider then (1) and (2) again. As in any other case, the details in the process for the semantic evaluation of these sentences depend among other things on decisions pertaining to their syntactic layout. For the purpose of this chapter, however, any plausible choice in this respect will do. I settle for concreteness’ sake on a simple-minded syntactic analysis of ‘alas’ and ‘hurray’ as (non-iterable) sentential operators, thus approaching (1) and (2) on the model of, respectively, ‘alas S’ and ‘hurray S’. Officially, then, let EX be a class of expressive prefixes (such as ‘alas’ or ‘hurray’), and define an expressive sentence as a sentence S of the form ‘ex T’, where ex ∈ EX, and T is a simple sentence, that is, a sentence that does not contain occurrences of expressive prefixes.3 Given an assumption of this kind, the compositional evaluation of cases such as (1) or (2) identifies as the semantic values for ‘alas’ and ‘hurray’

166

Stefano Predelli

functions that yield truth-conditional outcomes on the basis of the semantic value of the embedded sentence, in this case (3). Given the intuition that (1) and (2) are truth-conditionally on a par with (3), the function in question is identity: for any sentence S, ‘hurray S’ and ‘alas S’ are true with respect to c iff S is. It follows that ‘hurray’ and ‘alas’ share their character, namely the non-indexical character which, given any context c, yields the identity function on truth-values. Clearly, ‘alas’ and ‘hurray’ mean different things, and competent speakers do not employ (1), (2), or (3) indifferently. Thus, some aspects of the meaning of these expressions are not reflected in their character, and the peculiar features of their conventional profile must be recorded at a truth-conditionally irrelevant level. The next section presents the general background for the analysis of the non-character encoded aspects of the meaning of expressive prefixes. Section three discusses some consequences having to do with the logical properties of sentences such as (1) and (2).

3

Meaning and use

The semantic framework summarized in Section 1 may be applied to the semantic analysis of episodes of language use, on the basis of hypotheses pertaining to their representations in terms of expression-context pairs. Assume for instance that my use u of (3) on 1st August is representable in terms of (3) and of a context c containing the actual world and 1st August as its parameters: u = < (3), c>. It follows that u is true as long as (3) is evaluated as truec, that is, as long as Tony is actually re-elected some time after 1st August. Note that, although in this picture all instances of language use are representable as expression-context pairs (in particular, given my focus on uses of sentences, as sentence-context pairs), the converse is not the case. For instance, relatively uncontroversially, the pair consisting of (3) and a context c such that nobody ever speaks in c’s possible world is not the representation of any use of (3). In general, then, the class of contexts c such that the sentence-context pair <S, c> is a (representation of) a possible use of S, is a proper subclass of the class C of contexts, henceforth the class CU(S) of contexts of use for S. So, the definition for CU(S) follows the pattern: (4) CU(S) = {c ∈ C: K S(c)} where KS(c) is an appropriate condition on contexts, hereinafter a use constraint for S. The exact nature and composition of use constraints is not immediately relevant for my purpose here. Still, a few (negotiable) examples may be appropriate at this stage. So, for instance, a context is a context of use for any sentence only if its possible world is populated by linguistic

From the Expressive to the Derogatory 167

agents, and only if it contains instances of certain expression types. In this case, then, for any S: (5) c ∈ CU(S) only if, in cw, there exist speakers and expression tokens. Similarly, perhaps, a context c is a context of use for a sentence S only if, in c’s possible world, the agent of c tokens S (and thereby the expressions occurring therein) at c’s time.4 So, for instance, in the case of the simple sentence (3): (6) c ∈ CU(3) only if, in cw, ca tokens ‘Tony’ at ct. Be that as it may, it follows that certain sentences, though not entailed by a sentence S, end up being true at all contexts of use for a sentence S. So, assuming for the example’s sake the correctness of conditions (5) and (6), the sentence: (7) There exist speakers and expression tokens is truec for all c ∈ CU(S) for any S whatsoever, and the sentence: (8) I sometimes token ‘Tony’ is truec for all c ∈ CU(3). I say that (8), though obviously not entailed by (3), is (use) verified by (3). In general, then: (9) An expression e verifies a sentence S iffdef truec(S) for all c ∈ CU(e). The distinction between entailment and use-verification is worth repeating: (3) does not entail (8), because the former is truec as long as Tony ends up with a successful electoral result in cw, regardless of the occurrences of ‘Tony’ or any other token. Still, (3) verifies (8), in the sense that any use of the former inevitably takes place within a context with respect to which the latter is true.5 Indeed, unlike the notion of entailment (and validity), the idea of use verification is importantly grounded on non semantic regularities: the reason why (8) is verified by (3) has to do not only with the semantic profile of these sentences, but also with certain presumably general properties of language use. Still, the study of sentences such as (1) and (2) raises the further possibility that at least some use constraints be associated with special restrictions imposed by the meaning of certain expressions, rather than with purely extra-semantic properties, such as the fact that sentences need to be tokened in order to be used. In particular, the notion that (1) and (2) are truth-conditionally on a par with (3) does not entail that a use of any of these sentences is appropriate on any circumstance for the correct use of any of the other. So, for instance, (1) may appropriately be employed by a speaker who is unhappy with Tony’s re-election, that is, by a speaker for

168

Stefano Predelli

whom a use of (2) would not be suitable: utterances of (1) by someone enthusiastic at the prospect of Tony’s political failure are apparently somewhat ‘defective’ (in a sense of the term that does not need to be analyzed further for the purpose at hand). What this suggests is an approach to cases such as (1) and (2) grounded on a system of contextual constraints parallel to that employed in (4), the definition of CU(S). However, while (4) appealed to a use constraint KS(c) designed so as to reflect certain extra-semantic properties of language use, the contextual restriction now at issue must involve a constraint on c directly derived from the conventional meaning of expressive prefixes. We can thus define a class NU(1) of contexts of expressively non-defective uses for an expressive sentence such as (1), so that, at least as a first approximation: (10) NU(1) = {c∈CU(1) : at c w and c t, ca is unfavourably disposed towards Pc} where Pc is the content expressed by ‘Tony will be re-elected’ with respect to c. In plain English: (1) is non-defectively employed only by speakers unsympathetic to the prospect of Tony’s forthcoming political success. The class of expressively non-defective contexts for the (truth-conditionally equivalent) sentence (2) is presumably the distinct class: (11) NU(2) = {c ∈ CU(2) : at cw and ct, ca is favourably disposed towards Pc}, so that (2) may non-defectively be employed only in contexts whose agent approves of Tony’s victory. More generally, then: (12) (a) for any simple sentence S, NU(S) = CU(S) (b) for any expressive sentence S of the form ‘ex T’, NU(S) = {c ∈ CU(S) : Rex(T, c)} where Rex is a relation appropriate for ex, such as a relation between the agent of c and the content of T in c. As noted before, (10) and (11) result from properties of the conventional meaning of ‘alas’ and ‘hurray’: for instance, only linguistically incompetent speakers employ (2) (at least non-ironically) in the absence of a positive attitude towards Tony’s future victory. Thus, the difference in the clauses involved in (10) and (11) reflects the intuitive difference in meaning between ‘hurray’ and ‘alas’, undetectable at the truth-conditional level. It follows that the meaning of expressive prefixes may not be represented merely in terms of truth-conditional contributions, that is, in terms of character, but also in terms of expressive constraints, representable as restrictions on the class C of contexts, of the kind indicated in (12b). For instance, the meaning of ‘alas’ may be represented by means of the pair , where id is the identity

From the Expressive to the Derogatory 169

function on truth-values, and Ralas is the expressive constraint such that, for all sentences S and contexts c, Ralas(S, c) iff at cw and ct, ca is unfavourably disposed towards the content of S in c. Similarly, the meaning of ‘hurray’ may be represented by means of the distinct pair , where Rhurray(S, c) iff at cw and ct, ca is favourably disposed towards that content.6 The definition in (12) determines the class of expressively non defective uses of a sentence S by appealing both to the composition of CU(S) and to the effects of (non truth-conditional aspects of) the meaning of expressive prefixes. The exact nature of CU(S), as indicated before, depends on extra-semantic regularities having to do with the necessary conditions for the employment of a certain sentence. But the class of contexts c such that Rex(S, c) is meaning determined: though semantically irrelevant in a sense of semantics concerned with truth-conditions, this class is semantically relevant in a sense of semantics as the discipline devoted to the study of meaning. It is thus convenient to focus on this latter class, that is, on the class ND of non-defective contexts for a sentence S: (13) (a) for any simple sentence S, ND(S) = C (b) for any expressive sentence S of the form ‘ex T’, ND(S) = {c ∈ C: Rex(T, c)}. So, for instance, c is a non-defective context for (1) (or, as I also write, (1) is non-defective with respect to c) iff its speaker dislikes the prospect of Tony’s victory, regardless of the satisfaction in c of the conditions for the use of (1) – for instance, regardless of the existence of tokens of ‘Tony’ in cw. If the language under study is sufficiently well equipped, the expressive constraint for S may be presented in terms of the truth-conditions of certain sentences. In this case, let us say that: (14) A simple sentence W is an expressive witness for a sentence S iffdef truec(W) iff c ∈ ND(S). Informally, and focusing on expressive sentences: for any sentence S of the form ‘ex T’, W is an expressive witness for S iff W is true exactly on the condition established by Rex. So, for instance (roughly): (15) I am now unfavourably disposed towards Tony’s future electoral victory is an expressive witness for (1), since (15) is truec with respect to exactly all non-defective contexts for (1). Note incidentally that, at least on any non-trivial understanding of Rex, all and only simple sentences (i.e., sentences that do not contain expressive prefixes) have logical truths as expressive witnesses. Only simple sentences, since, from the assumption that, for any sentence S of the form

170 Stefano Predelli

‘ex T’, ND(S) ⊂ C, it follows that truec(P∨¬P) for some c ∉ ND(S). All simple sentences, since for any simple sentence S, ND(S) = C. (This result could be encoded ‘syntactically’: what I call a ‘simple sentence’ S could be represented by means of a ‘dummy expressive’ ∅, as in ‘∅ T’, where {c∈C : R ∅ (T,c)} = C).

4

The ‘logic’ of expressives

According to any semantic treatment that denies truth-functional relevance to expressive prefixes, both: (16) Alas, Tony will not be re-elected. Thus, Tony will not be re-elected and (17) Tony will not be re-elected. Thus, alas, Tony will not be re-elected are valid inferences, in the traditional sense that their conclusions are truec for any context c such that their premises are truec. Clearly, however, an appropriate semantic treatment ought to be compatible with the intuition that, in a certain sense, the step in (16) seems more naturally justifiable than that in (17). Considerations in terms of non-defective uses of expressions, rather than in terms of the truth-conditions of the expressions themselves, seem to be appropriate in this respect. For instance, although the premises and conclusions in (16) and (17) are semantically equivalent, the conclusion of (16) may appropriately be uttered whenever its premise may, but there exist contexts in which the premise of (17), but not its conclusion, may non-defectively be employed. This suggests a notion of ‘expressive validity’, defined in terms not only of truth-preservation, but, in a sense, also of ‘non defectiveness preservation’: for instance, all contexts c ∈ ND(‘alas Tony will not be re-elected’) are contexts such that c ∈ ND(‘Tony will not be re-elected’), but not the other way around. Since constraints on defectiveness are meaning-encoded, in the sense defined previously, this notion reflects the intuitive idea of ‘meaning guaranteed’ steps. Let us then say that: (18) A context c is a true non-defective context for S, c ∈ TND(S), iffdef Truec(S) and c∈ND(S). So, for instance, c is a member of TND(1) iff (i) in cw Tony is re-elected at some time after ct, and (ii) in cw and at ct, ca is unfavourably disposed towards Tony’s future re-election. Similarly, c is a member of TND(2) iff, again, Tony’s future success takes place in cw, but ca is well disposed towards that event. Clearly, the class of true non-defective contexts for any simple sentence S simply is the class c such that truec(S): thus, c is a member of TND(3) iff Tony eventually wins at cw, with no regard for ca’s

From the Expressive to the Derogatory 171

attitudes. Then: (19) An argument S1 ... Sm : Sn is expressively valid iffdef TND(S1) ∩ ... ∩ TND(Sm) ⊆ TND(Sn). A few general results may be appropriate at this stage, accompanied by suitable instances of arguments involving expressive sentences. Note first that: (20) A context c is a true non defective context for S iff truec(S & W), where W is an expressive witness for S. (Proof. ‘Only If’ direction: since c is a true non defective context for S, it follows from definition (18) that truec(S) and c ∈ ND(S); and from the fact that c ∈ ND(S) it follows from the definition of witness in (14) that truec(W); thus, truec(S & W). ‘If’ direction: if truec(S & W) then truec(S) and truec(W); and from the fact that truec(W) and the definition of witness it follows that c ∈ ND(S); thus, both truec(S) and c ∈ ND(S), so that c is a true non defective context for S). It is a consequence of (20) that: (21) An argument S1 ... Sm :. Sn is expressively valid iff the argument (S1 & W1) ... (Sm & Wm) :. (Sn & Wn) is valid where Wi is an expressive witness for Si. (Proof. S1 ... Sm :. Sn is expressively valid iff for all c such that c e TND(S1) and ... c e TND(Sm), it is also the case that c e TND(Sn). By (20), this holds iff for all c such that truec(S1 & W1) and ... and truec(Sm & Wm), it is also the case that truec(Sn & Wn)). It may also be worth pointing out that: (22) Any valid argument S1 ... Sm :. Sn whose conclusion Sn is simple is expressively valid. (Proof: given validity, Sn is truec for all c in which S1 ... Sm are true, and hence a fortiori for all c in which S1 ... Sm are true and non defective. As long as Sn is simple, it follows that any such c will also be non-defective for Sn). So, both: (23) Alas S, thus S and (24) Hurray S, thus S are expressively valid. Similarly expressively valid are ‘expressive introduction’ steps, as in: (25) Tony will be re-elected and I am unfavourably disposed towards Tony’s re-election, thus alas Tony will be re-elected.

172

Stefano Predelli

and more generally in: (26) S and W, thus ex(S) where W is an expressive witness for ex(S). (Proof: suppose by reductio that (26) is not expressively valid, i.e., that for some c, c ∈ TND(S) ∩ TND(W) but c ∉ TND(ex S). Then, either (i) falsec(ex S) or (ii) c ∉ ND(S). But if (i), then by the truth-conditional irrelevance of expressive prefixes, falsec(S), against the supposition that c ∈ TND(S). And if (ii), c ∉ TND(S), again in contradiction with that supposition). Note that expressive validity is not a form of ‘validity plus’: some expressively valid arguments are not valid. So, for instance, the argument: (27) Alas Tony will be re-elected, thus I am negatively disposed towards Tony’s re-election is logically invalid: the premise is truec with respect to exactly those contexts in which Tony will be re-elected, and in some of these contexts ca may well lack any attitude towards Tony’s success. But it is expressively valid: for any c ∈ TND(alas Tony will be re-elected), it must be the case that ca is negatively disposed towards Tony’s re-election in cw, and hence that the conclusion is truec. Moreover, since the class of non-defective contexts for the simple conclusion is the class of all contexts, for any such c, c ∈ ND(I am negatively disposed towards Tony’s re-election), so that c ∈ TND(I am negatively disposed towards Tony’s re-election). In general: (28) An invalid argument S1 ... Sm :. Sn is expressively valid only if, for any context c with respect to which the premises are true but the conclusion is not, c ∉ ND(Si) for some Si (1 ≤ i ≤ m). (Proof: assume that S1 ... Sm :. Sn is invalid but expressively valid, and that c is a context such that for all premises Si, truec(Si) but falsec(Sn). Suppose by reductio that for all premises Si, c ∈ ND(Si). Then, c ∈ TND(S1) ∩ ... TND(Sm) but c ∉ TND(Sn), against the assumption of expressive validity.)7 In the foregoing paragraphs, I discussed certain meaning-governed relationships between expressive sentences and other sentences, such as the relationships between ‘Alas S’, S, and W, where W is an appropriate witness. In the final paragraphs of this section, I rest satisfied with a few brief informal comments pertaining to some other features of the ‘logic’ of expressives. To begin with, although I explicitly eschewed iterations of expressive prefixes, expressive sentences may of course be incorporated within syntactically more complex sentences, according to customary syntactic regularities. So, for instance, if for any sentences S and T, ‘S and T’ is a sentence, it follows that, say: (29) George will run but, alas, Tony will be re-elected

From the Expressive to the Derogatory 173

is a sentence. In complex sentences such as this, the target for the relevant speaker’s attitude is apparently determined by syntactic scope. For instance, on any plausible analysis of the compositional procedure yielding the identification of ND(S) for a conjunctive sentence such as (29), c ∈ ND(29) only if ca is unfavourably disposed towards the content in c of ‘Tony will be reelected’, the sentence within its scope. By the same token, the conditions of non-defectivity for: (30) Alas, if George runs, Tony will lose have to do with the speaker’s dissatisfaction with Tony’s reliance on George’s absence, but those of: (31) If George runs, alas Tony will lose must deal with the speaker’s negative attitude towards Tony’s future defeat. Still, any acceptable theory ought also to recognize that the attitude in question is, in a manner of speaking, never subservient to the semantic contributions offered by other devices. For instance, there simply is no non-defective context for (31) whose speaker is actually indifferent at the prospect of Tony’s defeat, but would negatively be disposed towards that eventuality if George were to run.8 It is partly as a consequence of this evidence that the non-defectivity conditions for complex sentences are, in a sense, ‘intersective’: given a sentence S involving occurrences of S1 ... Sn, ND(S) ⊆ ND(S1) ∩ ... ND(Sn). (Equivalently, the witnesses for S are ‘conjunctive’, in the sense that W is a witness for S iff W is equivalent to a conjunction (W1 & ... & Wn), where Wi is a witness for Si). So, the class of non-defective contexts for any of the following: (32) George will run, but alas Tony will be re-elected (33) If George will run, alas Tony will be re-elected (34) Alas Tony will be re-elected, or George will run is the class C ∩ ND(alas, Tony will be re-elected), i.e., ND(1), and the class of non-defective contexts for any of the following: (35) Hurray George will run, but alas Tony will be re-elected (36) If hurray George will run, then alas Tony will be re-elected (37) Alas Tony will be re-elected, or hurray George will run is the class ND(hurray George will run) ∩ ND(alas Tony will be re-elected), that is, the class of contexts c such that, at cw and ct, ca is favourably disposed towards George’s participation, but unfavourably disposed towards Tony’s success. Another issue that deserves to be mentioned before I conclude this section has to do with the ideas of ‘content’ and of ‘having an attitude towards

174

Stefano Predelli

a content’, involved in the constraints for ‘alas’ and ‘hurray’. The reason why I could afford to remain noncommittal in this respect has to do with the fact that the logical relations I studied thus far did not involve, in a sense, logical moves ‘within’ the scope of an expressive. Yet, the analysis of the relationships between sentences of the form ‘alas S1’ and ‘alas S2’ is inevitably dependent upon alternative approaches to those notions. In particular, on what is sometimes called a ‘hyper-intensional’ view of ‘having an attitude’, ‘alas S1’ and ‘alas S2’ may well be associated with distinct nondefectivity classes even if S1 and S2 are logically equivalent (that is, the sense of ‘content’ in question must be sufficiently fine-grained to distinguish between the content of S1 in c and the content of S2 in c, even when S1 and S2 inevitably share their truth-value). So, in this view, c may be a non-defective context for, say: (38) Alas Tony will be re-elected but not for (39) Alas (Tony will be re-elected and T) for some sufficiently complex logical truth T, because ca fails to draw the relevant inference: ca is positively disposed towards Tony’s re-election but has no attitude towards the more complex content expressed by ‘Tony will be re-elected and T’. The detailed discussion of the relationships between hyperintensionality and expressivity goes beyond the limits of this chapter. I rest satisfied, before concluding this section, with a tentative stab against the hyper-intensional understanding of content summarized in the foregoing paragraph. For instance: (40) Alas (Tony will be re-elected), but hurray (Tony will be re-elected and T) seems inevitably defective (where T is, as before, logically true), notwithstanding the possibility that, in some context c, a sufficiently logically confused agent ca may be negatively disposed towards ‘Tony will be re-elected’, but react with elation at the prospect that ‘Tony will be re-elected and T’ be true. If this much is on the right track, then, what may be at issue in the definition of the use-constraints for ‘alas’ or ‘hurray’ may be relations to contents individuated along less demanding lines than in the hyperintensionalist proposal, so that, for instance: (41) Alas S1, thus alas S2 ought to be evaluated as non-defective (and obviously valid) whenever S1 and S2 are equivalent. By the same token, of course, the non-defectivity classes for the conjuncts in (40) ought to turn out to be disjoint, regardless of the logical acumen of the agent in this or that context.

From the Expressive to the Derogatory 175

5 Non-sentential expressives Presumably, an approach similar to that introduced thus far may be adapted to ‘non sentential expressives’, such as ‘damn’ in: (42) That damn Tony will be re-elected or (43) The damn Prime Minister will be re-elected. So, for instance, (42) is apparently truth-conditionally equivalent with: (44) Tony will be re-elected but, unlike (44), it is non-defective only with respect to contexts whose agent is unfavourably disposed towards Tony. In general, then, at least as a first approximation, for any sentence S containing occurrences of ‘that/the damn n’, for some appropriate singular term n, c ∈ ND(S) iff ca is unfavourably disposed towards i, where i is the semantic value of n in c.9 This approach generates ‘logical’ patterns parallel to those unveiled for expressive prefixes. So, the arguments: (45) That damn Tony will be re-elected, thus Tony will be re-elected (46) Tony will be re-elected and I am unfavourably disposed towards Tony, thus that damn Tony will be re-elected (47) That damn Tony will be re-elected, thus I am unfavourably disposed towards Tony are all expressively valid. These considerations may be extended to cases involving ‘general terms’, as in: (48) The damn communists will be re-elected. This sentence entails: (49) The communists will be re-elected and is expressively entailed by: (50) The communists will be re-elected and I am unfavourably disposed towards communists. There are, of course, many ways to convey one’s contempt for someone without the aid of expressives, for instance by means of adjectives whose semantic value is related to undesirable traits. So, for instance: (51) The Prime Minister is stupid is straightforwardly truth-conditionally committed to the Prime Minister’s lack of intelligence. But adjectives of this sort may be used expressively, as in

176

Stefano Predelli

a certain reading for the sentence: (52) The stupid Prime Minister will be re-elected. Indeed, (52) is ambiguous between an expressive and a non-expressive reading. The latter, presumably prominent in a scenario involving a multitude of Prime Ministers and the query ‘which Prime Minister will be reelected?’, straightforwardly entails ‘the Prime Minister who is stupid will be re-elected’, with a restrictive reading of the relative clause. The expressive reading, on the other hand, bears no truth-conditional relation to the Prime Minister’s intelligence or lack thereof: in this reading, (52) is equivalent with ‘the Prime Minister will be re-elected’, but is non-defective only with respect to contexts whose agent deems the Prime Minister to be stupid. Somewhat parallel considerations hold for: (53) That stupid Prime Minister will be re-elected. (Contrast the obvious, expressive reading of (53) with a reading appropriate for utterance accompanied by a demonstration, and intended as a reply to ‘which stupid Prime Minister will be re-elected?’) Interestingly, in some languages, this contrast is partially syntactically detectable. So, for instance, in Italian, the sentence: (54) Quello stupido di un capo di governo sará rieletto [that stupid of a Prime Minister will be re-elected] only allows for an expressive reading which lacks any entailment pertaining to the Prime Minister’s stupidity.10 Once again, parallel considerations are appropriate for examples involving general terms. So: (55) The stupid communists will be re-elected allows for a reading according to which truth is obtainable as long as a certain subclass of communists will be re-elected (as in scenarios that allow for felicitous stress on ‘stupid’) and: (56) Those stupid communists will be re-elected permits a non-expressive, demonstrative reading. More naturally obtainable are however expressive readings of these sentences, namely those that entail the re-election of the communists, and that are accompanied by a non-defectivity requirement to the effect that the speaker regards communists (or at least the group of communists of conversational relevance) as less than intelligent. It is a natural consequence of these considerations that, on their expressive readings, (55) and (56) involve, in a sense, non-restrictive occurrences of ‘stupid’. So, although the somewhat uncommon non-expressive uses of these sentences may be glossed by a restrictive relative clause, as in: (57) The communists who are stupid will be re-elected,

From the Expressive to the Derogatory 177

their more apparent expressive reading corresponds to something along the lines of: (58) The communists, who are stupid, will be re-elected (of course, assuming a truth-conditionally irrelevant contribution of the parenthetical). This point is spectacularly brought to light in English by the fortuitous syntactic distribution of ‘damn’ (as opposed to ‘damned’): ‘the damn communists’ does obviously not denote a subclass of the communists, and ‘the damn Prime Minister’, unlike certain readings of ‘the stupid Prime Minister’, is unquestionably committed to the existence of a unique Prime Minister. It is instructive (for reasons I explain in the next section) to note that these considerations are compatible with the possibility of true utterances of sentences such as: (59) Not all communists are damn communists (60) Vladimir is a communist, not a damn communist (61) Institutions that treat communists as damn communists are biased. So, for instance, (59) may presumably be employed as a reaction to someone’s characterization of communists as ‘damn communists’, and (60) seems appropriate in the mouth of someone reluctant to characterize Vladimir by means of ‘damn communist’. The reason for the unproblematic nature of these scenarios lies in their evident metalinguistic flavour. So, to cite parallel, uncontroversially metalinguistic instances, one may well write: (62) These are bright colours, not bright colors in order to correct an American’s unsuccessful attempt at British spelling, and one may well utter: (63) Institutions that treat rabbits as bunnies are run by childish treehuggers in order to convey one’s views of certain institutions, even on the (fairly uncontroversial) assumption that ‘colour’ and ‘color’, and ‘rabbit’ and ‘bunny’ are co-extensive.

6 From the expressive to the derogatory Examples such as (42) display a neat division of truth-conditional and ‘expressive’ labour: the character of ‘Tony’ determines the truth-conditional contribution of ‘that damn Tony’, whose non-defective contour is identified by the constraint associated with ‘that damn’.11 Other expressions, however, may well take on both tasks, in the sense that their meanings may involve

178 Stefano Predelli

both non-trivial characters and non-trivial non-defectiveness constraints. The choice of the examples appropriate in this respect is however an interesting empirical business, which involves decisions pertaining to the truthconditional relevance of the ‘emotive connotation’ in question. So, for instance, ‘nag’ may turn out to be equivalent with ‘old and feeble horse’, in the sense that its extension is a proper subclass of the class of horses, without being associated with any significant constraint on non-defective use. Or else, as perhaps Frege suggested, it may turn out to be semantically co-extensional with ‘horse’, but enriched with a constraint that fails to be appropriate for ‘horse’, for instance one having to do with the speaker’s negative opinion of the animal in question.12 The decision between these alternatives is empirical in the sense that it is grounded on brute facts about the conventional profile of that expression – in a sense, on the fact whether authoritative dictionaries define ‘nag’ as ‘an old and feeble horse’ or as ‘derog. a horse’. Throughout this chapter, in consonance with my interest with ‘alas’ and ‘that damn’, I focused on non-defectivity constraints related to the speaker’s attitudes: for instance, (1) is non-defective only with respect to contexts whose agent has certain dispositions towards Tony’s re-election. But, at least in principle, other, ‘non subjective’ types of constraints may be appropriate for other expressions, ‘nag’ being perhaps a case in point. Thus, the empirical analysis of the English meaning for ‘nag’ may well need to adjudicate between three options: the non-expressive treatment, according to which nags form a proper subclass of horses, and two expressive approaches, a subjective analysis, according to which the appropriate non-defectiveness constraint involves the speaker’s attitudes (as in the witness ‘I regard x as old and feeble’), or an objective hypothesis that selects something along the lines of ‘x is old and feeble’ as the relevant witness. In this sense, a use of: (64) That is a nag directed towards a healthy and vibrant horse would turn out to be false, though trivially non-defective, according to the first view; as true but defective in the mouth of certain speakers in the second view; and as true but inevitably defective in the third view. By the same token, of course, alternative verdicts of validity and/or expressive validity would be obtainable with respect to arguments such as: (65) That is a nag, thus that is old and feeble or (66) That is a nag, thus I regard that as being old and feeble. Incidentally, similar alternatives may in principle be available with respect to the semantics of certain morphological suffixes in languages other than

From the Expressive to the Derogatory 179

English, such as the pejorative suffix ‘-accio’ in Italian. So, given a pejorative suffix ‘pej’, the difference between a ‘merely extensional’ approach and alternative expressive treatments may be succinctly summarized in terms of the following three proposals. Let N be a noun of the form ‘n + pej’, extc(e) the extension of e with respect to c, and SN a sentence containing occurrences of N (and, for simplicity’s sake, no other device affecting conclusions of nondefectivity). Then: (67) (i) extc(N) ⊂ extc(n), and ND(SN ) = C. (ii) extc(N) = extc(n), and c ∈ ND(SN ) only if ca regards x as neg (iii) extc(N) = extc(n), and c ∈ ND(SN ) only if in cw x is neg. where neg abbreviates some appropriate negative connotation. So, if in the presence of a well groomed and friendly dog I say: (68) Quello é un cagnaccio [That is a dog + pej.] Option (i) deems that what I said is non-defective but false, option (iii) that it is true but defective, and option (ii) that it is true, and that its non-defectiveness hinges on my attitudes. Even as a native speaker of the language, I am unable to make up my mind with respect to the empirically correct choice here. Similar options may be applicable to derogatory slurs, such as racial slurs. Still, an extensional account in the spirit of (67i) strikes me as considerably less plausible in this case. In general, the use of a derogatory slur does not merely attribute undesirable features to a certain individual who happens to be a member of an ethnic group, but involves a certain negative connotation of that individual as a member of that group. In this sense, to cite Williamson’s example (Williamson forthcoming, p. 22), the use of: (69) Himmler was a Boche seems objectionable, notwithstanding the fact that Himmler did indeed belong to the class of Germans endowed with disagreeable traits. By the same token, the xenophobic contour of ‘Boche’ insistently filters through embedding devices, in a manner parallel to that detected for ‘alas’-sentences such as (31). For instance: (70) If Fritz is a Boche, then he plays the piano does not impart a merely conditional prejudice against the Germans: ‘the very use of the word generates the xenophobic implicature, irrespective of its position in the sentence’ (Williamson forthcoming, p. 22). If derogatory terms are indeed merely expressively derogatory, it is natural (though not inevitable) to suppose that they are truth-conditionally equivalent with their non-derogatory counterparts, and that their difference in meaning is exclusively encoded at the level of non-defectivity constraints.13 If ‘subjective’ constraints somewhat similar to those I proposed for ‘alas’

180

Stefano Predelli

are appropriate in this case, verdicts of non-defectivity would be derivable on the basis of the speaker’s attitudes towards certain individuals, presumably the (typical) members of a certain ethnic or national group. So, for instance, a derogatory general term d could be associated with a constraint (very roughly) along the lines of: (71) In cw and at ct, ca regards every typical member of extc(d) as neg,14 so that a sentence such as: (72) Some Boches play the piano could be employed non-defectively (and, given the musical talent of some Germans, truly) by anyone ill disposed towards the Germans. Alternatively, ‘objective’ constraints may be at issue, as in (again, roughly): (73) In cw and at ct, every typical member of extc(d) is neg, so that, perhaps in consonance with what is sometimes called a ‘silentist’ attitude towards derogatory terms, although there may well be true uses of (72), there exist no non-defective uses of that sentence. My sympathies fall within the subjectivist camp – at least on the assumption of the framework from the foregoing paragraph, according to which ‘objective’ constraints along the lines of (73) entail that (racial or xenophobic) slurs may never be employed non-defectively. Note in particular that, notwithstanding the confused and unpleasant attitude apparently conveyed by uses of slurs of that sort, these uses may hardly qualify as linguistically defective. For one thing, racist and xenophobic attitudes are empirically incorrect: there is no conceptual (and, more importantly, no meaning-encoded) difficulty in supposing that membership in an ethnic or national group provides satisfactory motivation for a hostile attitude. For another, race or national identity (and, for that matter, sexual orientation or religious conviction) are categories arguably deprived of linguistic significance: if derogatory overtones are linguistically appropriate for some general terms (and they surely are),15 it seems reasonable, though admittedly not inevitable, to suppose that they may be extended to predicates pertaining to race or nationality with no violation of the rules of English – though, obviously, in overt violation of the standards of common sense and decency.16 Leaving aside my tentative sympathy for an approach to slurs as in (71), I conclude this section with some considerations having to do with presumed evidence favouring a truth-conditional treatment of these expressions, as opposed to (subjectively or objectively oriented) analyses in terms of nondefectivity constraints. In a recent essay, C. Hom (forthcoming) claims that

From the Expressive to the Derogatory 181

a sentence such as: (74) He is Chinese, but he is not a Chink is appropriate in the mouth of someone with no hostile attitude towards the Chinese – indeed, in the mouth of someone who is objecting to discriminatory discourse. If this is correct, it may initially appear that the non-defectivity constraints I suggested thus far are on the wrong track (assuming the obvious parallel between ‘Chink’ and ‘Boche’). In particular, extending mutatis mutandis the foregoing treatment of (32)-(37) to ‘Chink’ entails that c ∈ ND(72) iff ca is ill-disposed towards the Chinese in cw (or, in the alternative considered before, iff typical Chinese display unfavourable traits in cw), so that (74) may never be non-defectively uttered by any unprejudiced individual (or never be non-defectively uttered toutcourt). In Hom’s essay, (74) is flanked by a variety of other examples, which I will assess in the next paragraphs. The particular case of (74) is however pedagogically instructive, because it is not only the least persuasive among Hom’s exemplars, but also hints towards the general guidelines for an independently motivated analysis, consistent with the approach to racial slurs I have defended thus far. Indeed, an expression I find much better suited in the mouth of non-xenophobic English speakers is the sentence resulting by omitting the contrastive conjunction in (74), as in: (75) He is a Chinese, not a Chink. The comparison of (74) and (75) is instructive, because these sentences conform to a pattern of ‘but’ distribution which, at least more often than not, significantly interacts with instances of ‘metalinguistic negation’ (Horn 1989, p. 405). So, to cite an unrelated instance: (76) This is a colour, but this is not a color has a tendency to sound as an inconsistent sentence (accompanied by a non-uniform choice of spelling), but: (77) This is a colour, not a color is a textbook example of metalinguistic negation.17 Although these considerations are far from conclusive, they do provide an initial, pedagogically instructive introduction to Hom’s counter-examples.18 Indeed, other cases interestingly similar to (74) are unequivocally deemed as instances of metalinguistic negation by diagnostics more solid than those relying on ‘but’-distribution, such as considerations having to do with morphologically ‘incorporated’ negation (Horn 1989, pp. 392–6) and with the

182

Stefano Predelli

distribution of negative polarity items (Horn 1989, pp. 396–402). So: (78) He did not nigger up his voice, he spoke in a dialect19 may well remain indifferent to the non-defectivity constraints associated with the racial slur, but is uncontroversially metalinguistic, as witnessed by its resistance to incorporate negation morphologically, as in: (79) * He unniggered up his voice, he spoke in a dialect. Similarly: (80) There were not some wops there, there were some Italians there may be used non-defectively (with an appropriate phonetic contour for ‘some wops’) in the absence of anti-Italian sentiments, but is unquestionably metalinguistic, as witnessed by the acceptability of ‘some’ in the first conjunct, as opposed to ‘there weren’t any wops there’.20 Not all of Hom’s examples are immediately amenable to the usual diagnostics for metalinguistic negation, but (even when negation is not present at all) they arguably display metalinguistic overtones, particularly apparent if considered side by side with the cases to which I alluded towards the end of Section 4. So, a presumably non-racist spoken utterance of: (81) Chinese people are not chinks is naturally accompanied by the kind of intonation appropriate for any non-contradictory use of: (82) /Tomahtoes/ are not /tomaetos/, for instance in the mouth of a British speaker with little tolerance for American pronunciation. Similarly, if: (83) Institutions that treat Chinese as Chinks are morally depraved is at all appropriate as an objection to racist practices, it is arguably so only in the sense in which (63), repeated (and renumbered) further on for readability’s sake, conveys an intelligible message, notwithstanding the relatively uncontroversial extensional equivalence of ‘rabbit’ and ‘bunny’: (84) institutions that treat rabbits as bunnies are run by childish treehuggers. Admittedly, none of these considerations prove that Hom’s sentences, and any other similar example one may come up with, must be read metalinguistically whenever uttered with no xenophobic or racist intent. They do however indicate that, in the absence of further arguments, instances such as (75) or (83) do not provide obvious, conclusive evidence in favour of the truth-conditional relevance of the derogatory element associated with racist and xenophobic slurs.

From the Expressive to the Derogatory 183

Notes 1. More precisely, in the classic double-index approach to indexical languages, [[e]]c,w is the semantic value of e with respect to a context c and a circumstance w, and a sentence S is truec iff [[S]]c,k = T, where k is the circumstance determined by c (Kaplan, 1977). 2. The controversial idea that proper names are associated with a constant character is assumed here merely for the sake of example. 3. ‘Simple’ is thus non-synonymous with ‘atomic’, in the usual sense of the term. For instance, ‘Tony will be re-elected and George won’t’ is simple but non-atomic. 4. Instances of recorded messages may provide counter-examples to this suggestion; see Sidelle (1991), Corazza et al. (2002), and Predelli (2005). 5. This is harmlessly sloppy. More pedantically: any use of (3) takes place under conditions appropriately representable by means of an n-tuple (context) c such that truec (8). 6. According to David Kaplan, as summarized in Potts (2005), ‘it may be that the primary problem in semantics is not what does this or that mean, but rather in what form should we attempt to say what this or that means’ (Potts 2005, p. 179). Although the approach I propose here may well be quite different from what Kaplan had in mind, it is no coincidence that I began being interested in the semantics of expressives after attending one of his presentations on ‘ouch’ and ‘oops’. 7. The ‘if’ direction of the conditional, on the other hand, does not hold; for instance, the argument: (*) Hurray, Tony will be re-elected. Thus, hurray, I am favourably disposed towards Tony’s re-election is invalid, and any context in which the premise is true but the conclusion is false must be a context c in which Tony succeeds, and I am not favourably disposed towards that eventuality. Yet, although any such c is such that c ∉ ND(hurray, Tony will be re-elected), (*) is not expressively valid, as confirmed by the possibility of a context c such that, in c, Tony will be re-elected and I am favourably disposed towards his success, but I am not favourably disposed towards my attitude in this respect 8. By the same token: (**) Tomorrow alas Tony will be re-elected is non-defectively used whenever the speaker presently dislikes the prospect of Tony’s re-election: for no reading of (**) is it the case that c ∈ ND(**) and ca is not negatively disposed towards Tony’s victory at ct (even if ca will be so disposed on the day after that containing ct). 9. It is doubtful that the object of the speaker’s hostility is inevitably the semantic value of the expression in its proximity. So, for instance, although a reading of this sort seems appropriate for (42), it may well be the case that what the speaker of ‘where did you put the blasted keys?’ is unfavourably related to are not the keys, but rather certain salient events in which that object is involved (Potts, 2005). Further complications may be raised by cases in which those expressions do not seem to carry the negative attitude, as in ‘this is a damn good cake’. 10. The sentence ‘quello stupido capo di governo sará rieletto’, on the other hand, preserves the ambiguity mentioned in the text. 11. ‘The semantic/pragmatic interpretation system treats expressive meanings as if they were on a different tier. Once computed, expressive meanings are barred

184

12.

13.

14.

15.

16.

17.

18.

Stefano Predelli from any further semantic composition of whatever kind’ (Kratzer 2004, p. 130). I remain neutral with respect to the correct interpretation of Frege’s views on the matter, though I tend to subscribe to Williamson’s conclusion that ‘Frege’s category of tone is too miscellaneous to take us very far in the analysis [of derogatory terms]’ (Williamson forthcoming, p. 19). For a dissenting voice, see Richard (2008). For Richard, ‘we cannot ascribe truth to utterances’ involving racial slurs (Richard 2008, p. 13): although ‘German’ and ‘Boche’ ‘classify’ the same individuals, it is a mistake to think ‘that classification is sufficient for truth or falsity’ (Richard 2008, p. 24). ‘Typical’ since, as Williamson puts it, ‘a xenophobe may easily say “He is a Boche, but he’s not cruel – he’s one of the few decent ones” ’. (69) may arguably need further strengthening: users of ‘Boche’ do not merely regard all actual Germans as inferior, but presumably consider (typical) Germans as intrinsically endowed of undesirable features. It is a lamentable historical accident that bigots and racists have taken the lead in the production of ‘lexically encoded’ slurs. Still, exceptions abound: ‘pigs’ may well have been an eminently usable derogatory term for typical police officers in the sixties (or at least for individuals belonging to certain repressive institutions). Moreover, the statistical prominence of racial, sexist, and xenophobic simple English slurs is easily compensated by the ‘expressive compositionality’ partially discussed: ‘damn fascist’ and ‘fucking racist’ are as subjectively expressive and linguistically non-defective as xenophobic slurs, but presumably eminently usable by at least some readers of this chapter. Williamson addresses a somewhat parallel issue, but his framework (grounded on the content of conventional implicatures) is different from mine – in particular, none of my considerations about linguistic non-defectivity are of relevance with respect to his views on the matter. He writes: ‘When someone uses the word “Boche” one can legitimately ask him to withdraw the anti-German implication; but if it is clear, as it may well be, that he does believe that Germans are cruel, then it is hardly legitimate to ask him to withdraw the implication that he has that belief’ (Williamson forthcoming, pp. 22–3). I tentatively disagree with this conclusion as well: I may well remain in a position to ask someone to withdraw a true implicature, just as I may legitimately invite a xenophobic conversant not to convey (in straightforwardly truth-conditional terms) her displeasure towards certain races or nationalities. Perhaps a similar point is to be found in Hornsby (2001): ‘Certainly there are occurrences of derogatory words that are utterly inoffensive: “He is not a nigger” can be said in order to reject the derogatory “nigger” ... [These] are examples in which it is part of the speaker’s message that she has no use for the word “nigger” ’ (Hornsby 2001, p. 129). Note that, according to Horn the form Y but not X can only be read along the lines of ‘descriptive negation’ (Horn 1989, p. 405). If Horn is right about this strong conclusion, the denial of an extensionalist approach demands the flat out rejection of Hom’s intuition that (74) may be used non-xenophobically. Although I am not sure that such a rejection is obviously out of place, I am also inclined to give a less decisive role to considerations of ‘but’ distribution. Be that as it may, it seems clear that, in a sense, (74) needs ‘all the help it can get’ in order to be even prima facie amenable to a non-xenophobic contour: stress and intonation play an important role in this respect. For further discussion of

From the Expressive to the Derogatory 185 metalinguistic negation, see for instance Burton-Roberts (1989 and 1990) and Carston (1996 and 1998). 19. The expression ‘nigger up one’s voice’ is cited in the Oxford English Dictionary, and is attributed to a 2000 entry in the San Francisco Bay Guardian. 20. Indeed, instances involving racial slurs occur among Horn’s initial prototypical examples of metalinguistic negation, as in ‘I beg your pardon, Lee isn’t an uppity Wop, he’s a strong vibrant Italian’ (Horn 1989, p. 372).

Bibliography Burton-Roberts, N. (1989) ‘On Horn’s Dilemma: Presupposition and Negation’, Journal of Linguistics, 25, 95–125. —— (1990) ‘Trivalence, Gapped Bivalence, and Ambiguity of Negation: A Reply to Seuren’, Journal of Linguistics, 26, 455–70. Carston, R. (1996) ‘Metalinguistic Negation and Echoic Use’, Journal of Pragmatics, 25, 309–30. —— (1998) ‘Negation, ‘Presupposition’ and the Semantics/Pragmatics Distinction’, Journal of Linguistics, 34, 309–50. Corazza, E., Fish, W., and Gorvett, J. (2002) ‘Who Is I?’, Philosophical Studies, 107, 1–21. Hom, C. (2008) ‘The Semantics of Racial Epithets’, The Journal of Philosophy, 105, 416–40. Horn, L. (1989) A Natural History of Negation (Chicago: University of Chicago Press). Hornsby, J. (2001) ‘Meaning and Uselessness: How to Think about Derogatory Words’, Midwest Studies in Philosophy, 25, 128–41. Kaplan, D. (1977) Demonstratives (ms.). Reprinted in J. Almog, J. Perry, and H. Wettstein (eds) Themes from Kaplan (Oxford: Oxford University Press, 1989). Kratzer, A. (2004) ‘Interpreting Focus: Presupposed or Expressive Meaning?’, Theoretical Linguistics, 30, 123–36. Potts, C. (2005) The Logic of Conventional Implicature (Oxford: Oxford University Press). Predelli, S. (2005) Contexts: Meaning, Truth and the Use of Language (Oxford: Oxford University Press). Richard, M. (2008) When Truth Gives Out (Oxford: Oxford University Press). Sidelle, A. (1991) ‘The Answering Machine Paradox’, The Canadian Journal of Philosophy, 21, 525–39. Williamson, T. (forthcoming). ‘Reference, Inference and the Semantics of Pejoratives’, in J. Almog and P. Leonardi (eds) Festschrift for David Kaplan (Oxford: Oxford University Press). Available at http://www.philosophy.ox.ac.uk/__data/assets/pdf_file/0011/1325/Reference.pdf/.

9 Analyticity in Externalist Languages Gillian Russell

1

Introduction

I am going to argue for a different way of thinking about the analytic/synthetic distinction (ASD). The view that I will present is complicated in some respects and it makes reference to unfamiliar entities, and so my strategy in arguing for it will be that of Bertrand Russell in ‘On Denoting’: I will present a number of puzzles for the traditional account of the ASD, and argue that my account can solve them all. I will not have much to say here about Quine’s objections to analyticity, though I have had much to say about them elsewhere. (Russell 2008) Rather, I will be presenting a series of new puzzles which rely on phenomena that have risen to prominence in the decades since the Quine-Carnap debate – phenomena such as direct reference, semantic externalism and indexicality. There is some awareness in the literature that these phenomena cause problems for the already beleaguered distinction (Putnam 1962; Kripke 1985/80a; Salmon 1993; Boghossian 1996; Williamson 2007), but if my central claim here is correct, then the ASD – properly understood – is compatible with all the new developments.

2

Analytic sentences

The ASD is sometimes introduced with the following story: there are two factors that go into making a sentence true, the way the world is, and what the sentence means. The sentence ‘grass is green’ for example, is true in part because it means what it does, and in part because grass has the colour it has. Sentences like this are synthetic. Analytic sentences, on the other hand, are those for which the meaning alone – independently of the input of the world – is sufficient to make them true. Commonly presented examples include the truths of arithmetic and logic, as well as so-called conceptual truths, such as ‘all bachelors are unmarried’ and ‘triangles have three sides’. 186

Analyticity in Externalist Languages

187

For many, the basic idea of an analytic sentence – though not the details or defence of any particular philosophical account – is deeply intuitive. When introduced to it, students often have the feeling of being reminded of something which at some level they already knew. The Socratic explanation for this might be that they were familiar with the ASD when they lived among the forms, but another explanation – more plausible to my way of thinking – is that the existence of analytic sentences is a natural consequence of an intuitive folk theory of meaning. The folk theory goes like this: linguistic expressions have meanings, and meanings play three roles. (1) they are what speakers have to know about an expression in order to count as understanding it. For example, my brother counts as understanding ‘alacrity’ if he knows what it means. (2) an expression’s meaning determines – in conjunction with the way the world is – the extension of that expression, such as its referent in the case of a singular term, or the set of objects which satisfies it in the case of a monadic predicate. (3) the meaning of an expression is what the expression contributes to what is said by a sentence containing it. The English sentence ‘muscle fibres are elastic’ says that (more fancily put: ‘expresses the proposition that’) muscle fibres are elastic, but if the word ‘elastic’ had had the meaning that the word ‘rigid’ has, the sentence would have said something different, namely, that muscle fibres are rigid. Thus the meaning of the sentence-part contributes to the meaning of the sentence. I call this sketch a folk theory, but in calling it such I mean neither to suggest that the folk would explain it in such technical language, nor that the conception of meaning it encapsulates is limited to the folk. Many of us begin our theorizing about language with common sense, and the gist of the picture is apparent in the work of philosophers as diverse as Immanuel Kant, Gottlob Frege, Rudolf Carnap, Jerrold Katz, David Lewis, and Frank Jackson. With this picture in the background, the traditional account of analyticity follows easily. Suppose we decide to add a new expression to our language, say ‘mimsy’. In order to introduce a new expression – as opposed to merely drawing attention to a string of letters or sounds – one has to give it a meaning. One way to do this is by providing a synonym, i.e., an expression which already has the meaning which we wish to give to ‘mimsy’. Let’s use ‘cold and slimy’ and introduce ‘mimsy’ by means of the following definition: something is mimsy if and only if it is cold and slimy. Assuming for the moment that everything has gone smoothly, ‘mimsy’ now means what ‘cold and slimy’ means. Given the folk theory, this has three important consequences. First, according to (2), meaning, together with the way the world is, determines extension. So, given that ‘mimsy’ and ‘cold and slimy’ mean the same thing, it must be the case that they have the same extension. This has the consequence that ‘every mimsy thing is cold and slimy’ is a true sentence.

188

Gillian Russell

It seems as if the truth of the sentence is a consequence of defining ‘mimsy’ to have the same meaning as ‘cold and slimy’ and hence it is not unnatural to call the sentence true in virtue of meaning. The folk theorist can draw even stronger conclusions. Given (2), if ‘mimsy’ and ‘cold and slimy’ mean the same thing, then relative to any possible world at all, they will have the same extension and hence ‘every mimsy thing is cold and slimy’ will be not only true, but necessary. Third, and perhaps a little less surely, we notice that according to (1), anyone who understands our new word ‘mimsy’ has to know what it means. This is also true of ‘cold and slimy’ of course. But then, isn’t it reasonable to assume that someone who understands both of these words will be able to tell that they mean the same thing? If so, then they will be able to go through the reasoning we went through previously and come to the conclusion that the sentence ‘all mimsy things are cold and slimy’ is true. Notice that if they can do that, they can come to know that the sentence is true without going out and examining mimsy things to see whether or not they are cold and slimy, but rather just by thinking about what ‘all mimsy things are cold and slimy’ means. There is some precedent for calling this kind of justification ‘a priori’. Sentences which appear to be true in virtue of meaning, necessary and a priori seem special and it isn’t unnatural to mark that specialness by calling them ‘analytic’. On this account analytic sentences are true in virtue of meaning, and because of that, they also have the property of being necessary and a priori. The account that I have presented here is a very stripped down account. Most authors make it more specific by supplementing it with a story about how to characterize the sentences which are true in virtue of meaning (i.e., those in which the concept expressed by the predicate concept is contained in the concept expressed by the subject concept [Kant 1992 and 1965]), or those which follow from the logical axioms supplemented with meaning postulates (Carnap 1958, etc.) and at different times the account has been burdened with various additional theses. Some authors have held that analytic truths must be not only a priori, but obvious, certain or indubitable, that they must be really ‘about language’ and not about the world at all, or that they fail to state genuine facts. But since the traditional conception of the analytic-synthetic distinction is my target in this chapter, I want to consider only the most plausible core of the account, and not turn my target into a strawman by supplementing it with unnecessary additional theses. Hence I shall not assume that analytic truths have any additional properties, such as obviousness or certainty.

3 Four puzzles about the analytic/synthetic distinction 3.1 Puzzle 1: Directly referential names To say that a name, such as ‘Hesperus’, is directly referential is to say that its referent is not determined by the content of the name, that is, it is not determined by the meaning which the name contributes to the proposition

Analyticity in Externalist Languages

189

expressed by the sentence containing it. This is not to say that the referent is not determined by anything, which would be rather magical, but rather that it is determined by something else – perhaps a description, or a causalhistorical chain – which is not a part of the proposition expressed by the sentence. When the question of what such a name does contribute to the proposition comes up, directly referential names are often taken to contribute the referent itself, with the consequence that sentences containing directly referential names express Russellian propositions. The problem that this creates for the ASD arises when we consider true identity claims using directly referential names, such as: (1) Hesperus is Phosphorus. Sentence (1) does not fit cleanly into either the class of analytic sentences or the class of synthetic sentences. There is a sense in which (1) is true in virtue of its meaning; given the meanings (referents and/or contents) of ‘Hesperus’ and ‘Phosphorus’, these two names have to refer to the same thing. Then given the meaning of ‘is’ the sentence has to be true. This, along with the fact that the sentence expresses a necessary truth, and indeed, seems to be necessary because it is true in virtue of meaning, argues in favour of classifying it as analytic. But, there are four serious problems with that classification. First, the sentence is not intuitively analytic – competent speakers are unlikely to unreflectively assign it to the ‘analytic’ list when asked to sort groups of sentences. Second, whether or not the sentence expresses an a priori proposition, it clearly lacks something of the epistemic property that we attributed to ‘all mimsy things are cold and slimy’ in the previous section; competent speakers cannot, for example, learn that the sentence is true just by thinking about it; they must get out their telescopes, or consult an astronomy text. Third, unlike, say ‘Hesperus is Hesperus’, ‘Hesperus is Phosphorus’ is not plausibly – no matter how rich one’s logic – a logical truth. This is a worry because analyticity is quite naturally regarded as a kind of natural language analogue to formally defined logical truth. And finally, though the sentence may be true in virtue of meaning in one sense, it isn’t quite the sense in which ‘all mimsy things are cold and slimy’ was true in virtue of meaning. For example, with the later, we can say, ‘no matter what the extension of ‘mimsy’, the sentence must be true’ (as ‘mimsy’ always gets the same extension as ‘cold and slimy’, because their extensions are determined in the same way.) That could not be said truly of ‘Hesperus’ and the sentence ‘Hesperus is Phosphorus’. So our first puzzle is this: is a true sentence which uses two directly referential names to state an equality analytic? There seem to be clear objections to both affirmative and negative answers, and that suggests that there is something wrong with our concept of analyticity. 3.2 Puzzle 2: Semantic externalism Sentences like (2) All cats are animals.

190

Gillian Russell

differ from (1) in two crucial ways. First, it is rather more plausible that (2) is analytic than it was that (1) was, and second, (2) is not an identity sentence; it does not contain names, and the relation it expresses is non-symmetric. Yet (2) raises problems of its own. Hilary Putnam has argued that (2) can be false, which fits ill with the traditional picture of analyticity. (Putnam 1962) His argument is developed over three cases, of which the last is the most important. In it we are asked to imagine that there have never been any animals resembling cats on earth, or indeed anywhere else, but that the creatures we call ‘cats’ are, and indeed always have been, robots (perhaps placed on Earth to spy by the Martians.) Thus all human acquaintance with cat-like objects has really been with things that are not animals. Putnam points out that were we to discover this fact we would say, not that there were no cats, but that cats had turned out to be very different than we thought. Unlike dogs, horses, and koalas, cats turned out not to be animals. And hence the sentence ‘all cats are animals’ would be false under those circumstances. This, of course, fits ill with the intuition that the sentence is analytic for, according to the traditional picture, analyticity is sufficient to guarantee necessary truth. Moreover it is hard to maintain that a claim is a priori when its truth depends on an a posteriori claim such as that, at some time in the past, there were cats that were animals, and not robot spies from Mars. 3.3 Puzzle 3: Minimalism about semantic competence One of the main themes of post-Quinean twentieth century philosophy of language has been the idea that linguistic competence might not require knowledge of a criterion for uniquely distinguishing things that fall into the extension of an expression. For example, Putnam suggested that ordinary competent speakers of English who use the words ‘elm’ and ‘beech’ need not be able to distinguish elms from beeches. (Putnam 1985/73) The ideas of reference via causal-historical chains, division of linguistic labour, and semantic deference provide models of language on which users of an expression need not have knowledge of anything (far less a meaning) that would uniquely determine the extension of the expression, and so they tend to allow for very minimal requirements on semantic competence (Kripke 1980b; Soames 2001; Putnam 1985/73; Burge 1986 and 1991/79). This lowering of the standards for semantic competence threatens to erode the epistemic import of analyticity. If, as the picture of analyticity suggested by the folk picture suggests, all analytic truths can be known on the basis of what you have to know to be semantically competent, then, more or less, the less you have to know to be semantically competent, the fewer truths will be analytic. Timothy Williamson argues for an even more acute version of such minimalism about semantic competence, according to which there is no particular thing that one is required to know for semantic competence.

Analyticity in Externalist Languages

191

A complex web of interactions and dependencies can hold a linguistic or conceptual practice together even in the absence of a common creed that all participants at all times are required to endorse. ... although disagreement is naturally easier to negotiate and usually more fruitful against a background of extensive agreement, it does not follow that any particular agreement is needed for disagreement to be expressed in given words. (Williamson 2007, p. 125) He then argues that even a sentence such as (3): (3) All vixens are female foxes. is not analytic, on the grounds that someone could be competent with all the expressions in the sentence without considering it to be true. He asks us to consider, for example, Peter, who holds two rather unusual, but well-defended, views: one about the semantics of the universal quantifier, according to which (3) entails ‘there is at least one vixen’, and a further odd conspiracy theory on which there are no vixens. Peter thus rejects (3) and Williamson concludes that since a competent speaker could reject it, the sentence is not analytic. The worry is, of course, that radical minimalism about semantic competence allows this strategy to be extended to every putatively analytic sentence. Hence puzzle 3 is this: if we accept some form of minimalism about semantic competence, how we can we avoid the conclusion that there are very few, or even no, analytic sentences? 3.4 Puzzle 4: Indexicality The final puzzle arises when one considers indexicals, such as ‘I’ and ‘now’. A sentence containing an indexical will say different things in different contexts, because, given different contexts, the indexical contributes something different to the proposition. For example, when GR utters ‘I am going swimming’ she expresses the proposition that GR is going swimming, and when RM utters the same sentence, he expresses the distinct proposition that RM is going swimming. What makes this possible is that the contribution made by ‘I’ to the proposition expressed by the sentence containing it changes as the context changes. In the case of pure indexicals like ‘I’, ‘here’ and ‘now’ the contribution made by the expression is systematically determined by a rule which determines a function from contexts to propositional contribution. In these cases we can, for the sake of argument, adopt Kaplan’s suggestion that the rule for ‘I’ is that it always contributes the agent of the context, the rule for ‘here’ is that it always contributes the place of the context and the rule for ‘now’ is that it always contributes the time of the context. ‘I am here now’ is another sentence that does not fit happily into the traditional analytic or synthetic categories. It seems to be true in virtue

192

Gillian Russell

of meaning in just the kind of way traditional analytic sentences are. Moreover, the sentence is as plausibly a priori as any other analytic sentence. Yet on the traditional conception of analyticity, all analytic sentences express necessary truths and the proposition expressed by ‘I am here now’ is often contingent. Other examples of the contingent analytic include ‘The actual postmaster general is the postmaster general’ and ‘that [the shortest spy] is the shortest spy’, and these sentences raise the same issue: how can there be contingent analytic truths? 3.5 The source of the trouble The four puzzles arise out of conceptions of the way language works that are simply inconsistent with the folk picture of meaning. Direct reference is a view on which what an expression contributes to propositional content is distinct from the thing that determines referent. On Kaplan’s view of indexicals what the speaker has to know is not the same as what the expression contributes to the proposition. Minimalism denies that speakers know anything that determines reference, and externalism allows for the referent of an expression (relative to a particular world and time) to be determined by something other than what the speaker knows. Since the folk picture cannot accommodate these ubiquitous phenomena, perhaps it should not be so surprising that the account of analyticity which it supports is also inadequate in the face of them.1

4

Reviving the analytic/synthetic distinction

In presenting the new picture, I will present two definitions of analyticity and these two definitions are not co-extensional. Rather, the first is an approximation to analyticity expressed using more or less familiar tools, such as possible worlds and functions from contexts. The basic idea will be that the meanings of analytic sentences determine functions that return the value true, regardless of the context in which they are uttered and no matter what the world was like when they were introduced. But modal definitions of semantic concepts have had a tendency to fall short; the idea of a singular term that is directly referential, for example, is not completely captured by saying that it refers to the same object with respect to every possible world. To better capture direct reference, we follow Kaplan in pointing to a ‘metaphysical picture’ on which directly referential terms are those whose propositional-contribution and referent are identical. (Kaplan 1989b; King 2005) I will show that the first definition of analyticity runs into similar problems, and so the second definition of analyticity will make reference to a more metaphysical picture that underlies and explains the distinctive modal profile of analytic sentences.

Analyticity in Externalist Languages

193

4.1 A different approach to meaning To avoid the charge of ad hocness, I will begin by looking at a picture of meaning that can replace the folk view. Instead of assuming that expressions have a single thing – their meaning – which determines extension, is known to speakers and contributed to propositions, we will allow that these three roles may be played by distinct things. If there may be three different things there, it behooves us to have three names for them, and so in place of the more general word ‘meaning’ I will use the following three technical terms: reference determiner, content and character. The reference determiner of an expression is what determines the function from the way the world is to the extension of the expression. Though the expression reference determiner may be new and strange, the idea that it stands for is old and familiar: when expressions have referents, that fact is not a miracle; ‘elastic’ has the extension it has because of what it means (its reference determiner). The usual facts about the reference-determining aspect of meaning all hold; if two expressions have the same reference determiner, they have the same referent with respect to all possible worlds. Hence if expression A applies to object c at w1, and B does not, then the reference determiners of A and B must be different. The content of an expression is what it contributes to the proposition expressed by a sentence containing it. This is the notion of content familiar from Kaplan. (Kaplan 1989b) But where Kaplan used ‘character’ to refer to an aspect of meaning that was both what was known to speakers and what determined reference, I will reserve it for the former. Hence character is what a speaker has to know to understand an expression. The new picture allows for content, character and reference determiner to come apart, but there is no prohibition against two or more roles being played by the same thing. 4.2 Truth in virtue of what? Once we have replaced the unified folk notion of meaning with three things, one might wonder what has become of truth in virtue of meaning. I contend that what has been called ‘truth in virtue of meaning’ is really truth in virtue of reference determiner.2 Truth in virtue of reference determiner fits well with some of the stories that we tell about analyticity, and even where it does not have the properties that earlier philosophers have attributed to analyticity, it is possible to see why those philosophers thought that it did.3 At the beginning of this chapter I noted that there are two things that go into making a synthetic sentence true – what it means, and the way the world is. It is less commonly noted that this can be generalized to subsentential expressions and their referents. For example, ‘the longest book in the British Museum’ has the referent it has in part because of what it means, and in part because of the way the world is; had the description

194

Gillian Russell

meant what ‘the longest sarcophagus in the British Museum’ means, or had there been a different set of books in the British Museum, the description could have picked out a different object. Similarly, for ‘bachelor’. Things get trickier when we consider indexicals and names. If we stipulate that by ‘meaning’ we mean content, then the story fails to get off the ground. Whatever possible world we are considering, ‘Hesperus’ refers to the same object; it is a rigid-designator. The same goes for indexicals like ‘I’. The other half of the story fails too: it is downright misleading to say that by changing the content of ‘Hesperus’ or ‘I’ one can change the referent, because in both these cases the content is the referent. In the case of indexicals there is an obvious alternative: we specify that by ‘meaning’ we speak this time not of content, but of reference determiner, and by ‘world’ we mean not a circumstance of evaluation, but rather a context of utterance. On this construal it is straight-forwardly true that by varying either the meaning or the world, we can vary the extension of the expression: for example, had you been the agent of the context (this is varying the ‘world’), then ‘I’ would have referred to you, instead of to me, and had the reference determiner of ‘I’ been ‘the hair-colour of the agent of the context’ (this is varying the meaning), then ‘I’ would have referred to a colour, and not to me. This suggests two things: first, that the meaning in the story told to introduce analyticity is reference determiner, as opposed to say, content, referent or character; and second, that that meaning can determine a richer kind of function which may take contexts of utterance, and not just ordinary possible worlds, as arguments. Moral support for this idea can be found in Kaplan’s adaptation of the idea of analyticity/logical truth to encompass indexicals and the contingent analytic.4 Kaplan suggests that a sentence is analytic if it is true with respect to every context of utterance. Hence ‘The red book is red’ and ‘I am here’ are analytic in Kaplan’s sense, because there is no context in which they can be uttered falsely, and ‘Snow is white’ and ‘I am in St Louis’ are not, since there are contexts of utterance with respect to which they are false.5 The main problem with Kaplan’s suggestion – as an approach to defining analyticity – arises when we turn to considering sentences containing names and natural kind predicates. The sentence ‘Hesperus is Phosphorus’, being necessary but non–context-sensitive, returns the same truth-value with respect to every context of utterance. Hence it is analytic on Kaplan’s definition, and this is a counterexample to that definition. 4.3 Contexts of introduction Kaplan’s definition can be adapted in a natural way, however, to give us the first of the two definitions of analyticity that I want to endorse in this chapter. Names and natural kind predicates may not be context of utterance–sensitive but they do have extensions and referents, and those have to be determined somehow. What is distinctive about words such as ‘cat’ and

Analyticity in Externalist Languages

195

‘Hesperus’ is that their referent depends on the state of the world when they were introduced. Suppose, for example, that the word ‘Hesperus’ was introduced when someone pointed to a bright spot in the sky in the evening and said ‘Let’s call that “Hesperus”.’ Then ‘Hesperus’ refers to the thing that was that bright spot (the planet Venus) regardless of which possible world we are discussing and regardless of context of utterance. Yet it remains true that had the bright spot in the sky been the planet Mars, the name ‘Hesperus’ would have referred to Mars and not to Venus even though its referent was determined in the same way (pointing and speaking the same words with the same meaning) as in the actual world. Call the state of the world at the time when the expression was introduced the ‘context of introduction’. Then I am claiming that the extensions of names and natural kind predicates can vary with context of introduction. It is our implicit awareness of this fact that tells us in Putnam’s third case that where there could not have been an animal around when the word ‘cat’ was introduced, ‘cat’ must apply to non-animal creatures. Kaplan taught us to think of character (in the loose, modal sense) as a function from contexts of utterance to contents, which were in turn functions from circumstances of evaluation to extensions (meaning that we might just as well think of characters as a function from contexts of utterance and circumstances of evaluation to extensions). I suggest that we think of a reference determiner for an expression as a function from contexts of introduction and contexts of utterance to contents, or what amounts to much the same thing, as a function from a context of introduction, context of utterance and a circumstance of evaluation to a referent. Contexts of introduction can be thought of as providing contents for names and natural kind predicates, contexts of utterance as providing contents for indexicals, and circumstances of evaluation as providing a truth-value for the proposition that results. A sentence will be analytic just in case its reference determiner returns the value true for all argument pairs consisting of a context of introduction and a context of utterance. At a rough intuitive approximation, that means that a sentence is analytic if it would express a true proposition, even if the context of utterance had been different, even if you’d uttered it on Twin (or some n-tuplet) Earth, where the extensions of names and natural kind predicates are different.6 Definition 1 (Analyticity (modal definition)) A sentence S is analytic just in case for all pairs of context of introduction (c i ) and context of utterance (c u = 〈ac u, pc u, tc u, wc u〉), the proposition expressed by S with respect to 〉 〈ci, c u〉 is true at wc u and tc u .

5 Back to the puzzles The new definition of analyticity resolves puzzle 1; ‘Hesperus is Phosphorus’ is not true in virtue of meaning because it is not true with respect to all context pairs (ui, uc). Here is a counterexample: if ui is a context in which

196

Gillian Russell

reference determiner for ‘Hesperus’ (say, the evening star as seen from London on 27th July 1789) picks out Mars and the reference determiner for ‘Phosphorus’ (say, the morning star on 1st January 1540) picks out Mercury, then the sentence ‘Hesperus is Phosphorus’ is false. Hence we need not be concerned that it fails to have the other properties associated with analyticity, because it is not analytic. Puzzle 2 is resolved in a similar way: ‘cats are animals’ is not analytic, because there are context-pairs with respect to which it may be false, for example, the one where the objects we call ‘cats’ have all always been robots and not animals. Putnam’s thought experiment itself is a demonstration that the sentence is non-analytic. In the puzzle 2 case, more should be said, for if this sentence is not analytic, why does it seem analytic? The folk picture can provide an error theory here. Many people think both that i) they are competent speakers and that ii) (as the folk picture requires) in order to be competent, they have to know how the referents of the various expressions are determined. Couple this with the psychological fact that people who think that they must know the answer to some question are apt to think that the very first answer that comes to mind is the right one, and we can sketch a story about why ‘all cats are animals’ seems analytic: What is the reference determiner for ‘cat’ ? If you implicitly accept the folk picture, you could be forgiven for answering something like ‘smallish furry animal with pointy ears, four legs and a tail’7 But if this were correct, then of course all ‘cats are animals’ would be analytic. The apparent analyticity of ‘cats are animals’ is an artefact of the folk theory of meaning. So far I have said nothing about the epistemic and metaphysical consequences of a sentence’s being analytic. The new picture is one on which we have to be very careful about making the distinction between sentences and the propositions the sentences express. The relationship between the two is no longer 1–1, or even many–1, but many-many, since an indexical sentence may express many different propositions and the same proposition may be expressed by different sentences. Since it is sentences that are analytic, but propositions that are necessary or a priori, we need to be very careful about exactly what we are attributing modal or epistemic properties to as a consequence of the analyticity of some sentence. Since ‘I am here now’ is analytic on this definition, it should be clear that it is not a consequence of my account that analytic sentences express necessary propositions – this is the solution to puzzle 4. But in saying this we leave ourselves open to the following objection: analyticity is supposed to entail necessity. If your new account of it does not allow for this, then it is not an account of analyticity. I disagree. Even on my view, analytic sentences are such that they will be true no matter what the world is like. Is it any wonder that earlier writers have sometimes confused this with the property of expressing a necessary

Analyticity in Externalist Languages

197

proposition? If one restricts one’s attention to sentences which express the same proposition regardless of context, then an analytic sentence will express a necessary proposition. Earlier writers did not have as good a grip on indexicality, or causal theories of reference as we do, and so they were unfamiliar with the main counterexamples to the claim that sentences express the same proposition regardless of context. It is no surprise then, that they tended to slip from the claim that a sentence could not be false, to the claim that what it said couldn’t be false. But that is no reason to stick analyticity with the claim that analytic sentences express necessary propositions – rather, we are now in a position to see that what analytic sentences really have is a distinctive kind of modal profile: they are true regardless of context. What of their epistemic status? The special status of analytic truths in the traditional account depended on two facts. First, that it was the meaning of the sentence that was (fully) determining the truth value of the sentence, and second, that that meaning was something that was known to competent speakers. Somehow we were meant to conclude from this that competent speakers would be able to tell that the sentence is true. But this suggestion of an argument will not survive the new account because it equivocates on the word ‘meaning’; there is no guarantee, on the new account of meaning, that the thing that competent speakers must know (character) is the thing that is determining the truth-value (reference determiner.) I want to suggest that what is epistemically special about analytic sentences is that someone who knows the reference determiners for the expressions contained in them (as well as some basic background facts about how the language works, such as how the reference determiners of complex expressions are computed from their parts), is in a position to work out that the sentence is true, that is, he is in possession of the premises of an argument to the conclusion that the sentence is true. Where the sentence is of the kind that expresses the same proposition regardless of context, he will also be in a position to deduce the proposition it expresses, i.e., not just that ‘all bachelors are unmarried’ is true, but that all bachelors are unmarried. This knowledge of reference determiners is not normally a priori, and nor is it generally linked with the idea of semantic competence. But then, in suggesting that analyticity is a matter of truth in virtue of reference determiner, I am suggesting that analyticity has much less to do with semantic competence than is usually thought. What is epistemically interesting about analytic sentences is that from knowledge of facts about language, i.e., that ‘bachelor’ applies to an object iff it is an unmarried male, one can conclude something non-linguistic, i.e., that all bachelors are unmarried. An advantage of having defined analyticity independently of its epistemic properties like this is that we can now ask whether there are any examples of analytic sentences which fail to have the epistemic properties

198

Gillian Russell

usually attributed to analytic sentences. I think there are, and to see this it will help to consider the phenomenon of parasitic reference determination where the reference determiners are, for the most part, unknown to speakers. Consider a situation (perhaps not so different from the actual one) in which the name ‘Cassius’ was introduced to the language when Mr Clay, indicating his newborn son, said ‘You’re right; it’s a good name. Let’s call him “Cassius”.’ The name ‘Muhammed Ali’, on the other hand, is introduced in a different way that causes its reference determiner to be parasitic on the reference determiner for ‘Cassius’. At a certain point in his boxing career and journey towards the Muslim faith, Cassius’ spiritual advisor says ‘From now on, Cassius will be known as “Muhammed Ali”.’ Cassius and the rest of the community adopt this name. As a result of this second stipulation, ‘Muhammed Ali’ will now refer to whoever ‘Cassius’ referred to in the advisor’s utterance. But, of course, the reference determiner for ‘Cassius’ is not sensitive to the context in which the advisor did his uttering, but only to the context in which ‘Cassius’ was introduced. As a result, the reference determiner for ‘Muhammed Ali’ will be sensitive to the context in which ‘Cassius’ was introduced. As evidence for this, consider what we would say if Mrs Clay’s baby had been replaced with an alien baby spy whilst still in the womb, whilst her biological baby went off to live a new life on Mars. Alien Spy Baby is eventually born, baptized ‘Cassius’ by Mr Clay, and grows up to become a great boxer. Later on his advisor says ‘From now on, Cassius will be known as “Muhammed Ali”.’ In these circumstances we would say that Cassius is an Alien Spy, but also, crucially, that Muhammed Ali is an Alien Spy, which demonstrates that the referent of the second name is sensitive to changes in the context of introduction for the first. Now consider what the modal definition of analyticity says about the sentence ‘Cassius is Muhammed Ali’ compared to what it says about ‘Hesperus is Phosphorus’. ‘Hesperus is Phosphorus’ was not true in virtue of meaning because there are contexts of introduction with respect to which ‘the evening star’ picks out Mars and the morning star picks out Venus. But the parallel situation cannot arise with ‘Cassius is Muhammed Ali’ so that sentence will be true in virtue of meaning, though clearly it will not be a priori to many speakers that Cassius is Muhammed Ali. ‘I am here now’ is also true in virtue of meaning, and epistemically it is an interesting case. Indexicals are rather unusual expressions in that speakers normally know how to work out what the referent is, given the context – they have to implicitly know what the reference determiner is. But this knowledge means that they know the premises of the argument to the conclusion that the sentence ‘I am here now’ is true. Hence semantic competence does give one access to the fact that this sentence is true. But in a different respect both ‘Cassius is Muhammed Ali’ and ‘I am here now’ are on an epistemic par: it is the case that if one knows the reference

Analyticity in Externalist Languages

199

determiners for all the expressions in the sentence, one can work out that the sentence is true. It is this property that is the epistemic consequence of analyticity, and it takes the natural but false assumption that competent speakers know the reference determiners for expressions to derive the false claim that competent speakers can work out that analytic sentences are true. For these reasons I hold, pace Williamson, that minimalism about semantic competence does not lead to minimalism about analyticity itself, but rather to minimalism about the access that semantic competence gives to analytic truths.

6 Beyond the modal definition The previous definition of analyticity was given more or less in terms of possible worlds along with some useful concepts from set theory. Despite the real progress represented by possible worlds analyses of semantic concepts, they have had a tendency to fall short. There is no satisfactory definition of direct reference, or content in terms of possible worlds. Newtonian mechanics is powerful and useful, but tends to get things wrong at really high relative velocities. The possible worlds approach to semantics is similarly powerful and useful, but it tends to get things wrong in the presence of non-linguistic necessity (Thomason 1974; Kaplan 1989a; Soames 1987; 1998, 1999, 2001, 2004; King Summer 2005). My own modal definition of analyticity is no exception, and, where nonlinguistic necessity enters the picture, we can have sentences which have the same distinctive modal profile as an analytic sentence, even though they didn’t get that profile because of something special about their meaning, but rather because of the substantive metaphysical facts. For theists, one example might be ‘there is a god’, which (at least if we insist on a certain non-indexical interpretation of the quantifier) contains no names or indexicals, and so if it expresses a necessary truth, no variation in the contexts of introduction or utterances will make it false. Or suppose, for a crazy but particularly clear example, that there are only 15 possible worlds, and in every possible world, the evening star is the morning star. Then ‘the evening star is the morning star’ would express a necessary truth, and do so without being analytic. Even if these are not very convincing examples, I hope that the problem they illustrate is clear: our modal definition is hostage to the facts about metaphysical necessity. But whether or not a sentence is analytic should depend on facts about its reference determiners alone, and not on facts about the modal world. Unlike physicists working with Newtonian mechanics, we do not yet have a well-developed alternative to the possible worlds approach to semantics. The best alternative is mostly restricted to the concepts of content and direct reference, and it is the Russellian/Kaplanian ‘metaphysical picture’ according to which propositions are set-theoretic sequences which take ordinary

200 Gillian Russell

objects and perhaps properties as elements. I propose to extend this metaphysical picture so that it can be used to provide a stricter definition of analyticity. I will only have space to sketch the basic idea here. Sentences such as ‘Cassius is Cassius’ and ‘bachelors are bachelors’ seem to be analytic (prescinding from worries about reference failure in the first sentence for now) because the reference determiner for the first occurrence of ‘Cassius’ is the same as the reference determiner for the second occurrence of ‘Cassius’ and similarly for ‘bachelors’. The same will hold if the reference determiners for ‘bachelors’ and ‘unmarried men’ are known to be the same, even though the words are different: ‘bachelors are unmarried men’ will be analytic. It seems that identity of reference determiner, used appropriately in a sentence, can induce analyticity. But so can something else. ‘Mohammed Ali is Cassius’ and ‘bachelors are unmarried’ are analytic, but in these sentences it is not that the reference determiner of one expression is identical with that of another, but that the reference determiner of one is parasitic on, or part of, or contained in the other. Having noticed this, we see that identity is simply the limit case of containment – it is containment that is the more general notion. There is a long history of using something like containment to define analyticity. Before we had possible world semantics, this was the concept for which we instinctively reached. For example, Locke writes that a claim is ‘trifling’ when, ‘a part of the complex idea is predicated of the name of the whole’ (Locke 1993/1690) and Kant that, ‘Either the predicate B belongs to the subject A, as something that is (covertly) contained in this concept A; or B lies outside the concept A, although it does indeed stand in connection with it. In the one case I entitle the judgments analytic, in the other synthetic.’ (Kant, First Critique, A7/B11). I want to suggest that the metaphysical picture for analyticity should include reference determiners, and containment relations between reference determiners. It will not suffice, however, to replace the words ‘belongs to’ in Kant’s definition with ‘contains’ and ‘predicate B’ with ‘reference determiner of B’, because 250 odd years have shown up some other technical problems with Kant’s approach (Frege 1980/1884; Quine 1951 and 1961; Katz 1967 and 1974). However we can generalize Kant’s notions of subject and predicate to the notions of logical subject expression (LSE) and logical predicate expression (LPE) and to invoke a third sentence-part, the modifier (M) of the sentence. Roughly the (LSE) is the expression, or sequence of expressions that tell us what the sentence is making a claim about. The (LPE) is the expression in the sentence that tells us what the sentence is saying about the referent of the (LSE). In the simplest case, a sentence will be true just in case the objects which satisfy the (LSE) meet the condition specified by the (LPE), e.g., ‘snow is white’ will be true just in case the objects which satisfy the reference determiner for ‘snow’ meet the condition specified by ‘is white’. It seems

Analyticity in Externalist Languages

201

clear that something special will happen if meeting the condition specified by the (LPE) is a prerequisite for satisfying the (LSE): so long as something satisfies the (LSE) the sentence cannot help but be true. Hence we might be tempted to say that a sentence is analytic just in case the reference determiner for the (LPE) is contained in that of the (LSE). But this would not be quite right. Consider what should be said about putatively analytic sentences like ‘no bachelor is married’, ‘it is not the case that any bachelor is married’, ‘Mary walks with those with whom she herself strolls’ and ‘poor people have less money than rich people’, all of which are putatively analytic, but do not meet the suggested definition. Moreover, more has to be said about the effects of failure to refer. Finally, the proximity of analyticity to logical truth suggests that analyticity might be generalizable to arguments, as in: Sam is a bachelor. Therefore, Sam is unmarried. If so, analytic truth might turn out to be a special case of analytic consequence and analytic truths might then be divisible into two kinds, i) core analytic truths in which a particular containment relation holds between the subject and the predicate, and ii) analytic consequences of core analytic truths. More work needs to be done on all these topics, but I will finish by simply presenting and illustrating my second definition: Definition 2 (Analyticity (metaphysical picture)) A sentence that consists of modifier (M), logical subject expression (LSE) and logical predicate expression (LPE), is analytic if (i) the sentence can be true even if the reference determiner for the (LSE) is not met by anything, and (ii) either (M) is positive and the reference determiner for (LSE) contains the reference determiner for (LPE) or M is negative and the reference determiner for (LSE) excludes the reference determiner for (LPE). At an intuitive level, the modifier (M) is a part of the sentence that can modify the relation required between the object(s) picked out by the (LSE) and the condition specified by the (LPE) in order for the sentence to be true. For example, while ‘bachelors are unmarried’ will be true just in case the objects picked out by the (LPE) (bachelors) meet the condition specified by the (LSE) (being unmarried), ‘no bachelors are married’ will be true just in case the objects picked out by ‘bachelors’ fail to meet the condition specified by ‘married’. This modification can get pretty complicated, for consider ‘some bachelors are unmarried’, ‘four bachelors are unmarried’ etc. However, modifiers can be split into three kinds, and only two are important when it comes to analyticity. Positive modifiers require that the condition specified by the (LPE) be met by every object which meets the (LSE) in order for the

202 Gillian Russell

sentence to be true. Examples include ‘all’ and ‘every’. Negative modifiers require that no object which meets the (LSE) satisfy the condition specified by the (LPE). The paradigm case is ‘no’ in ‘no bachelors are married’. Every other kind of modifier is neutral. Clause (i) in the definition is simply intended to rule out cases where the sentence will be false if the (LSE) is not met by anything.8 Clause (ii) in the definition specifies two ways in which a sentence can be analytic. Either the modifier is positive and the reference determiner for the LSE contains that of the LPE. The sentence ‘all bachelors are unmarried’ is analytic in this way. Alternatively the modifier is negative and the reference determiner for the LSE excludes that of the LPE. The sentence ‘no bachelor is married’ is analytic in this way.

7

Conclusion

In this chapter I have presented four puzzles for analyticity. They are not the usual objections presented by Quine and his followers, but rather puzzles that have arisen out of post-Quinean philosophy of language. Where there is an awareness of these problems for the traditional conception of analyticity, it is usually assumed that they provide one more reason to give up on the ASD. I have argued that this is not the case, and that the puzzles arise because of phenomena which undermine the folk theory of meaning that supported the traditional account of analyticity. I have argued that revamping that folk account of meaning into one which is compatible with the new phenomena will give us a new account of truth in virtue of meaning, in effect, a new account of the ASD. In part my account has been presented in terms of the usual functions between contexts and possible worlds, and I have shown how this account can solve all four puzzles. But the definition given in these terms is susceptible to a problem that besets all semantic definitions given in modal terms – the definition tends to go awry in the presence of substantive necessity. My response to this problem has been to sketch the beginnings of a metaphysical picture – on the model of Kaplan’s metaphysical picture for direct reference – and use it to define analyticity. This last section of the chapter has been a bit wild and sketchy but, I hope, not without interest anyway.

Acknowledgements I would like to thank Frank Artzenius, Gilbert Harman, John Hawthorne, Sarah Sawyer, and Scott Soames for helpful discussions of the material presented in this chapter, as well as the members of the Bay Area Philosophy of Language Discussion Group, including Kent Bach, Michael Caie, Michael Glanzberg, and Ken Taylor. Thank you also to Iain Russell for assistance with manuscript preparation.

Analyticity in Externalist Languages

203

Notes 1. One might wonder whether accommodating these phenomena requires a complete rejection of the folk picture; at times Putnam seems to suggest that it is not that the folk view is wrong, so much as that it only applies to a restricted type of expression (e.g., to expressions like ‘chair’ and ‘bachelor’ and not to ‘water’.) If so, then we can imagine a theory according to which there were several ways in which expressions could work, and one picture of the way meaning worked was right for certain expressions, and another for others. This is a view that reminds me strongly of the famous passage from the Investigations in which Wittgenstein writes: Think of the tools in a tool-box: there is a hammer, pliers, a saw, a screwdriver, a rule, a glue-pot, glue, nails and screws. – The function of words are as diverse as the functions of these objects. (Wittgenstein 1953, section 11) Wittgenstein’s view is a much more radical one than the one I am entertaining here, yet the metaphor is a useful one: even within the part of language that is used for the tasks of asserting, describing and naming we can think of different expressions as having different kinds of uses: names can be the hammers, the definite descriptions the paints etc. small wonder then that they have different semantic, metaphysical and epistemic properties when they are used for such different things. If we were to adopt this picture, we might stage a rescue of the ASD by claiming that it only applies to certain parts of language – the parts where the traditional story about meaning is correct – and it has no place where there are indexicals, names, natural kind predicates, etc. or where phenomena like semantic deference and division of linguistic labour have taken a hold. That conservative approach would be a mistake, however. One reason is that there are analytic-sentences which contain, and even rely upon, the unfolky phenomena, such as ‘all red water is water’ and ‘I am here now’. A second is that there is reason to doubt that there are any expressions that function as the naive folk view holds that all expressions function, especially as liberalism about semantic competence has a tendency to leak all over the language. Williamson provides numerous examples to support the thesis that liberalism about semantic competence is not limited in its scope to a particular subset of words. (Williamson 2008, chapter 4) 2. One way to think about the kind of claim I am making here is this: there are real, intuitive phenomena – of which analyticity is one – out there that we, as philosophers, are aware of and attempt to characterize. This is difficult and even the best of philosophers will often misdescribe the phenomenon in which they are interested, and attribute properties to it that are really properties of something else, or properties of nothing. Twentieth-century philosophy of language provided us with a surprising thought, namely that the reference determiner of an expression had often been misdescribed as the meaning of an expression, whereas in fact it is often – say in the case of names and natural kind predicates – something that is distinct from the expression’s meaning, because it is neither known to speakers nor a part of what is said. I merely extend this view: truth in virtue of reference determiner has often been misdescribed as truth in virtue of meaning. Still, it is truth in virtue of reference determiner that is the phenomenon that is really of interest. 3. Occasionally I have heard the suggestion that since there are four different kinds of meaning (when one includes referent/extension) on my view, there might be four

204

4.

5.

6.

7. 8.

Gillian Russell

different kinds of truth in virtue of meaning. Perhaps, but only truth in virtue of reference determiner is the important one for analyticity. One argument against taking truth in virtue of meaning to be truth in virtue of character is Williamson’s: since characters have turned out to be minimal or non-existent, very little is true in virtue of character, and so that approach does nothing to account for the familiar notion of analyticity – all we’re left with is an error theory. An argument against taking truth in virtue of meaning to be truth in virtue of extension is that varying the extension of an analytic sentence appears to have no effect on its analyticity: ‘all bachelors are unmarried’ comes out analytic independently of which people are bachelors: a few more people marrying or having sex-changes will not affect the analyticity of the sentence. An argument against taking truth in virtue of meaning to be truth in virtue of content is that in cases of analytic sentences where the content and reference determiner of their component expressions come apart, such as with analytic sentences containing indexicals, varying the content seems to have no effect on the analyticity of the sentence: ‘I am here now’ is analytic whether it expresses the content that GR is at work, or whether it says RM is at home. Strictly speaking, the Kaplan of Demonstratives sometimes calls the phenomenon ‘the contingent analytic’ and sometimes ‘the contingent a priori’. I hold that the former was a more apt description. Formally, the world of evaluation makes an appearance as a part of the context of utterance, construed as a quadruple (a, p, t, w) in which the fourth element is a possible world. A sentence is true in a context if the proposition it expresses in that context is true at the world of that context. The only admissible contexts are those in which the agent (a) is located at the place (p) and time (t) of the context in the world (w) and hence ‘I am here now’ comes out true at all contexts. A sentence is true relative to a context of introduction and context of utterance pair just in case it is true relative to the triple of that context of introduction, that context of utterance, and the circumstance of evaluation that is given by the time and world of the context of utterance. This is how we get the two-place function from the three-place one. This of course, cannot be right, if only because cats who have lost a leg are still cats. This clause deals brutally with reference-failure cases, such as ‘Muhammed Ali is Cassius Clay’. They do not count as analytic, because the sentence would not be true if say Mrs Clay had had a phantom pregnancy and was referring to her own hallucination when she introduced the name ‘Cassius’. Overall, this seems too brutal to me, and we can soften the blow by introducing a class of pseudo-analytic sentences, which meet only clause ii) of the definition of analyticity, i.e., Definition 3 (Psuedo-Analyticity) A sentence that consists of modifier (M), logical subject expression (LSE) and logical predicate expression (LPE), is pseudo-analytic if (M) is positive and the reference determiner of the (LSE) contains the reference determiner of the (LPE) or M is negative and the reference determiner of the (LSE) excludes the reference determiner of the (LPE).

Bibliography Boghossian, P.A. (1996), ‘Analyticity Reconsidered’, Nous, 30(3), 360–91. Burge, T. (1986) ‘Individualism and Psychology’, Philosophical Review, 95, 3–45. —— (1991/79) ‘Individualism and the Mental’, reprinted in D. Rosenthal (ed.) The Nature of the Mind (London: Oxford University Press). Carnap, R. (1958) ‘Meaning Postulates’, in Meaning and Necessity, 2nd edn (Chicago: University of Chicago Press).

Analyticity in Externalist Languages

205

Frege, G. (1884) The Foundations of Arithmetic, 2nd edn (Evanston, Ill.: Northwestern University Press). Kant, I. (1965) Critique of Pure Reason, trans. Norman Kemp-Smith (New York: St. Martin’s Press). —— (1992) ‘Jäsche Logic’, in J.M. Young (ed.) Lectures on Logic (Cambridge: Cambridge University Press). Kaplan, D. (1989a) ‘Afterthoughts’, in J. Almog, J. Perry, and H. Wettstein (eds) Themes from Kaplan (New York: Oxford University Press). —— (1989b) ‘Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology of Demonstratives’, in J. Almog, J. Perry, and H. Wettstein (eds) Themes from Kaplan (New York: Oxford University Press). Katz, J. (1967) ‘Some Remarks on Quine on Analyticity’, Journal of Philosophy, 64, 36–52. —— (1974) ‘Where Things Now Stand with the Analytic-Synthetic Distinction’, Proceedings of the Aristotelian Society, 28, 387–94. King Summer, J.C. (2005) ‘Structured Propositions’, in E.N. Zalta (ed.) The Stanford Encyclopedia of Philosophy (Stanford: The Metaphysics Research Lab, Stanford University). URL: http://plato.stanford.edu/archives/sum2005/entries/ propositions-structured/ Kripke, S. (1985/80a) ‘Naming and Necessity’, in A.P. Martinich (ed.) The Philosophy of Language, 4th edn (Oxford: Oxford University Press). —— (1980b) Naming and Necessity (Oxford: Blackwell). Locke, J. (1993/1690) An Essay Concerning Human Understanding (London: Everyman). Putnam, H. (1962) ‘It Ain’t Necessarily So’, Journal of Philosophy, 53, 658–71. —— (1985/73) ‘Meaning and Reference’, in A.P. Martinich (ed.) The Philosophy of Language, 4th edn (Oxford: Oxford University Press). Quine, W.V.O. (1961) ‘Two Dogmas of Empiricism’, in W.V.O. Quine, From a Logical Point of View, 2nd (revised) edn (Cambridge, Mass.: Harvard University Press). —— (1951) ‘Two Dogmas of Empiricism’, Philosophical Review, 60, 20–43. Russell, G. (2008) Truth in Virtue of Meaning: A Defence of the Analytic/Synthetic Distinction (Oxford: Oxford University Press). Salmon, N. (1993) ‘Analyticity and A Priority’, Philosophical Perspectives, 7, 125–33. Soames, S. (1987) ‘Direct Reference, Propositional Attitudes, and Semantic Content’, Philosophical Topics, 15(1). —— (1998) ‘Facts, Truth Conditions and the Skeptical Solution to the Rule-Following Paradox’, Language, Mind and Ontology, 12, 313–48. —— (1999) Understanding Truth (New York: Oxford University Press). —— (2001) Beyond Rigidity (Oxford: Oxford University Press). —— (2004) Reference and Description: The Case against Two Dimensionalism (Princeton, N.J.: Princeton University Press). Thomason, R. (1974) ‘Introduction’, in R. Thomason (ed.) Formal Philosophy: Selected Papers of Richard Montague (New Haven: Yale University Press). Williamson, T. (2007) The Philosophy of Philosophy (Oxford: Blackwell). Wittgenstein, L. (1953) Philosophical Investigations, 3rd edn (Englewood Cliffs, N.J.: Prentice Hall).

10 The Modified Predicate Theory of Proper Names Sarah Sawyer

1

Introduction

Tyler Burge (1973) argues for what he calls the ‘modified predicate view’ of proper names – the view that proper names are predicates in their own right. The view contrasts with the traditional predicate view according to which a name is an abbreviated or manufactured predicate; and it contrasts with the rather more popular view that names have the semantical role of individual constants.1 The modified predicate view is constituted by three central theses: (T1) A proper name is a predicate true of an object if and only if the object was given that name in an appropriate way. (Burge 1973, p. 428) (T2) Proper names in singular unmodified form, functioning as singular terms, involve a demonstrative element. (Burge 1973, p. 432) (T3) A proper name occurring as a singular term in a sentence used by a person at a time designates an object if and only if the person refers to that object at that time with that proper name, and the proper name is true of that object. (Burge 1973, p. 435) Burge is concerned primarily with the logical role of proper names in a semantical account of natural languages; his chosen framework is Tarskian truth theory2 as applied to the sentences of a person at a time; and he states as a condition of adequacy on any such semantical account ‘that the theory of truth be fully formalized – that is, that the sense and reference (if any) of every expression of the theory should be unambiguously determinable from its form’ (Burge 1973, pp. 425–6). This means that while context can be relied upon to determine the referent of a particular occurrence of an indexical, demonstrative or ambiguous expression in the object-language, the meta-language must be free from such context-dependence: it must instead provide an analysis of the role context plays in the object-language and be free from ambiguity.3 206

Modified Predicate Theory of Proper Names

207

With this much as background, Burge’s modified predicate view has three clear virtues. First, the view provides a unified treatment of both singular and non-singular uses of proper names. Names are accorded the same semantical role whether in singular, unmodified form – as exemplified by ‘Alfred’ in (1)4: 1. Alfred studies in Princeton. – or in plural or modified form – as exemplified by ‘Alfred’ in (2)–(5): 2. 3. 4. 5.

There are relatively few Alfreds in Princeton. An Alfred Russell joined the club today. The Alfred who joined the club today was a baboon. Some Alfreds are crazy; some are sane.

The unified treatment is made possible by treating non-singular uses of proper names as semantically primary and singular uses as semantically derivative – as is indicated by (T1) and (T2) respectively. This seems to go against the common intuition that singular uses of proper names are more fundamental than non-singular uses. However, the common intuition can be accommodated by understanding it as an expression of the pragmatic or epistemic primacy of singular uses of proper names. And the pragmatic or epistemic primacy of singular uses is of course consistent with the semantic primacy of non-singular uses. On this understanding, non-singular uses of proper names display their logical form on their sleeves, so to speak, whereas singular uses of proper names do not. Singular uses such as ‘Alfred’ as it occurs in (1) are to be understood as semantically equivalent to demonstrative expressions such as ‘that Alfred’, and hence as falling within the same semantic category as explicit demonstrative-predicate constructions such as ‘that cat’. Proper names in all their uses, then, are predicates. A unified treatment of proper names such as that afforded by the modified predicate view is virtuous because it explains the obvious connection between singular and non-singular uses of names. This is illustrated by the intuitive validity of inferences of the following kinds: 6. Alfred Jones lives in Princeton, so there is at least one Alfred living in Princeton. 7. Alfreds tend to be over 50, so Alfred Jones is probably over 50. Sentence (6) involves a move from a singular to a non-singular use of a proper name, and (7) relies on a move from a non-singular to a singular use of a proper name. As such, the validity of each inference relies on there being a semantic connection between singular and non-singular uses of a given proper name. In the absence of such a semantic connection the intuitive validity of the inferences would be inexplicable.5

208 Sarah Sawyer

The second virtue of the theory – and one that is, in a sense to be made clear, related to the first virtue – is that it allows for unrestricted quantification over individuals who share a name. In essence, the view assigns the correct truth-conditions to sentences containing plural occurrences of names such as (2) and (5) as well as (8) and (9) that follow6: 8. There will be many Alfreds next century. 9. There are many Alfreds to whom I shall never effect singular reference. A correct understanding of such sentences depends upon understanding that the plural use of ‘Alfred’ in each case concerns all Alfreds including those who are either not known to the speaker or known to the speaker but not by that name. The sentences are not correctly understood, for example, as concerning merely the set of individuals called ‘Alfred’ by the speaker. Indeed, (9) is barely intelligible on such an understanding. As Hornsby says: ‘The set of Alfreds an individual calls by name is most likely a proper subset of the extension of the predicate “Alfred”, but to use the name correctly predicatively he must know that anything which is called “Alfred”, even something quite unknown to him, is an Alfred’ (Hornsby 1976, p. 233). The third virtue of the theory is that it meets the stated condition of adequacy. A name is strictly true of many individuals – all those given that name in an appropriate way – and which individual is referred to by a singular use of a name will depend on context in just the same way as the referent of a use of a demonstrative expression will depend on context. In order that the condition of adequacy be met, this context-dependence of the objectlanguage cannot be carried over into the meta-language, since this would yield a theory of truth that was not fully formalized; but the context-dependence of the object-language can be accounted for in the meta-language by means of a set of reference clauses that determine which object is referred to by a singular use of a name by a speaker at a time, as follows: (x)(y)(Reference (x), & By(x,p) & At(x,t) & With(x, ‘Aristotle1’, ‘Aristotle is human’) & To(x,y) → (‘Aristotle is human’ is true with respect to p at t ↔ Human ([y] Aristotle (y)))) Read: For all x and y, if x is an act of reference by person p at time t to y with the first occurrence of ‘Aristotle’ in ‘Aristotle is human’, then ‘Aristotle is human’ is true with respect to p at t just in case the object which is y and is an Aristotle is human. (Burge 1973, p. 433) In this chapter – and against the background of these three virtues – I defend the modified predicate view. In Section 2, I rehearse briefly the arguments that favour the modified predicate view over both the traditional predicate view and the individual constant view. In Section 3, I consider a series of

Modified Predicate Theory of Proper Names

209

objections that have been levelled against the modified predicate view and argue that none is successful. Finally, in Section 4, I outline two further benefits of the view: one concerns the prospects for a unified treatment of singular thought; the other concerns Kripke’s puzzle about belief. I conclude briefly in Section 5.

2 The alternatives The modified predicate view is to be distinguished from the traditional predicate view according to which a name is a definite description. There are three main versions of the traditional predicate view. The first is the Russellian view according to which a name abbreviates a complex definite description: thus ‘Aristotle’ is taken to be semantically equivalent to a definite description such as ‘the pupil of Plato and teacher of Alexander the Great’.7 The second is the Quinean view according to which a name is a definite description involving a manufactured predicate: thus, ‘Aristotle’ is taken to be semantically equivalent to ‘the aristotelizer’.8 According to the third, meta-linguistic version, a name is taken to be a meta-linguistic definite description: thus ‘Aristotle’ is taken to be semantically equivalent to ‘the individual named “Aristotle” ’.9 As Burge points out, each version of the traditional predicate view violates intuition on a number of counts: (i) the Russellian version violates the intuition that names do not describe the objects they name; (ii) all three versions render names incomplete symbols and thereby violate the intuition that names play the semantical and grammatical role of singular terms; and (iii) all three versions violate the intuition that some sentences that involve failures of designation are neither true nor false. The modified predicate view, in contrast, can accommodate all three of these intuitions. Although names are predicates, the only descriptive content they have is that the individuals of which a given name is true bear that name, which is uncontroversial. Second, names in their singular use (being demonstrative-predicate expressions) remain singular terms semantically and grammatically.10 Finally, the view allows for sentences containing singular terms that fail to designate an object to be neither true nor false, as can be seen by (T3).11 Moreover, the modified predicate view can explain how names can be used to designate their bearers rigidly, whereas no version of the traditional predicate view can.12 If a singular unmodified use of a name is a complex demonstrative, then it will rigidly designate the object it designates in virtue of containing an implicit demonstrative which designates rigidly. The definite article, in contrast, does not designate rigidly, and hence neither does a complex term that consists of the definite article and a predicate.13 But there is a more fundamental concern with the traditional predicate view: each version falls foul of one of the three virtues identified in Section 1. I will start by looking at the manufactured predicate view and

210

Sarah Sawyer

the meta-linguistic predicate view, both of which fail to satisfy the formal adequacy requirement. If a name is semantically equivalent to a definite description, then its predicative element must be uniquely satisfied if it is to denote an individual. However, manufactured predicates such as ‘aristotelizer’ and meta-linguistic predicates such as ‘individual named “Aristotle” ’ are general terms that are not uniquely satisfied. Consequently, context must determine which individual a use of a name refers to on a given occasion. This reliance on context is appropriate in the object-language, but in the meta-language there must be no such reliance on context if the condition of adequacy is to be met. However, if names are definite descriptions involving either manufactured or meta-linguistic predicates, then there is no way to provide the required analysis in the meta-language of the role context plays in determining singular reference. The modified predicate view has the resources to yield an analysis of context in the meta-language, and hence has the resources to satisfy the condition of adequacy, because it treats singular occurrences of names as involving a demonstrative element which is explicit at the level of logical form. These versions of the traditional predicate view, in contrast, make no allowance for context in the logical form of a name but instead treat a singular occurrence of a name as referring to an individual independent of context and hence neither require nor allow room for a context-dependence in the meta-language. This means that neither the second nor the third version of the traditional predicate view satisfies the condition of adequacy. The concern here arises from combining two factors: (i) a name is a definite description, and (ii) the predicative element of a definite description which is semantically abbreviated by a name is not uniquely satisfied.14 Insofar as the Russellian view accepts (i) and (ii) it is subject to the same concern. The most natural interpretation of the Russellian view, however, rejects (ii) and insists instead that the predicative element of a genuine name be itself uniquely satisfied. As such, the Russellian account satisfies the condition of adequacy but faces a dilemma in virtue of the insistence that the predicative element of a name be uniquely satisfied: either it cannot provide a unified account of singular and predicative uses of names, or it cannot allow for unrestricted quantification over individuals who share a name. Both the manufactured predicate view and the meta-linguistic predicate view can accommodate the predicative use of names in a straightforward fashion precisely because the predicates ‘aristotelizes’ and ‘individual named “Aristotle” ’ can be true of many individuals, even individuals unknown to the speaker (as can ‘is an Aristotle’). Singular uses can then be seen as derived from predicative uses by involving the same predicative element and coupling it with the definite article. In contrast, the Russellian view under consideration treats each singular occurrence of ‘Aristotle’ as abbreviating a definite description involving a complex predicate that is true of only one individual. This leaves only a superficial connection

Modified Predicate Theory of Proper Names

211

between the term ‘Aristotle’ as it applies to Aristotle the ancient philosopher, the term ‘Aristotle’ as it applies to the neighbour’s cat, and so on. In effect, names must be construed as multiply ambiguous expressions. But if names are multiply ambiguous expressions – to be disambiguated in the meta-language in order to meet the condition of adequacy – then there is no sense to be made of quantification over individuals who share a name, since no two individuals do share a name. There is then no account of nonsingular uses of proper names, and hence no unified account of singular and non-singular uses of proper names. It might be thought that this objection could be overcome by treating nonsingular uses of a proper name by a speaker at a time as constructed from the set of singular uses of same-sounding names available to that speaker at that time. The basis of the construction would have to be restricted in this way to singular uses of same-sounding names by a speaker at a time since the relevant complex predicates involved in the uses of same-sounding names by every speaker would outstrip the semantic knowledge of any individual speaker and hence could not be employed in stating a truth-theory for that speaker at that time. However, the necessary restriction of the construction base would render the quantificational domain of plural occurrences of names different for each individual speaker and different from the intuitively correct domain, which is not restricted in this way. Consequently, the Russellian view that treats predicative occurrences of names as constructed from the set of same-sounding names available to a speaker at a time can provide a unified treatment of proper names if and only if it assigns incorrect truth-conditions to sentences containing plural and modified uses of proper names. The difference between intuitive truth-conditions and those assigned to utterances by the theory is brought out particularly well by sentences such as (9). A speaker uttering (9) intuitively says something coherent and true. However, if a non-singular use of a proper name by a speaker at a time is assumed to be constructed from the set of same-sounding singular names available to her, then (9) must be assigned truth-conditions which render any given utterance of it false, and will portray the speaker as verging on incoherence. Effectively, (9) as uttered by S at t would be true just in case the set of Alfreds to whom S can effect singular reference at t (because S has a proper name for each of them at t) contains a significant proportion of Alfreds to whom S will never effect singular reference (because she will never have a proper name for them). To avoid this unwelcome consequence, a non-singular use of a proper name by a speaker at a time would have to be treated as entirely distinct from any proper names conceived as singular. On this understanding, (6) and (7) would involve invalid inferences. The view in any case has an artificiality about it since there is no rationale for quantifying over individuals who share a same-sounding name other than by implicit assumption that names that sound the same form a semantic type, which comes close to an admission that names are at root predicates.

212

Sarah Sawyer

The discussion of the traditional predicate view has been complex. One way to understand the fundamental problems facing the traditional predicate view is in terms of a dilemma. Either singular reference is made possible in the object-language by the definite article in conjunction with context, in which case the theory fails to satisfy the condition of adequacy that the theory be fully formalized, or singular reference is made possible by the predicative element of the definite description, in which case there is no unified account of singular and predicative uses of names that yields the correct quantificational domain for predicative uses. Let us then move to a consideration of the individual constants view. In contrast to the traditional predicate view, the view that proper names have the semantical role of individual constants has gained wide-spread support – largely in the form of Millianism, according to which names contribute the objects they denote to propositions expressed by sentences containing them.15 The Millian view (just like the Russellian view) has to treat names as ambiguous in the object-language and index the names in the truth-theory in order to eradicate the ambiguity at that level. Since the semantic value of a name is the object to which it refers, and hence understanding a name consists in knowing its referent, the individual constants view is committed to a specification in the truth-theory of the referent of each of the indexed names.16 As Burge points out, the view has certain immediate drawbacks. First, the number of denotation rules required for any given name is vast in comparison to the single satisfaction rule and set of primitive demonstrative reference clauses required by the modified predicate view. Second, ‘the truth-theorist for the idiolect of a person at a time would be presented with the awesome task of actually tracking down and specifying each of the individuals that a person knows’ (Burge 1973, p. 438). Moreover, the Millian view faces the same dilemma as the Russellian version of the traditional predicate view: if a truth-theory for a speaker at a time contained an axiom for each of the distinct individuals for whom the speaker had a name at that time, then the theory would either fail to provide a unified treatment of proper names, or be unable to accommodate quantification over unnamed or unknown individuals. The concern here is not confined to Millianism but applies more generally to any version of the logical constants view of proper names. For example, consider Larson and Segal’s version of the individual constants view which incorporates a certain predicative element.17 According to their view, which they call ‘FLIC’ (the Free Logic/Individual Concepts view), the axioms in the truth-theory for a speaker at a time would have the following form: (x)(x satisfies ‘Boris Karloff’ iff x = Boris Karloff) (x)(x satisfies ‘Aristotle1’ iff x = Aristotle 1 (the philosopher)) (x)(x satisfies ‘Aristotle2’ iff x = Aristotle 2 (the magnate))

Modified Predicate Theory of Proper Names

213

Although the theory has some benefits over the Millian view – most notably by being able to offer a straightforward account of intensional contexts and empty terms – it faces the same dilemma as the Millian view (and the Russellian view) in virtue of resolving a perceived ambiguity in names by indexing them in the truth-theory. If non-singular uses of a name are to be semantically connected to singular uses, where the singular uses are conceived as primary – as they must be on any individual constants view – then the quantificational domain of non-singular uses will be inappropriately restricted. To illustrate the general nature of the concern, consider Steven Boer’s attempt to provide a unified account of names that takes the singular use rather than the predicative use as primary18 (Boer does not himself endorse the view): (LC1) X is the referent of ‘Alfred’ in (1) if X uniquely satisfies the identitycriterion which the utterer of (1) associates on that occasion with the use of ‘Alfred’ in (1). (LC2) X is in the extension of ‘Alfred’ in (2)–(5) if and only if the generalized intersection of the sets of identity-critera which the utterer of (2)–(5) associates with different singular-term-uses of ‘Alfred’ provides a criterion satisfied by X. (Boer 1974, p. 395) The predicative use of ‘Alfred’ is here defined in terms of the referential (i.e., singular) use of ‘Alfred’. Boer goes on to say: If the speaker’s identity-criteria for various singular-term-uses of ‘Alfred’ agree only in requiring that the referent should have gotten the name ‘Alfred’ in some appropriate way, the generalized intersection of such criteria will provide the single requirement that an entity is in the extension of ‘Alfred’ qua predicate just in case that entity got the name ‘Alfred’ in some appropriate way. Hence all of Burge’s examples would be accounted for. (Boer 1974, p. 395) However, while the account suggested by Boer does provide a unified account of proper names, it do so only by assigning incorrect truth-conditions to sentences containing names used predicatively. On this view, the extension of a speaker’s use of ‘Alfred’ is the set of Alfreds for whom the speaker has associated identity-criteria rather than the set of Alfreds simpliciter. The problem arises precisely because of the construction of predicative uses from singular uses, and it is in virtue of this that any theory that treats singular uses of names as primary will be inadequate. I have argued that the Russellian view and the individual constants view face a dilemma: either sentences containing names used predicatively are assigned incorrect truth-conditions or names used predicatively are

214

Sarah Sawyer

semantically unconnected to singular uses. The modified predicate view escapes the dilemma in virtue of treating predicative uses of names as primary and truly general: names are predicates that apply to all individuals who bear the name. More broadly, the only view that possesses all three virtues identified in Section 1 is the modified predicate view.

3 Objections to the modified predicate view In this section I focus on a series of objections that have not yet been responded to in the literature.19 According to the modified predicate view, names are predicates. As such, names can be true of many distinct individuals. And yet a name can on an occasion be used to refer to just one individual. Such singular reference is made possible by the presence of an implicit determiner in each singular unmodified use of a name. According to Burge – as stated in (T2) – the implicit determiner is to be understood as a bare demonstrative.20 If this is right, then (10) and (11) are semantically equivalent: (10) Mary had fish for lunch. (11) That Mary had fish for lunch. James Higginbotham has objected to this alleged semantic equivalence on the grounds that different truth-conditions are to be assigned to (10) and (11).21 He writes: ‘On seeing a woman emerge from the seafood restaurant, and taking her for my friend Mary, I might volunteer either [(10)] or [(11)]. Suppose that the woman is a Mary, but not my friend Mary. Then [(11)] is true if the woman had fish for lunch, but [(10)] is not verified thereby.’ (Higginbotham 1988, p. 36). The objection has led Gabriel Segal to suggest that if names are predicates, the implicit determiner involved in a singular use of a proper name ought to be understood as the definite article rather than as a demonstrative.22 According to this proposed modification, ‘Alfred’ in its singular use is to be understood as equivalent to ‘the Alfred’ rather than, as Burge suggests, ‘that Alfred’. According to Segal this version of the predicate view avoids Higginbotham’s objection. Segal goes on to say: If the implicit determiner is ‘the’ then it is probably a referential ‘the’. That would be the simplest and most natural explanation of the genuinely referential nature of proper names. ... [The view] thus fits best with an ambiguity theory of the definite article, holding that while ‘the’ sometimes functions as a quantifier, it can also work as a kind of demonstrative, much like ‘this’ and ‘that’. (Segal 2001, p. 551) However, if the proposal is to treat the definite article as a demonstrative whenever it plays the role of the implicit determiner in a singular use of a

Modified Predicate Theory of Proper Names

215

name, it is unclear why the implicit determiner shouldn’t rather be treated as a straightforward demonstrative. Moreover, the suggestion that the implicit determiner be the definite article understood referentially opens itself up to Higginbotham’s objection. Since the only evidence Segal offers in favour of preferring the definite article over the demonstrative as the implicit determiner is to avoid Higginbotham’s objection and since the demonstrative version is simpler Burge’s original proposal is to be preferred.23 It remains, then, to address Higginbotham’s objection. Higginbotham is right that in the scenario he describes (10) and (11) might differ in truthvalue. However, this is consistent with the claim that they are semantically equivalent. Let us sharpen the scenario to explain why. Let us call the known Mary ‘Mary K.’ and the Mary seen exiting the seafood restaurant ‘Mary S.’. Further, let us assume that Mary K. did not have fish for lunch while Mary S. did. In such a scenario an utterance of (10) would be false while an utterance of (11) would be true. This is because Mary K. is the referent of ‘Mary’ in (10) while Mary S. is the referent of ‘That Mary’ in (11). Higginbotham’s objection assumes that this difference in reference implies a semantic difference. However, there is an alternative explanation for the difference in reference that is consistent with the claim that names are predicates with a demonstrative element in their singular use. It is commonplace to note that a demonstrative utterance can be true in one context and false in another even though there is no semantic difference between them – the difference in truth-value being due entirely to a difference in context of utterance. Higginbotham’s case, however, is a case where the context of utterance for (10) and (11) is assumed to be the same. If the context of utterance for (10) and (11) are identical and yet the truth-values differ, the difference must be due to a difference in semantic content. Hence Higginbotham’s conclusion. However, there is an alternative explanation for the difference in truthvalue consistent with identity of both semantic content and context of utterance. Because the demonstrative element in (10) is left implicit, the name is best understood as referring to the Mary who is conversationally most salient, which may not be the Mary who is perceptually most salient. In contrast, the demonstrative element in (11), being explicit, has the effect that the expression ‘That Mary’ is best understood as referring to the Mary who is perceptually most salient. Note that if an utterance of (10) were accompanied by a nod of the head towards Mary S., it would also concern Mary S. and hence be true. The nod of the head here would have the effect that the explicit demonstrative does in (11). In general, the demonstrative element in a singular use of a name will determine a different referent depending on whether it is left implicit or made explicit. One way to put this is to say that which contextual features of an utterance are relevant to determining the referent of the singular use of a proper name will be determined in part by whether the demonstrative element is left implicit or made explicit. As a result, the mere fact that (10) and (11) differ in truth-value does not imply

216

Sarah Sawyer

that they differ semantically even if the context of utterance is taken to be the same. A second objection to the modified predicate view has been raised by Segal.24 Segal argues that singular uses of proper names do not function like ordinary determiner-predicate constructions, which casts doubt on the hypothesis that names are predicates. Thus consider the following: (12) a. b. c. d.

That Churchland argued for materialism. Churchland argued for materialism. That philosopher argued for materialism. *Philosopher argued for materialism.

According to Segal, the sentences in (12) illustrate that the determiner in a [determiner [name]] construction can be dropped, whereas dropping the determiner from a [determiner [common noun]] construction renders a sentence ungrammatical. More precisely, Segal claims that if there is an implicit determiner involved in the singular use of a proper name then first, there must be special rules about when it can be dropped, and second, there must be special rules about when it can appear. Segal offers two examples to establish his claim. The first example runs as follows. In response to the question ‘Where do you live’ posed in a context in which ‘London’ would naturally be interpreted as London, England, (13a) would be appropriate whereas (13b) would ‘sound bizarre’ (Segal 2001, p. 561): (13) a. I live in London. b. ??I live in that London. In contrast, (13b) would be appropriate if uttered in a context in which London Ontario is not out of the question and the speaker could point out of the window of the plane or point at the location on a map. Second, consider the following sentences: (14) a. This is that John I mentioned yesterday. b. *This is John I mentioned yesterday. (14a) is grammatical, whereas (14b) is not.25 It would seem that it is sometimes inappropriate to make the hidden determiner explicit (as in (13b)), and sometimes ungrammatical to leave the implicit determiner implicit (as in (14b)). As Segal says, if there is a hidden determiner involved in a singular use of a proper name there must be ‘special rules governing when it can appear on the surface and when it cannot’ (Segal 2001, p. 561).

Modified Predicate Theory of Proper Names

217

Borrowing from the proposed solution to Higginbotham’s objection, here is a first pass at the special rules. (R1) If the conversational context is such that the referent of a singular use of a name would be evident without an accompanying demonstration, then the implicit determiner should not be made explicit. (R2) If the conversational context is such that the referent of a singular use of a name would not be evident without an accompanying demonstration, then a demonstration should be provided, whether or not the implicit determiner is made explicit. Example (13) is explained by (R1), and example (14) is explained by (R2). Finally, although (12a) and (12b) should be understood as grammatical and as semantically equivalent, each will be appropriate in different circumstances: (12a) would be appropriate if both Paul and Patricia Churchland were conversationally salient, but (12b) would be appropriate if only Paul Churchland, for instance, were. This is, I think, how the sentences are most naturally read.26 Finally, let us turn to four considerations against the modified predicate view offered by Jeff King.27 King’s first and second considerations are offered as evidence for the claim that singular uses of names differ semantically from predicative uses of names and hence that names are not predicates in all their uses. His first consideration relies on the claim that singular and predicative uses fall into different syntactic categories: names function as noun phrases in their singular use and count nouns in their predicative use. This is reflected, he claims, in the different proforms in each of the following28: (15) Many Jeff Kings live in LA and Oriana knows one of them. (16) Jeff King lives in LA and Oriana knows him. However, this consideration begs the question against the modified predicate view by assuming that the grammatical subject of (16) is a bare name and not a complex demonstrative. According to the modified predicate view ‘Jeff King’ as it occurs in (16) functions as a predicate and a count noun, just as it does in (15). Syntactically, ‘Jeff King’ in (16) is analogous to ‘that man’ in (17): (17) That man lives in LA and Oriana knows him. Consequently, an alleged semantic difference cannot be grounded in an alleged syntactic difference: both differences would be rejected by the modified predicate theorist.

218 Sarah Sawyer

Second, King claims that singular and predicative uses of names have different modal profiles. Consider the following two sentences: (18) Doug Stanley lacks a name. That’s false but it might have been true. (19) Every Doug Stanley lacks a name. That’s false but it might have been true. King claims that intuitively the second sentence of (18) is true whereas the second sentence of (19) is false. Unfortunately he does not explain his intuition, which I do not share. Intuitively, I would maintain, the second sentence of (19) is also true. As yet, then, there is no reason to think that singular and predicative uses of names differ in their modal profile and hence fall into different semantic categories. King’s third and fourth considerations are intended to discredit the claim that names in their singular use involve an implicit determiner. According to King, (20) is grammatical while (21) is not: (20) That Glenn Bunting is happy but this one isn’t. (21) *Glenn Bunting is happy but this one isn’t. This contrasts with (22) and (23), both of which are grammatical: (22) Dogs are kind. (23) Dogs are kind but this one isn’t. The fact that (22) and (23) are grammatical supports the claim that generic count nouns are fronted by an implicit determiner, whereas the fact that (21) is ungrammatical tells against the claim that names in their singular use are fronted by an implicit determiner. However, the example is of the same kind as Segal’s examples discussed above and is to be explained in a similar fashion by invoking rules (R1) and (R2).29 King’s final argument against the modified predicate view concerns designation. He claims that ‘a “bare” (singular) name in subject position can designate a particular individual in the extension of the name qua count noun’ (King 2006, p. 149), whereas other count nouns cannot designate particular individuals in their extension. The difference is brought out by the following: (24) Sarah Sawyer lives in Sussex. (25) Dog is kind. The term ‘Sarah Sawyer’ as it occurs in (24) can be used, according to King, to designate a particular Sarah Sawyer, but the term ‘dog’ in (24) cannot be used to designate a particular dog. There are two steps to the response here. First, a “bare” (singular) name in subject position cannot on its own designate a particular individual in

Modified Predicate Theory of Proper Names

219

its extension. Names do not designate their bearers; they are true of their bearers. Designation is a complex relation consisting of the true of relation, which is semantic, and speaker reference, which is pragmatic. Thus ‘Sarah Sawyer’ as it occurs in (24) does not designate an individual any more than ‘dog’ in (25) does. There is a still a disanalogy between (24) and (25) because a speaker can refer to a particular individual by the use of a bare singular name, whereas a speaker cannot refer to an individual by the use of any other kind of predicate except by adding an explicit demonstrative, as in ‘that dog’. However, the disanalogy can now be seen merely as an instance of the disanalogy identified by Segal above: the determiner that accompanies a singular use of a name can be left implicit while the determiner that accompanies the singular use of any other kind of predicate cannot. And once again, that is to be explained by (R1) and (R2). No doubt there is more to be said. However, if the demonstrative element in a singular use of a name can determine a different referent depending on whether it is left implicit or made explicit, then the modified predicate view of names can withstand Higginbotham’s objection; if something like (R1) and (R2) are correct, then the modified predicate view of names can withstand Segal’s objection and King’s objections.

4 Two benefits of the modified predicate view The modified predicate view has two further advantages: first, it promises a unified account of singular (de re) thought; second it offers a straightforward solution to Kripke’s puzzle about belief. I will consider each advantage in turn. There is an intuitive distinction between thoughts that are de re – ones that relate a thinker to an object directly – and thoughts that are de dicto – ones that relate a thinker to an object if at all only indirectly. De re thoughts are generally thought to be expressed by sentences that contain either demonstratives or singular uses of proper names. A unified semantic treatment of demonstratives and proper names, then, would provide a unified account of de re thought by providing an explanation as to why sentences containing these two kinds of expressions are alike in expressing de re thoughts. Demonstratives and proper names are typically given different semantic treatments – demonstratives are treated as determining a referent only relative to a context, whereas names are treated as singular terms and hence as determining a referent independently of an application in a context. If a unified semantic treatment of demonstratives and names is to be forthcoming, then, either demonstratives will need to be treated as context-independent or proper names will need to be treated as context-dependent. The first option is explored by Evans and McDowell.30 The second option emerges naturally from the modified predicate view of names. This is the option I will urge here.

220

Sarah Sawyer

Consider the following two sentences: (26) That book is tatty. (27) The oldest book is tatty. An utterance of (26) by a subject S1 is naturally taken to express a de re belief, while an utterance of (27) by that subject is naturally taken to express a de dicto belief. The difference is plausibly due to the presence of a demonstrative in the first utterance and a definite description in the second. One way to represent (26) and (27) that does justice to the difference is by (28) and (29) respectively31: (28) Br (S1, ,  x is tatty ) (29) Bd (S1,  the oldest book is tatty ) In this notation, ‘Br’ represents that the subject has a de re belief, and ‘Bd’ that the subject has a de dicto belief. The pointed brackets in the de re case mark out the object or objects the belief is about,32 and serve to indicate that it is the object or objects themselves which are of importance, rather than any description of them. There is no equivalent in the representation of a de dicto attribution. The corner quotes, in contrast, denote the proposition (in the de dicto case) or propositional fragment (in the de re case) expressed by the terms enclosed. (28) and (29) embody a certain conception of the de re/de dicto distinction: de re attributions relate the subject in part to an incomplete proposition expressed by an open sentence and in part to a res, whereas de dicto attributions relate the subject to a complete proposition, expressed by a closed sentence; and de re thoughts are those which require a non-conceptual contextual relation to determine the object or objects they are about, whereas de dicto thoughts are those which are fully conceptualized. This conception fits well with our understanding of demonstrative thought. On this understanding, demonstrative thoughts can be conceptually identical (identical in content) and yet be about distinct individuals because the conceptual elements of a demonstrative thought do not determine an object of thought independently of context. In contrast, individual constants and Russellian definite descriptions serve to single out an object independently of context. Consequently, demonstratives and names thus conceived do not form a unified semantic group. This means that if sentences containing demonstratives and names are alike in expressing de re thoughts, there is no semantic unity to the class of terms that serve to express de re thoughts. That is, if names are conceived as individual constants or Russellian definite descriptions, the explanation for why a sentence containing a name expresses a de re thought must be different from the explanation for why a sentence containing a demonstrative expresses a de re thought.

Modified Predicate Theory of Proper Names

221

If, on the other hand, a demonstrative element in singular uses of proper names is recognized, the problem is overcome. This is precisely what the modified predicate view provides. Consider again sentence (1): (1) Alfred studies in Princeton. If the modified predicate view is correct, this sentence is incompletely interpreted: it has a truth-value only relative to a context of application in much the same way as the explicitly demonstrative sentence (26). Such sentences can be used to express thoughts about different individuals on different occasions. Which thought is expressed by (1) will depend on which Alfred (if any) is contextually identified. Thus S1 might utter (1) referring to Alfred Smith, S2 might utter (1) referring to Alfred Jones, and S3 might utter (1) attempting to refer to her imaginary friend. Assuming our earlier notation, the correct attributions of thought would then be as follows: (30) Br (S1, ,  x is an Alfred and y is a Princeton and x studies in y ) (31) Br (S2, ,  x is an Alfred and y is a Princeton and x studies in y ) (32) Br (S3, < , Princeton>,  x is an Alfred and y is a Princeton and x studies in y )33 The conceptual elements of the thoughts thereby attributed no more single out a particular Alfred (or a particular Princeton) than the conceptual elements of the thoughts expressed by demonstrative utterances single out a particular object of thought: context is needed in each case. If de re attitudes are taken to be those that are not fully conceptual, and the modified predicate view is correct, then both demonstrative utterances and sentences containing singular uses of proper names express de re thoughts alike in virtue of containing a demonstrative element. Names and demonstratives will, then, fall within the same general semantic category. Consequently, the modified predicate view provides a unified theory of de re thought in virtue of providing a unified semantic theory of demonstratives and singular occurrences of proper names. The second benefit of the modified predicate view is that it provides a straightforward solution to Kripke’s puzzle about belief.34 Kripke presents a case in which Peter has heard of Paderewski the musician and of Paderewski the statesman, but believes incorrectly that they are distinct individuals. Thinking he is talking about two distinct individuals, Peter assents to both of the following: (33) Paderewski had musical talent. (34) Paderewski had no musical talent.

222

Sarah Sawyer

If it is assumed that a normal speaker sincerely assenting to ‘p’ expresses the belief that p, then it looks as if Peter has inconsistent beliefs in virtue of the fact that both occurrences of ‘Paderewski’ refer to the same individual. Moreover, reporting Peter’s beliefs in a certain way can render the reporter herself inconsistent in virtue of asserting that Peter both believes that Paderewski had musical talent and does not believe that Paderewski had musical talent. The modified predicate view avoids both troublesome consequences by recognizing a demonstrative element in singular uses of proper names. Thus adopting the notation given above the attribution to Peter of the beliefs he expresses by utterances of (33) and (34) respectively would be: (35) Br (Peter, <Paderewski>,  x is a Paderewski and x had musical talent) (36) Br (Peter, <Paderewski>,  x is a Paderewski and x had no musical talent) On this understanding it is clear that Peter’s thoughts are indeed inconsistent since inconsistent predicative elements are predicated of one and the same individual by Peter. However, there is no inconsistency between the conceptual elements of Peter’s thoughts, which means that there is no irrationality. The conceptual elements of Peter’s thoughts, being de re, function like open sentences and are not truth-evaluable independently of an application in a context. The contextual element that determines that the thought attributed to Peter by (35) is about Paderewski differs from the contextual element that determines that the thought attributed to Peter by (36) is about Paderewski. The fact that singular uses of names involve a demonstrative element means that context is required to determine the object of thought. It is clear that an individual can be related to the same individual via different contextual relations, as Peter is. Indeed, the different contextual relations that obtain between Peter and Paderewski are presupposed in the generation of the apparent paradox. But there is nothing paradoxical in predicating inconsistent properties of an individual if the predications are tied to different contextual applications. In addition, an explicit mention of context in the attributions of thought to Peter ensures that any perceived inconsistency in the attributions is also eradicated. Consequently, the apparent inconsistency both on the part of Peter and on the part of the attributer disappears and the paradox is resolved.35 Kripke is wrong to say that there can be no correct answer to the question, ‘What does Peter believe about Paderewski?’, that does not unjustly impugn Peter’s rationality and logical acumen. Given the modified predicate view, one can say that Peter believes of Paderewski that he is a Paderewski who has musical talent, and that he is a Paderewski who has no musical talent. Peter’s rationality and logical acumen are not thereby impugned because of the implicit demonstrative involved in any singular use of ‘Paderewki’.

Modified Predicate Theory of Proper Names

223

5 Conclusion I have argued that the modified predicate view of proper names is to be preferred over both the traditional predicate view and the more popular individual constant view. It provides a unified account of singular and nonsingular occurrences of proper names; it yields the correct truth-conditions for sentences containing proper names; it provides a unified treatment of singular thought; and it provides a solution to Kripke’s puzzle about belief.

Acknowledgements Many thanks to Ray Elugardo and Michael Morris for helpful comments.

Notes 1. These views are discussed in more detail in Section 2. 2. See Tarski (1956). 3. While it is possible to reject the adequacy constraint, one who does so reject it must think seriously about what it is a semantic theory should do. 4. Examples (1)–(5) are taken from Burge (1973, p. 429). 5. If the modified predicate theory is correct, the following inferences are analogous at the level of logical form to (6) and (7) respectively: (6’) That cow lives in a barn, so at least one cow lives in a barn. (7’) Dogs tend to have fleas, so that dog probably has fleas. The fact that (6’) and (7’) are also intuitively valid provides evidence that proper names function as general terms. Thanks to Ray Elugardo for drawing my attention to this. 6. Examples such as (8) and (9) are given in Hornsby (1976) p. 232. 7. See Russell (1911 and 1918). 8. See Quine (1953). 9. See Russell (1911 and 1918), and, more recently, Bach (1987). 10. Not everyone will accept the orthodox view that demonstratives are singular terms. See, e.g., King (2001). This will also have implications for the claim that demonstratives are rigid. 11. For the details see Burge (1973, section III). 12. See Kripke (1980) for a series of criticisms of predicate theories along these lines. 13. There is a standard way of getting some descriptive theories of names to meet Kripke’s objections about rigidity: actualize the descriptions (‘the person who actually was taught by Plato and taught Alexander’, ‘the actual aristotelizer’, ‘the individual actually named “Aristotle”‘). However, there tend to be problems with such theories in complicated embeddings, for which see Soames (2002) pp. 46–50. 14. The modified predicate view avoids the concern by rejecting (i). 15. See, e.g., Salmon (1986) and Soames (2002). 16. Although the Russellian view must also index names in the truth-theory, the truth-theory need not specify the referent of each name since semantic understanding on this view consists in knowing the predicate that an individual

224 Sarah Sawyer

17. 18. 19. 20. 21. 22. 23. 24. 25.

26.

27. 28. 29.

30. 31. 32.

33. 34. 35.

uniquely satisfies, and hence the referent of each name is determined by a specification in the meta-language of the relevant predicate. See Larson and Segal (1995). Boer (1974, p. 395). For other objections together with responses see Elugardo (2002). See also Recanati (1993). See Higginbotham (1988). The example is his. See Segal (2001). Segal does not himself endorse the view, but suggests the modification as a friendly amendment. On might also refer back to the series of problems connected with the traditional predicate view detailed in Section 2. See Segal (2001) pp. 560–1, from which the examples in the chapter are also taken. There is a parallel between (14a)/(14b) and the following pair: (14c) This is that horse I mentioned yesterday. (14d) *This is horse I mentioned yesterday. The parallel is indirect evidence that proper names function as general terms, since ‘horse’ is clearly a general term. This point is due to Ray Elugardo. It might be thought that examples (13) and (14) are different in character because (13b) is pragmatically inappropriate in the circumstances but grammatical nonetheless, whereas (14b) is simply ungrammatical. This would matter if (R1) and (R2) were rules of linguistic etiquette rather than rules of grammaticality since they would then provide an explanation for the bizarreness of (13b) but not an explanation of the ungrammaticality of (14b). However, I think both examples are best understood as bizarre rather than as ungrammatical. See King (2006). I am heavily indebted to Ray Elugardo for bringing this article to my attention and for helpful discussion of the material. Examples (12)–(13), (18)–(23), and (25) are King’s. Moreover, as Ray Elugardo has pointed out in conversation, it is not clear why (21) should be regarded as ungrammatical. For instance, on being asked whether a particular Glenn Bunting is happy, the parent of a whining child might point to his child while uttering (21). In such a context, the utterance might be true even if the whining child is not a Glenn Bunting. See Evans (1982) and McDowell (1977), (1984) and (1986) for a defence of the view. The understanding of de re attitudes that follows, together with the accompanying notation, is due to Burge (1977). ‘Br’ marks a relation between a thinker, a propositional fragment and the members of the ordered n-tuple denoted by the pointed brackets and terms it contains: it does not mark a relation between a thinker, a propositional fragment and an ordered n-tuple. If it did, intuitively empty thoughts would in fact be thoughts about the empty set. This has the result that de re attitudes are multigrade relations. While any utterance of (1) is grammatical, some utterances will express incomplete thoughts which are neither true nor false, as is indicated by (32). See Kripke (1979). There is a worry here that a form of Kripke’s puzzle can be constructed for any expression which someone can count as understanding while mistakenly supposing to be ambiguous. If that is right then more will need to be done to solve Kripke’s puzzle: the proposed solution given here depends upon identifying a

Modified Predicate Theory of Proper Names

225

demonstrative element in proper names, but the proposal that there is a demonstrative element in predicates seems unwelcome.

Bibliography Bach, K. (1987) Thought and Reference (Oxford: Oxford University Press). Boer, S. (1974) ‘Proper Names as Predicates’, Philosophical Studies, 27, 389–400. Burge, T. (1973) ‘Reference and Proper Names’, The Journal of Philosophy, 70, 425–39. —— (1977) ‘Belief De Re’, The Journal of Philosophy 74, 338–62. Elugardo, R. (2002) ‘The Predicate View of Proper Names’, in G. Preyer and G. Peter (eds) Logical Form and Language (Oxford: Clarendon Press). Evans, G. (1982), The Varieties of Reference (ed.) J. McDowell (Oxford: Oxford University Press). Higginbotham, J. (1988) ‘Contexts, Models, and Meanings: A Note on the Data of Semantics’, in R. Kempson (ed.) Mental Representations: The Interface between Language and Reality (Cambridge: Cambridge University Press). Hornsby, J. (1976) ‘Proper Names: A Defence of Burge’, Philosophical Studies, 30, 227–34. King, J. (2001) Complex Demonstratives (Cambridge, MA: MIT Press). —— (2006) ‘Singular terms, Reference, and Methodology in Semantics’, Philosophical Issues, 16, Philosophy of Language, 141–61. Kripke, S. (1979) ‘A Puzzle about Belief’, in A. Margalit (ed.) Meaning and Use (Dordrecht: D. Reidel). —— (1980) Naming and Necessity (Cambridge, MA: Harvard University Press). Larson, R. and Segal, G. (1995) Knowledge of Meaning: An Introduction to Semantic Theory (Cambridge, MA: MIT Press). McDowell, J. (1977), ‘On the Sense and Reference of a Proper Name’, Mind, 86, 159–85. —— (1984), ‘De Re Senses’, in C. Wright (ed.) Frege: Tradition and Influence (Oxford: Basil Blackwell). —— (1986), ‘Singular Thought and the Extent of Inner Space’, in P. Pettit and J. McDowell (eds) Subject, Thought and Context (Oxford: Clarendon Press). Quine, W.V.O. (1953) ‘On What There Is’, in his From a Logical Point of View (New York: Harper). Recanati, F. (1993) Direct Reference: From Language to World (Oxford: Blackwell). Russell, B. (1911) ‘Knowledge by Acquaintance and by Description’, Proceedings of the Aristotelian Society, 11, 108–28. —— (1918) ‘The Philosophy of Logical Atomism’, in R.C. Marsh (ed.) Logic and Knowledge (London: Macmillan). Salmon, N. (1986) Frege’s Puzzle (Cambridge, MA: MIT Press). Segal, G. (2001) ‘Two Theories of Names’, Mind and Language, 16, 547–63. Soames, S. (2002) Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (Oxford: Oxford University Press). Tarski, A. (1956) ‘The Concept of Truth in Formalized Languages’, in his Logic, Semantics, Metamathematics (New York: Oxford University Press).

11 Introduction, Transmission, and the Foundations of Meaning Jeff Speaks

1

Introduction

In his 1956 correspondence with Roderick Chisholm, Wilfrid Sellars held the view that ‘the metalinguistic vocabulary in which we talk about linguistic episodes can be analyzed in terms which do not presuppose the framework of mental acts’, in part because ‘the categories of intentionality are nothing more nor less than the metalinguistic categories in terms of which we talk epistemically about overt speech as they appear in the framework of thoughts’.1 The past half century has not been kind to this view. Instead, the vast majority of philosophers of mind and language have sided with Chisholm, who held that we can ‘explicate the intentional characteristics of language by reference to believing and to other psychological attitudes’.2 By far the most widely held and well-worked out views of the foundations of facts about the meanings of expressions in public languages take these facts to be reducible to, or constituted by, or explained by, facts about the propositional attitudes – for example, the beliefs, desires, and intentions – of speakers of those languages. Proponents of this sort of view hold to the thesis that facts about mental content are prior to facts about public language meaning; for convenience, I’ll call these views mentalist theories of meaning. An initial question about mentalism asks what this thesis means – can we explain what ‘reducible to’, ‘constituted by’, or ‘explained by’ mean in this sort of case?

2

The analysis of meaning

One reasonably clear answer to this question is that the mentalist claims to provide an analysis of the facts about meaning in terms of facts about mental content – an analysis of what it is for a fact of the former sort to obtain in terms of facts of the latter sort. To give an analysis of this sort is to provide metaphysically necessary and sufficient conditions for an expression 226

Foundations of Meaning

227

to mean such-and-such in terms of some collection of facts about mental content, and to defend the view that these conditions are, in some sense or other, more fundamental or basic than the facts about meaning to be analyzed. I’ll call this view analytic mentalism. So put, the analytic project presupposes rather than provides an understanding of ‘priority’; but this needn’t be too serious a problem. We might have some grip on what explanatory priority means, and be able to recognize cases in which one sort of fact is prior to another, without being able to give an analysis of the relation of explanatory priority itself. Most mentalists have engaged in the analytic project, in this sense. The first question which any analytic mentalist must answer is: Which facts about mental content get into the analysans? The two most well-developed versions of the analysis of meaning in terms of mental content differ in their answers to this question. The first, Gricean analysis of meaning focuses on the communicative intentions of speakers of the language, while the second analysis of meaning focuses on the beliefs of speakers of the language. Though consideration of every variant of these two sorts of analyses of meaning is beyond the scope of this chapter, in what follows I will argue that each of these versions of analytic mentalism faces obstacles whose significance has not been fully appreciated. 2.1 The Gricean program Grice thought that (1) facts about what expressions mean are to be explained, or analyzed, in terms of facts about what speakers mean by utterances of them; and he thought, further, that (2) facts about what speakers mean by their utterances can be explained in terms of the intentions of speakers. These two theses comprise the ‘Gricean program’ for reducing meaning to the contents of the intentions of speakers. In what follows I’ll focus on part (2) of the Gricean program, the attempt to explain speaker-meaning in terms of audience-directed intentions. The basic version of the Gricean analysis of speaker-meaning in terms of the intentions of speakers is the following: [G] a means p by uttering x ⬅ a intends in uttering x that (1) his audience come to believe p, (2) his audience recognize this intention, & (3) (1) occur on the basis of (2)3 One way to see the intuitive motivation behind Grice’s analysis is to begin with the idea that meaning something by an utterance is a matter of trying to convey one’s beliefs. Trying to convey one’s beliefs can be thought of as intending someone to share one’s beliefs; but, fairly clearly, you can intend by an action that someone form a belief that p without meaning p by your action. An example here might help. Suppose I turn to you and say,

228 Jeff Speaks

‘You’re standing on my foot.’ I intend that you hear the words I am saying; so I intend that you believe that I have said, ‘You’re standing on my foot.’ But I do not mean by my utterance that I have said, ‘You’re standing on my foot.’ That is my utterance – what I mean by it is the proposition that you are standing on my foot, or that you should get off of my foot. I do not mean by my utterance that I am uttering a certain sentence. This sort of example indicates that speaker meaning can’t just be a matter of intending to cause a certain belief – it must be intending to cause a certain belief in a certain way. But what, in addition to intending to cause the belief, is required for meaning that p? Grice’s idea was that one must not only intend to cause the audience to form a belief but also intend that they do so on the basis of their recognition of the speaker’s intention. This condition is not met in the previous example: I don’t expect you to believe that I have uttered a certain sentence on the basis of your recognition of my intention that you do so; after all, you’d believe this whether or not I wanted you to. This is all to the good. However, even if [G] can be given a fairly plausible motivation, and fits many cases rather well, ultimately it seems unlikely that there is any connection between speaker-meaning and intended effects stable enough to ground an analysis of the sort that Grice envisaged. To see why, focus first on clause (1) of [G], the requirement that the speaker intend that his audience come to believe the proposition that he means by his utterance. There are many cases in which we speak meaningfully without intending that our audience form the relevant belief, as when we remind someone of something – in such cases, the audience already has the belief, and so can’t come to have it. An initially plausible fix is to change (1) from ‘believes’ to ‘occurrently believes’ or ‘actively believes’. This may help with cases of reminding, but has no obvious application to other counterexamples to (1). As Grice himself noted, instances of confessing something to someone who’s caught you red-handed, or taking an oral exam, are both cases in which we mean something by our utterances without intending that our audience come to occurrently believe the proposition we mean.4 In both cases, the audience already has the relevant occurrent belief. Perhaps the Gricean should modify (1), not by focusing on the distinction between occurrent and non-occurrent beliefs, but rather by changing his view of the content of the belief which the speaker must intend to cause. One idea along these lines is that when I mean p by an utterance, I don’t always intend that my audience come to believe p, but do always intend that my audience come to believe that I believe that p.5 Clause (1) of the account could be modified accordingly. But this doesn’t address obvious variants of the previous cases. Suppose, for example, that my daughter confesses to having given her pet fish too much food. In this case, she knows that I already believe that she gave the fish too much food, and that I already believe that she believes this. (She might also know that I occurrently believe these

Foundations of Meaning

229

things.) Nonetheless, she might still mean by her utterance that she gave the fish too much food. The idea that meaning p is partly a matter of intending that an audience believe that the speaker believes p also leads to some other unattractive consequences. It seems clear that in some cases, I can mean by an utterance that I believe something, without also meaning the proposition which is the content of that belief. For example, if I say ‘I believe that Indiana is the cultural capital of the United States’ I might mean to say something just about my own beliefs; I might have no intention at all of meaning by my utterance that Indiana is in fact the cultural capital of the United States. In this case I will likely intend that my audience come to believe that I believe that Indiana is the cultural capital of the United States. But then I will, on the present view, satisfy the conditions for meaning by my utterance that Indiana is the cultural capital of the United States (since I intend that my audience come to believe that I believe this proposition) – and this contradicts our assumption that I meant by my utterance only that I believe this proposition. The Gricean might at this point turn to more drastic revisions; perhaps the key to speaker meaning is not the intention to bring about a belief, but rather the intention to bring about some other sort of mental state. For example, perhaps, as Neale (1992) has suggested, the relevant intentions are intentions to cause agents to entertain certain propositions. There are two problems here. The first is that the modification leads to yet more counterexamples. Consider, for example, utterances of disjunctions. Speaker-meaning clearly does not distribute over disjunction – I can mean by my utterance that either arithmetic is incomplete or the moon is made of cheese without meaning by my utterance that the moon is made of cheese. However, I might never intend someone to entertain a disjunction without intending them to entertain the disjuncts – I might, after all, realize that it is impossible for someone to do the former without also doing the latter.6 The second problem is that this modification to clause (1) makes clause (3) impossible to satisfy in many cases. Usually when you hear a certain sentence which in the context means p, you immediately, almost reflexively, entertain the proposition p; this is the condition that normal speakers are in with respect to sentences they understand. Suppose, I, the speaker, know this, then I can hardly intend that you come to entertain p on the basis of your recognition of my intention when I utter this sentence, as clause (3) requires – I know that you will entertain this proposition irrespective of your recognition of this intention. The Gricean might reply (as Neale does) that clause (3) is in any case an implausible condition on speaker-meaning, so that dropping this clause from the account is not much of a cost. To get a sense of the problems caused by (3), consider Grice’s own example of an utterance of the conclusion of an argument. Clause (3) of [G] requires for a speaker to mean p not only that she intend that her audience come to believe p and that her audience

230 Jeff Speaks

recognize this intention, but also that her audience come to this belief on the basis of that recognition. The problem is that, when giving an argument of some kind, a speaker will typically intend that her audience come to believe the conclusion on the basis of belief in the premises, and not on the basis of a recognition of the speaker’s intention.7 Further, if we pursue this suggestion and drop clause (3) from the Gricean analysis, this effectively reduces speaker meaning to the intention to simultaneously cause two effects in an audience: a mental state, and the belief that the speaker intends that the audience come to be in that mental state. But, as an example discussed previously shows, this can’t be right. Remember the case of my utterance of ‘You’re standing on my foot.’ As noted before, given that I want you to hear me, I will intend you to believe that I am uttering this sentence. But given that I also want you to know that I am uttering this sentence intentionally, I’ll want you to recognize that I intend that you believe this. But this means that the proposition that I have uttered the sentence ‘You’re standing on my foot’ is such that (1) I intend that you believe it, and (2) I intend you to recognize this intention. If this were enough for speaker meaning, I would mean by my utterance that I have uttered the sentence, ‘You’re standing on my foot.’ But I don’t – I mean that you are standing on my foot and that you should get off my foot, but I don’t mean anything about my own utterances. So again the Gricean faces a dilemma: given the present problem, condition (3) must be included in [G]; but, as noted before, if (3) is included, the possibility of meaning something by statements of the conclusions of arguments looks puzzling. This dialectic of revision and counterexample could be continued.8 And in fact there is a persistent thought to the effect that, even if we have not yet arrived at the right form of the Gricean account, some true version is out there to be discovered. The discussion of the problems raised by revising clause (1) of Grice’s original account has been designed to discourage this thought: each counterexample to a version of the Gricean analysis leads to a revision, which in turn opens the door to new counterexamples. We don’t have here a progressive narrowing of problem cases; rather, each version of the analysis creates as many problems as it solves. For this reason, even the range of cases considered so far seems to indicate that while (of course) in many cases there is a connection between what a speaker means by an utterance and the mental states they intend to cause in their audience, there is no connection of this sort stable enough to ground an analysis of speakermeaning in terms of intended effects. Nonetheless, it would be good to have a general argument against Gricean analyses, rather than a piecemeal argument by counterexamples. I think that such an argument can be given, by exploiting a variant on Moore’s Paradox. Moore drew attention to the oddness of uttering sentences of the form [M] S, but I do not believe that S.

Foundations of Meaning

231

While there are contexts with respect to which instances of [M] are true, it is, as Moore pointed out, difficult to imagine a scenario in which a speaker who understands this sentence could utter it sincerely and seriously. The same can be said of a similar sort of sentence, which replaces ‘do not believe’ with ‘do not intend that you will believe’: [I] S, but I do not intend you to believe that S. As with Moore’s original paradoxical sentence, it is not easy to imagine cases in which a speaker could naturally use a sentence of the form of [I] seriously and sincerely in conversation. But there clearly are cases in which a speaker could utter a sentence which entails an instance of [I] seriously and sincerely. Consider for example, sentences of the form [K] S, but I know that you will not believe that S. Unlike the first two, it is easy to imagine cases in which instances of [K] are uttered seriously and literally. Consider, for example the following: An unfaithful husband is found out by his wife; he might deeply regret his act and love his wife, but know that there is no way to convince her of this. He says to her, ‘I love you, though I know you won’t believe that.’9 Not only might the husband say this seriously and literally; what the husband says might well be true: it might well be the case that he really does love her, and also be the case that he knows that she will not believe that he does. What follows from the fact that the husband is speaking seriously and literally, as well as truly? From the fact that he is speaking seriously and literally it follows that among the things he means by his utterance of this sentence is the proposition expressed by the sentence in the context: the proposition that he loves his wife, and knows that she will not believe that he loves her. Moreover, it seems that speaker-meaning distributes over conjunction;10 it is impossible for a speaker to mean that p and q without also meaning that p. (Just as it is impossible to assert, or believe, a conjunctive proposition without asserting, or believing, the conjuncts.) From this it follows that the husband not only means by his utterance that he loves his wife and knows that she will not believe that he loves her, but also (the first conjunct) that he loves her. So, (a) The speaker means that he loves his wife by the utterance. But recall that in the previous case the speaker not only spoke seriously, but also truly. So the conjunctive proposition that he loves his wife, but knows

232 Jeff Speaks

that she will not believe this, is true. So both of the conjuncts are true, and, in particular, the second is true: (b) He knows that his wife will not believe that he loves his wife. The problem for the Gricean is that it seems plausible that knowing that some event will not occur as the result of one’s action seems inconsistent with intending that the event occur: one cannot intend to bring about effects which one knows will not obtain.11 So, from (b) it follows that (c) He does not intend that his wife will believe that he loves his wife. But from (a) and (c) it follows that there is some proposition – namely, the proposition that he loves his wife – which is such that the speaker means this proposition by his utterance, and yet he does not intend that his audience believe this proposition. But the Gricean analysis entails that there can be no proposition with both of these properties; so the Gricean analysis must be false. Of course, we have by now discussed a number of other cases in which are also counterexamples to [G] and its variants – what makes this one special? The interest of the present argument is that it is more than an isolated counterexample: it amounts to a recipe for generating counterexamples in response to various revisions of the Gricean analysis. For example, suppose that a Gricean changes the propositional attitude specified by clause (1) of the analysis to some other propositional attitude, R. Then all that is needed is an example of a sentence analogous to the form of the one uttered by the husband – S, but I know that you will not believe that S – to one of the form: S, but I know that you will not bear R to the proposition that S. It seems plausible that, for any propositional attitude, we’ll be able to find a sentence and context such that the sentence may be seriously and truly uttered in that context. So this is a class of counterexamples with substantial generality; it seems likely that, whichever audience-directed intention is seized upon by a particular version of the Gricean analysis, we could construct variants of [K] appropriate to refute that analysis which can be uttered seriously and literally, as well as truly, by speakers.12 2.2 Meaning and belief So the Gricean analysis of meaning in terms of speaker-meaning, and speaker-meaning in terms of intentions, faces serious problems. But there’s no reason why the analytic mentalist should have to focus on communicative intentions; the analytic mentalist can reduce meaning to any propositional attitude. And while most attention since the 1960s has focused on the Gricean analysis and its descendants, there is at least one other very natural option for the analytic mentalist to pursue. There seems to be a

Foundations of Meaning

233

tight connection between the meaning of a sentence and the contents of the beliefs that speakers have when they utter the sentence seriously. In particular, it seems to be a necessary truth that if a speaker sincerely utters a sentence in a context in which the sentence expresses a certain proposition, the speaker must believe that proposition. Why not try to use this necessary connection between meaning and belief to analyze facts about linguistic meaning in terms of the contents of the beliefs of speakers?13 The burden of such a theory will be to say what it takes for a sentence to inherit its meaning from the content of a particular belief. The simplest attempt at doing this is to say that a sentence is correlated with a belief just in case were an agent to accept (endorse, assert) the sentence, then the agent would form the relevant belief. A first try at formulating this sort of analysis of meaning might be something like the following: [B] (S means p in a community C) ⬅ ;a(a is a member of C 䊐→ (a accepts S ; 䊐→ a believes p)) There are delicate issues about the interpretation of the counterfactual on the right-hand side of [B]. It is tempting to argue that it is falsified by sentences like ‘0 = 1’ which are so obviously false that everyone who understands them rejects them. Consider the nearest possible world in which I accept the sentence ‘0 = 1’. Isn’t it a world in which the sentence has a different meaning more similar to the actual world than one in which I believe, absurdly, that zero and one are identical? If so, [B] yields the incorrect result that ‘0=1’ does not express the proposition that zero equals one. However, it is plausible that counterexamples of this sort can be blocked by stipulating away ‘backtracking’ evaluations of the counterfactual in [B], in which one rejects a counterfactual on the grounds that, had the event mentioned in the antecedent of the counterfactual occurred, something in the past (relative to that event) would also have been different, where this latter makes the consequent of the counterfactual false.14 The real problem with [B] is that facts about the meanings of sentences are, in a certain sense, more fine-grained than facts about the beliefs of agents; and this makes it difficult to give an account of the former in terms of the latter. Belief has the feature attributed previously to speaker-meaning: it distributes over conjunction.15 Now consider a conjunctive sentence S whose meaning is the conjunction of the propositions p and q. Then each of the following claims will be true: ;a ((a is a member of C) 䊐→ (a accepts S ;→ a believes p & q)) ;a ((a is a member of C) 䊐→ (a accepts S ;→ a believes p)) ;a ((a is a member of C) 䊐→ (a accepts S ;→ a believes q))

234 Jeff Speaks

The problem is that it follows from [B], along with the truth of these three claims, that S is three ways ambiguous: S means p & q, S means p, and S means q. After all, if [B] is to be true, then anyone who accepts S must believe the conjunctive proposition p & q; but the distribution of belief over conjunction entails that they will also believe p and believe q. So all three beliefs are correlated with the conjunctive sentence; and if meaning is defined in terms of correlated beliefs, the sentence has as its meaning each of these three propositions. But this is incorrect: conjunctive sentences are not, in general, three ways ambiguous just by virtue of being conjunctions.16 While this does show that [B] is false, it seems that there is an easy way to revise it in answer to this objection. We can limit the account of meaning to non-conjunctive sentences, and give a separate account of the meanings of conjunctive sentences in terms of the meanings of their parts. But this is only a superficial solution; there are counterexamples similar in form to those based on the distribution of belief over conjunction even for simple sentences. To generate such a counterexample, one needs only a sentence S and two non-conjunctive propositions p, q such that every speaker in the relevant community who would believe p upon accepting S would also believe q. For simplicity, I’m focusing on cases in which believing q is a metaphysically necessary consequence of believing p; but the class of pairs of propositions which can generate counterexamples to [B] is much larger than this. But even if we restrict ourselves to cases in which having one belief strictly entails having another, there are plenty of plausible examples. Consider, for example, the proposition that Lassie was a brown dog (or collie, or puppy) and the proposition that Lassie was a dog. Is it possible for an agent who accepts, and therefore understands, the sentence ‘Lassie was a brown dog’, and who believes that Lassie was a brown dog, not to believe that Lassie was a dog? If not, [B] entails that ‘Lassie was a brown dog’ is ambiguous between meaning that Lassie was a brown dog and meaning that Lassie was a dog; this is clearly incorrect. Colours provide a similar example – is it, e.g., a necessary truth that if someone believes that an object is scarlet, she also believes that the object is red? If so, then we can generate a similar counterexample in the obvious way. Another such class of cases exploits the trivial equivalence – paradoxical cases aside – of a proposition p and the proposition which attributes truth to p. Suppose that an agent accepts the sentence ‘It is true that Lassie was a dog’, and so by the disquotational principle believes that it is true that Lassie was a dog. Is it possible that such an agent could yet fail to believe that Lassie was a dog? Again, it seems unlikely. One response to these kinds of cases is to try to come up with some criterion for separating simple from complex sentences, and to hold that (i) [B] is only applicable as a theory of meaning for simple sentences, (ii) the meanings of complex sentences are a function of the meanings of the simple sentences, and (iii) all the previous cases turn on illicit applications of

Foundations of Meaning

235

[B] to complex sentences. This strategy faces two main challenges. First, we need some criterion which classes together all of the sentences which generate counterexamples of the previous form; second, we need this criterion to be loose enough to count as simple enough sentences to provide the resources to supply the missing theory of meaning for complex sentences. The prospects for finding a criterion to meet both constraints do not appear to me to be good.17 It is tempting to try to solve this problem by appealing to a distinction between beliefs gained ‘directly’ as a result of accepting a sentence, and beliefs gained ‘indirectly’. In one sense, this is clearly correct; if S means p, then there is, intuitively, a more direct link between the acceptance of S by an agent and that agent’s believing p than between the agent’s acceptance of S and the agent’s coming to believe any other proposition. The problem, though, is not to find intuitions of directness; the problem is to explain what these intuitions are tracking in a noncircular way. One can’t, after all, appeal to causal differences; it is not as though, in the sorts of cases described before, the agent comes to believe p and that an effect of this is their coming to believe q – in the cases described, the former condition is metaphysically sufficient for the latter.18 It is important here not to be so focused on the counterexamples as to lose sight of the source of the objection. The basic point is that it seems to be in the nature of belief that facts about the beliefs of agents come, so to speak, in clumps: for many propositions, it is a necessary truth that any agent who believes such a proposition must also believe one or more distinct but related propositions. But facts about meaning just are not ‘clumpy’ in this way: there are no propositions which are such that it is a necessary truth that, if a sentence has it as its meaning, it must have one or more distinct propositions as its meaning as well. Because facts about meaning are, in this sense, more fine-grained than facts about belief, it is difficult to see how we could give an account of the former in terms of the latter.19

3 Supervenience and the priority of thought Of course, further variations on [G] and [B], our two versions of analytic mentalist theories of meaning, are possible. But at this point I want to shift attention to another sort of route that the mentalist might take. One might well think that the mentalist as such needn’t be overly worried by these difficulties with the analytic project – any more than one who thinks that physical facts are explanatorily prior to mental facts need be worried, as such, about difficulties in giving an account of what it is to feel pain in terms of purely physical properties. Such cases suggest that we can have explanatory priority without analysis; so why not just hold to the priority of mental content over meaning without holding any commitment to the success of the analytic project?

236

Jeff Speaks

This raises two related questions for the mentalist. First, how can we establish the relevant priority claim without providing an analysis of meaning in terms of facts about mental content? And, second, if the mentalist is not claiming that a given fact about linguistic meaning can be analyzed in terms of facts about mental content, then what does the claim that mental content is prior to meaning amount to? A plausible and popular answer to this pair of questions is that saying that the facts about mental content are prior to the facts about meaning is saying that the facts about meaning are ‘nothing over and above’ the facts about mental content, and that we can argue for this by showing that the facts about meaning supervene (in a sense to be specified) on the facts about mental content. Let’s call this loosely defined position supervenience mentalism. What sort of supervenience of meaning on mental content would be sufficient to show the priority of the latter over the former? Let’s define the supervenience of one class of properties on another as follows: The A properties supervene on the B properties ⬅ df for any two worlds w and w*, every one-to-one mapping of individuals in w and w* such that each individual in one world alike with respect to the B-properties of the individual with which it is paired in the other world is also such that the paired individuals are alike with respect to their A-properties.20 Clearly, we can’t say that supervenience in this sense is sufficient for priority; priority is an asymmetric relation, but supervenience in this sense is not. But we can use the preceding definition of supervenience to formulate an asymmetric relation which might appear to be a good indicator of priority: Asymmetric Supervenience The B properties are prior to the A properties iff (1) the A properties supervene on the B properties, and (2) the B properties don’t supervene on the A properties. Intuitively, the idea is that once we fix the B-properties in a world, we’ve thereby determined the distribution of A-properties, but that fixing the distribution of the A-properties still leaves open some questions about which individuals instantiate which B-properties. This seems to afford the B-properties a kind of privileged position with respect to the A-properties; it seems to show that the B-properties are at least in part independent of the A-properties, whereas the A-properties are determined by the B-properties. Moreover, it seems that if the Asymmetric Supervenience Test is a good one, things look pretty good for supervenience mentalism. Remember, the supervenience mentalist claims that facts about the beliefs, intentions, desires, and

Foundations of Meaning

237

other propositional attitudes of subjects are prior to facts about what the words of their language mean. So, let’s focus on the following two sets of properties: MC: properties of standing in certain propositional attitude relations, like believing, desiring, or intending, to contents. S: properties related to the semantics of one’s language, like speaking a language in which such-and-such expression has such-and-such a meaning. It certainly seems as though S asymmetrically supervenes on MC. It is quite difficult to see how two worlds could be alike with respect to all of the intentions, beliefs, and other mental states of every individual, but different with respect to the meaning of some linguistic expression; but it’s easy to imagine a pair of worlds mentally alike but for the fact that in one Bob has a passing desire for a cupcake which are nonetheless identical with respect to the semantic properties of the expressions of every language spoken in those worlds. But the Asymmetric Supervenience Test can’t, as it stands, be sufficient for priority. In most cases in which we’re interested, the relevant supervenience relations will not hold between a pair of properties, but between a pair of classes of properties. For example, in the present case we’re interested in the supervenience relations between mental properties and properties to do with the meanings of words in the languages of speakers. But if we take the Asymmetric Supervenience Test as our guide, whether one property is prior to another seems to depend on apparently arbitrary decisions about how we group properties into classes. Suppose, for example, that the A properties asymmetrically supervene on the B properties. Now consider some class C of properties on which the B properties asymmetrically supervene. If the B properties asymmetrically supervene on the C properties, they’ll also asymmetrically supervene on the union of the C properties and the A properties. So are the B properties prior to the A properties, or not? If all we have to go on is the Asymmetric Supervenience Test, then it seems that the B properties are prior to the A properties, but less fundamental then the A properties together with the C properties. But this can’t be right; the facts about what is ontologically prior to what can’t depend on our decisions about which properties to consider together. It is not easy to know how to reform the Asymmetric Supervenience test to block this sort of result. An initial thought is simply to add the following condition to our test: (3) There is no class of properties C such that the B-properties asymmetrically supervene on the union of the A- and C-properties.

238 Jeff Speaks

But this would be much too strong; as long as the B-properties asymmetrically supervene on some set of properties or other, there will be such a class C. A more refined test would be to append, not (3), but (3*) There is no class of properties C such that the B-properties asymmetrically supervene on the union of A- and C-properties but do not asymmetrically supervene on the C-properties alone. The problem for the supervenience mentalist is that, if we add (3*) to the Asymmetric Dependence test, it is not clear that the modified test yields the result that MC is prior to S. After all, her opponent is likely to say that, even if S does asymmetrically supervene on MC, there is a class of properties such that MC asymmetrically supervenes on the union of that set with S, even though MC does not supervene on that set alone. Consider, for example, two further sets of properties: D: dispositional properties, like properties of agents being disposed to perform certain actions. P: properties to do with perceptual representation of one’s environment. One sort of opponent of mentalism is likely to think that MC asymmetrically supervenes on the union of S, D, and P; and this is not, on the face of it, an implausible claim. It is hard to imagine two worlds alike with respect to the distribution of semantic, perceptual, and dispositional properties but which differ with respect to the beliefs, desires, or intentions of individuals in those worlds. If this is right, then we have two supervenience claims: S asymmetrically supervenes on MC MC asymmetrically supervenes on S & D & P There’s no inconsistency in affirming each of these two claims about asymmetric supervenience; but there is an inconsistency in taking each of the two to be priority-revealing. And of course in deciding which, if either, is priority revealing, the Asymmetric Supervenience Test is no help. So, from the supervenience facts alone, it is less than obvious that we get a determinate result as to whether properties in MC are prior to the properties in S.21

4

World history supervenience and world slice supervenience

So far, our conclusions have been negative. The project of analytic mentalism faces serious problems, and the supervenience relations between meaning and mental content considered in the previous section don’t seem to

Foundations of Meaning

239

give us an answer to the ‘priority’ question. I think that we can take some positive steps toward understanding the relationship between meaning and mental content by considering a certain sort of failure of supervenience; or so I’ll argue in this section. Previously we defined the supervenience of A-properties on B-properties as follows: For any two worlds w and w*, every one-to-one mapping of individuals in w and w* such that each individual in one world is alike with respect to the A-properties of the individual with which it is paired in the other world is also such that the paired individuals are alike with respect to their B-properties. Since the properties we’re interested in are ones that individuals have at some times but lack at others, one might wonder how the role of time in this formulation should be understood. Let’s say that an order-preserving mapping of times in w and w* is a one-to-one mapping of times such that for any two times x, y in w, x is earlier than y in w iff the time in w* to which x is mapped is earlier in w* than the time to which y is mapped. Given this, it is natural to understand the one-to-one mapping mentioned in the preceding definition of supervenience as constrained by an orderpreserving mapping of times: when we say that a certain one-to-one mapping of individuals is such that each pair is alike with respect to their A-properties, what we really mean is that a certain one-to-one mapping of individuals is such that each pair is alike with respect to their A-properties at each time, where for them to be alike at each time is for an individual in w to have the same A-properties at t as the individual with whom he is paired in w at the time paired with t by some order-preserving mapping of times. 22 Let’s call this world history supervenience, since to say that one property supervenes on another in this sense is to say that any two worlds alike with respect to the distribution of the subvening class of properties over the course of the history of those worlds are also alike with respect to the distribution of the supervening class of properties over the course of the history of those worlds. But just as we can define a supervenience relation based on world histories, we can also define a supervenience relation based on world slices: the states of possible worlds at particular times. Thinking of a world slice as a pair of a world and a time, we might define one such supervenience relation as follows: World slice supervenience For any two world/time pairs w, t and w*, t*, every one-to-one mapping of individuals in w at t and w* at t* such that each individual in one world is alike with respect to the A-properties of the individual with which it is

240 Jeff Speaks

paired in the other world is also such that the paired individuals are alike with respect to their B-properties. It is possible for one property to supervene on another in the world-history sense without supervening in the world-slice sense. In fact, our two sets of properties – MC and S – are so related. I’ll argue this point first, and then turn to a discussion of its significance for our understanding of the relationship between mental and linguistic representation. Consider the following example: In 2695, a historian of American sports is going through some old magazines in the archives, and comes across a story about someone named ‘Pete Rose’. The story talks about how he was banned from baseball for betting on games, etc. etc. The historian then leaves the archives and says to a colleague, ‘I read about this guy, Pete Rose. He was an interesting character ...’ When the historian utters this sentence, it seems that the name ‘Pete Rose’ has a reference for him, and that it refers to the same man to whom it refers for me and other English speakers familiar with the name: a certain late twentieth-century professional baseball player. And this can be so even if in 2695 there are only a few historians of American sports, and so that, immediately prior to this historian delving into the archives, no one had any thoughts, beliefs, or other propositional attitudes about Pete Rose. However we think of the semantics of names, it seems plausible that if no one has any thoughts about an object, then no one has any thought whose content includes the concept expressed by that name. So, if the name had a reference, and hence a meaning, before being encountered by the historian, it seems that the meaning of the name was not a concept which figured in the thoughts of any person alive at that time. But does the name have a reference before being encountered by the historian? It is hard to see how it could fail to have a reference. After all, if it did not have a reference in, say, 2694, then the historian would be effectively introducing a new name into the language. And it would then be puzzling how the historian could do this: how could the historian be in a position introduce a name which genuinely refers to Pete Rose – as the historian’s uses of the name seem to? (We can imagine that the archives do not provide uniquely identifying information, if that matters.) To put the same point another way: if the name lacked a meaning when the last person with beliefs about Pete Rose died, and it acquired a meaning in 2695, what gave it the meaning it acquired in 2695? It’s hard to see how it could ‘automatically’ reacquire the same meaning it had unless it simply had that meaning all along. But if this is right, then meaning does not supervene in the world-slice sense on mental content. For imagine the world described in 2694, before the discovery. We can imagine a world-slice exactly the same as this one

Foundations of Meaning

241

with respect to the mental representations of each individual in the world, but in which ‘Pete Rose’ either is not a meaningful name, or is the name of someone else. So there can be a difference in the semantic properties of expressions at a time without any corresponding difference in the facts about mental content at that time. This sort of phenomenon can clearly arise with respect to any sort of name, whether of a person, or a place, event, or something else; it also seems to apply to names for properties, and other abstract singular terms. It is also easily extended to the case of natural kind terms; imagine a case like the previous one in which the historians of American sports are replaced by historians of the fauna of North America, and ‘Pete Rose’ is replaced by a term for an extinct species, like ‘passenger pigeon’. The historians could come to have de re thoughts about the passenger pigeon, just as the historians in the previous example could come to have de re thoughts about Pete Rose. In fact, it seems possible to run this sort of example for at least many ordinary predicates. Consider colour words. Suppose that over time humans lose the ability to distinguish shades of red and green, so that eventually all language users are red/green colour blind. The words ‘red’ and ‘green’ might then, let’s imagine, fall into disuse and be forgotten; and eventually it might become the case that no one had any thoughts involving the concepts expressed by these terms. But the terms might then be rediscovered in a way analogous to the rediscovery of ‘Pete Rose’ described previously; those who rediscovered the terms might note that they were used as colour terms in a way similar to words familiar to them, like ‘yellow’ and ‘blue’, and eventually learn about the events which brought about the disappearance of ‘red’ and ‘green’ from the language. Couldn’t these people then use sentences like ‘I wonder which things are red, and which things are green’ to express thoughts which are about the colours, red and green? If so, then the argument runs as follows: because the relevant language users are in no position to introduce a term to refer to the properties red and green, ‘red’ and ‘green’ must have referred to these properties prior to their rediscovery. But if they referred to these properties, then they must have had a meaning which determined these properties as reference. But, since by hypothesis no one has thoughts about these properties at the time t immediately prior to the rediscovery of ‘red’ and ‘green’, no one has thoughts involving the concepts expressed by these terms at t. And so, as mentioned before, it is easy to imagine a world-time pair in which we hold fixed the thoughts of everyone in the world described before at t, but change the reference (and hence meaning) of ‘red’ and ‘green’. To be sure, this sort of case is not as straightforward as the original cases of names and kind terms. But it is anyway a bit awkward to hold that the meanings of ordinary predicates do supervene in the world-slice sense on the facts about mental content, whereas the meanings of names do not. After all, if we think of the intension of a predicate as fixed by a (possibly complex) property, then it is plausible that, for any predicate F, we can introduce an abstract

242 Jeff Speaks

singular term n as a name for the property which determines the intension of F. Then we can use this term to form predicates of the form ‘instantiates n’ which will have the same intension as F. Sophisticated language users apt to form predicates of this sort will of course regard them, correctly, as sharing their intension with the corresponding predicate. But notice that in the previous examples of world slice supervenience failure, it’s not just that no one has any propositional attitudes involving the relevant concept; it’s also true that no one has any propositional attitudes involving any concept which shares its intension with the concept in question. But if this is right, then the failure of world slice supervenience for abstract singular terms entails the failure of world slice supervenience for predicates, given that those predicates share their intensions with predicates formed using abstract singular terms which we already know fail to world-slice supervene on the facts about mental content. So, for at least many sorts of linguistic expressions, their meaning at a time does not supervene on the propositional attitudes of users of the relevant language at that time.23 What does this supervenience failure show us about the relationship between meaning and mental content? For one thing, this supervenience failure amounts to what Brian McLaughlin has called a refutation by appeal to a FIST (false implied supervenience thesis) of standard versions of analytic mentalism.24 Standard analyses of meaning of either the Gricean or belief-based sort analyze the meaning of a word at t in terms of the propositional attitudes of users of the language at t; in this way, at least, such theories have to be revised. Second, and perhaps more importantly, the failure of world-slice supervenience puts some pressure on the supervenience mentalist’s claim that, in some sense or other, the facts about meaning are ‘nothing over and above’ the facts about mental content. After all, in all of the standard examples of cases in which the supervenience of one sort of property on another is taken to show that instantiation of the supervening property is ‘nothing over and above’ instantiation of the subvening property, we have no gap between world-history supervenience and world-slice supervenience. Consider the following examples: The supervenience of the shapes presented on a TV screen on the colours of the pixels. The supervenience of shape properties on microphysical properties. The supervenience of phenomenal character on intrinsic physical properties. The supervenience of dispositional properties on categorical properties. One might of course dispute some of these supervenience claims; but what seems clear is that if these supervenience relations hold in the world-history sense, they also hold in the world-slice sense.

Foundations of Meaning

243

But these negative points leave unanswered the more interesting question: what does the fact that meaning supervenes on mental content over world histories but not world slices show us about the relationship between meaning and mental content? We can get some guidance from other instances of world-history supervenience without world-slice supervenience: The supervenience of the laws in effect in a democracy on the actions of legislators. The supervenience of the facts about artworks on the intentions and actions of artists. Plausibly, two world-slices could be alike with respect to the actions of legislators in those worlds, but different with respect to which laws are in effect; but this will always be due to some difference in legislator actions at some previous time in the history of the world from which the slice is taken. This indicates that the first of these two supervenience claims holds in the world history sense, despite failing in the world slice sense. Parallel remarks apply to the supervenience of art works on the intentions and actions of artists. In each case, the explanation of why world-history supervenience holds but world-slice supervenience does not is fairly obvious. If we ask what it is for a law in a certain sort of democracy to be in effect, our answer has two parts: we must say what it is for a law to be instituted, or put into effect, and what it is for a law to continue to stay in effect over time. And our theories of law-institution and law-transmission will not, of course, be the same; certain actions of legislators are required to institute a law, but legislators need not be constantly engaging in these actions for the law to stay in force. A face-value interpretation of the facts about supervenience of meaning on mental content suggests that our answer to the question, ‘What is it for a word to have a given meaning at a time?’ should also have two non-interchangeable parts: an explanation of what it is for a word to be introduced with a given meaning, and of what it is for it to maintain that meaning over time. This is hardly a new idea. One of the guiding ideas of Naming and Necessity, after all, was that in order to understand the reference of names and kind terms at a time, we need to divide the story about how they acquire a reference into an account of how they are introduced with a given reference, and how they maintain that reference over time and across speakers. And, if we accept Kripke’s views about reference, it seems inevitable that we extend them to the case of meaning. After all, given that meaning determines reference, if some other theory of meaning were available, Kripke’s theory of reference would be dispensable. Given the widespread acceptance of Kripke’s ideas, it is surprising that theories of the foundations of meaning have in general not taken this two-part structure. Perhaps the reason for this divergence is that people have thought that the distinction between introduction and transmission was relevant only

244 Jeff Speaks

to the sorts of names and kind terms that Kripke considers. But, first, the previous examples indicate that this is a mistake, and that the distinction between introduction and transmission is relevant not just to names and kind terms, but also to standard predicates. And second, names and kind terms are hardly a negligible proportion of natural language expressions; one would think that if theories of meaning were meant to exclude these expressions from their scope, this should be made explicit. But let’s suppose that the right account of the foundations of meaning will consist of a pair of theories: an account of what it is to introduce a term with a given meaning, and an account of how meaning is preserved, or transmitted, between speakers and over time. How does this bear on the question of mentalism? An immediate consequence of the sorts of world-slice supervenience failures mentioned before is that meaning transmission needn’t involve any speaker being in any propositional attitude state whose content includes the content of the expression in question. This rules out the most obvious sorts of mentalist theories of meaning transmission, and makes room for meaning to have a certain priority over mental content in particular cases: as in the example of ‘Pete Rose’, speakers can be put in a position to have thoughts with certain contents in virtue of their coming into contact with a linguistic expression with that content – even if there is no one who exists at that time who has a mental representation of any sort with that content. Mentalist theories of term introduction still remain a live possibility; but the focus on term introduction, rather than on simply using a term with a certain meaning, changes which sorts of mental representations might be relevant to the determination of meaning. In particular, it makes it more plausible that perceptual states could play a role here. It is very implausible that to use a name with a certain meaning, one must have perceptually represented the referent of the name at some point – we often use names for things we’ve never encountered in perception. But it is significantly more plausible – though still, of course, controversial – that one can introduce a name for an object only on the basis of perceptual representation of it. Many predicates might also be introduced via perceptual representations of the properties which are their contents.25 Not all predicates can have their meaning explained in this way, since not all properties which are the contents of predicates are the sorts of properties that can be perceptually represented. Perhaps in these cases, the right account of term introduction can be given by a use theory of the sort which has been developed by Paul Horwich. An example of a predicate whose content plausibly cannot be given by a perceptual experience is ‘is true’. As Horwich suggests, we might think of the rule for introducing a term with the meaning of our truth predicate as the adoption of a disposition to provisionally accept every instance of the corresponding T-schema.26

Foundations of Meaning

245

This sort of mixed account of term introduction is, of course, just a sketch of a theory. But suppose that some such theory is correct. Would we then say that a mentalist account of meaning has been vindicated, or not? In one sense, this sort of view clearly captures some core mentalist commitments: there’s a clear sense in which it locates the foundations of intentionality in a kind of mental representation – namely, perceptual representation. In this sense, the mentalist view that all linguistic representation is ‘derived intentionality’ would turn out to be correct. But, on the other hand, the previous discussion suggests that the relationship between mental and linguistic representation might well prove more complicated than mentalist accounts suggest – and not just because of the need to take account of the distinction between term introduction and term transmission. Perhaps we should understand perceptual representation as prior to and explanatory of certain kinds of linguistic representation; but this needn’t entail a mentalist view of the relationship between linguistic meaning and the contents of judgements and beliefs. A view of term introduction and transmission of the sort sketched before opens the door to a view of these sorts of cognitive mental states as inheriting their contents from perception on the one hand, and the semantic properties of public language expressions, on the other. If this turns out to be right, then even if something of the mentalist’s thesis of the priority of the mental is correct, the mentalist’s view of the relationship between meaning and thought is not.

Acknowledgements Thanks to the participants in my seminar on the metaphysics of meaning in the spring of 2008 at Notre Dame and, for helpful discussion of supervenience, to Fritz Warfield.

Notes 1. See Chisholm and Sellars (1958). 2. Of course, not everyone has agreed with Chisholm; important dissenters include Gauker (1994), McDowell (1994), and Brandom (1994). 3. See Grice (1957 and 1969). For reasons we will not go into, this should be understood as requiring a single, self-referential intention, rather than three separate intentions. See Harman (1974). Obviously, [G] won’t, without modification, handle non-assertoric sentences like imperatives and questions; in what follows I ignore this sort of problem. 4. See among other places Grice (1969). 5. See Grice (1969, p. 111). 6. A possible fix here is to appeal to the notion of a ‘primary intention’, in Schiffer’s sense: ‘To specify one’s primary intention in doing X ... is to give one’s reason for doing X’, as opposed to specifying an intention which one has because one has a given primary intention (Schiffer 1972, p. 62). It might be one’s primary

246

Jeff Speaks

intention, in this sense, that one’s audience entertain a disjunctive proposition and not that they entertain one of the disjuncts; one might intend the latter only because one intends the former. But this modification of the Gricean theory seems too restrictive: Suppose that your car runs out of gas, and I say to you, ‘There’s a gas station around the corner.’ Clearly, though I do intend that you come to believe that there is a gas station around the corner, I only have this intention because I intend you to come to believe that you can fill your car up around the corner; so the former is a secondary intention and not, in the relevant sense, the reason for my utterance. Nevertheless, it does seem as though I meant by my utterance that there is a gas station around the corner. 7. Another class of counterexamples to clause (3) of the account, though a less significant one, comes from a sub-class of the class of cases of reminding, discussed in the chapter. Suppose that you cannot remember your friend’s name, though you feel as though it is on the tip of your tongue. In such a case, I can’t intend that you come to form a belief about the name on the basis of recognition of my intention; I know that hearing the name alone will cause you to form the belief. 8. Indeed, I’ve skipped some of the most notorious problem cases, those to do with audience-less uses of language, and especially uses of language in thought. See for discussion Harman (1974) and Chomsky (1975). 9. Thanks to Jonathan Beere for this example. 10. That is, indexicality aside, a sentence of the form α means that σ and σ′ entails  α means that σ and α means that σ′ . 11. If you don’t think that knowledge places this kind of constraint on intending, the example can be changed to some other epistemic state which might seem more clearly inconsistent with intention. For example, the husband might have easily have said: ‘I love you, but I am completely certain that you will not believe me.’ So it seems that the Gricean can only avoid the argument by claiming that there are no epistemic constraints on intention at all, which seems implausible. 12. I think that this sort of argument can also be used against neo-Gricean analyses, like that defended in Davis (2002), but that is beyond the scope of this chapter. I hope to develop this argument further elsewhere. 13. I think that Ramsey (1927) had a view of this sort in mind. See also Loar (1981). 14. For discussion, see Lewis (1979). In the present case, it is natural to think that, if I had accepted ‘0 = 1’, this could only have been because ‘0 = 1’ expressed some nontrivially-false proposition; and this latter consideration is what makes it false that, in the world under consideration, I believe that 0 = 1. But, by ruling out backtracking evaluations of this sort, perhaps we can put cases like this to the side. A residual worry here is that in making this stipulation, we are illegitimately building facts about what sentences mean into the class of facts held fixed for purposes of evaluation of the counterfactual. I am unsure whether this is a genuine circularity. 15. That is, it is a necessary consequence of the truth of a sentence of the form α believes that σ and σ′ that, in that context, the following sentences are also true:  α believes that σ and α believes that σ’ . 16. Actually, matters are a bit worse than this for [B]. Normal cases of ambiguous sentences are ones in which the sentence is sometimes correlated with one belief, and sometimes with another; ‘I am going to the bank’ is sometimes correlated with beliefs about financial institutions, and sometimes with beliefs about the sides of rivers. But in this case we have a sentence which is always correlated in the relevant way with three beliefs. So the result is not just that conjunctive sentences are ambiguous; it’s that they are ambiguous in a heretofore unimagined way. Thanks to Tom Feeney for pointing this out.

Foundations of Meaning

247

17. Both challenges seem to me to pose serious problems. I am reluctant to think of ‘is scarlet’, e.g., as being syntactically complex; and even if it were, I am unsure how the assignments of meanings to simple sentences could be extended to a complete theory which applies to complex sentences as well. A related move for the mentalist is to try to solve the problem via a distinction between the various propositions which are candidate meanings for a sentence. One proposal which several people have suggested to me is that, if it is a necessary truth that someone who accepts a sentence S believes a set of propositions p1 ... pn, we should assign as the meaning of S the strongest of these propositions. This works nicely for some of the cases discussed in the chapter, such as the conjunctive sentence. Because the conjunctive proposition entails but is not entailed by its conjuncts, the present proposal assigns, correctly, the conjunctive proposition as the meaning of the conjunction. The solution isn’t general though, as is shown by a small variant on the preceding one: if one of the conjuncts is a necessary truth, then neither the conjunction nor the other conjunct will be uniquely strongest; but, for all that, the conjunction is not ambiguous. The example of a proposition and the proposition which ascribes truth to it is similar. And there are some cases in which the weaker of two propositions is the right interpretation. E.g., consider the sentence ‘I am actually a teacher.’ I utter this only when I believe that I am actually a teacher – i.e., a teacher in the actual world – but whenever I believe this, I also believe that I am a teacher. Even so, the sentence ‘I am actually a teacher’ does not mean that I am a teacher; it means that I am a teacher in the actual world. 18. Another possibility is to explain the direct/indirect distinction in terms of sentences in an internal language of thought, or in terms of an internalized semantic theory. While this is a revision of [B] worth pursuing, it’s beyond the scope of this chapter. A short reply to this view is that it seems implausible to say that, if we discovered that we were all possessed of internalized semantic theories which mapped conjunctions in English onto each of their conjuncts as well as a conjunctive sentence of the language of thought, we would then conclude that conjunctions were three-ways ambiguous. I discuss these issues further in Speaks (forthcoming). 19. This sort of worry is also a problem for a familiar sort of variant on [B], which attempt to analyze meaning in terms of correlated beliefs along with mutual beliefs by speakers about those beliefs. Lewis (1975), among others, gives a theory of this sort. But we can imagine a linguistically sophisticated community, in which, e.g., all speakers know – and know that each other know – that belief distributes over conjunction. (Or fill in your favourite of the previous examples.) For such a community, the examples given before will work as well against a convention-theoretic account of Lewis’ sort as they do against the simpler [B]. 20. This is strong global supervenience, in the sense of Bennett and McLaughlin (Fall 2006) and Stalnaker (1996). I’m focusing on global supervenience to make room for the view, shared by many mentalists, that the meaning of expressions in one’s lexicon depends not just on the contents of one’s own mental states, but also those in one’s linguistic community. For present purposes, I’m not sure that much hangs on the distinctions between different sorts of global supervenience claims. 21. For further objections to the idea that metaphysical dependence or priority should be understood in terms of supervenience, see Bennett and McLaughlin (Fall 2006, section 3.5) and Schaffer (2009). The supervenience mentalist might object to the foregoing by claiming that the properties in P belong with the properties in MC. But of course which properties belong with which is part of what is at issue. The supervenience mentalist

248

22. 23. 24. 25.

26.

Jeff Speaks might also object to the inclusion of D in the anti-mentalist’s supervenience base, on the grounds that intentional actions are ultimately to be analyzed partly in terms of beliefs, desires, and intentions. Plausibly, the anti-mentalist who takes facts about beliefs, for example, to be explained in terms of linguistic meaning rather than the other way around is already committed to rejecting this sort of causal-psychological theory of intentional action. But it must be conceded that if some such theory of intentional action turns out to be right, the sort of ‘parity’ argument given before is harder to make. However, perhaps D could be replaced by the class of dispositions to perform certain bodily movements, rather than intentional actions, in certain circumstances. This is the way that distributions of properties over time are handled in the appendix to Paull and Sider (1992). We needn’t assume that this holds for all expressions; it is difficult to construct an analogous case for logical constants. See McLaughlin (1984). This way of putting things assumes a Russellian view of semantic contents; but the Fregean might think of perceptual experiences as providing modes of presentations of objects and properties which could then go on to serve as the contents of names and predicates, respectively. For a plausible argument that in at least some cases it is impossible to introduce a term with a given content without having a perceptual experience involving that content – since otherwise certain sorts of de re claims would, implausibly, be knowable a priori – see Soames (2005). See, among other places, chapter 2 of Horwich (2005). Horwich, of course, intends his use theory to be a theory of meaning simpliciter, rather than a theory of what it is to introduce a term with a certain meaning. But Horwich’s theory seems open to the same sort of refutation by a FIST discussed: the meaning facts at t do not globally supervene on the Horwichian use facts at t, and so can’t be analyzed in terms of them. Each of the examples developed work as well against Horwich’s theory as against the supervenience mentalist.

Bibliography Bennett, K. and McLaughlin, B. (2006) ‘Supervenience’, in E.N. Zalta (ed.) Stanford Encyclopedia of Philosophy (Stanford: Stanford University Press). Brandom, R. (1994) Making It Explicit (Cambridge, Mass.: Harvard University Press). Chisholm, R. and Sellars, W. (1958) ‘Correspondence on Intentionality’, in A. Marres (ed.) Intentionality, Mind, and Language (Urbana: University of Illinois Press). Chomsky, N. (1975) Reflections on Language (London: Temple-Smith). Cranston Paull, R. and Sider, T. (1992) ‘In Defense of Global Supervenience’, Philosophy and Phenomenological Research, 52, 4, 833–54. Davis, W. (2002) Meaning, Expression, and Thought (New York: Cambridge University Press). Gauker, C. (1994) Thinking Out Loud: An Essay on the Relation between Thought and Language (Princeton, N.J.: Princeton University Press). Grice, P. (1957) ‘Meaning’, in Grice (1989). —— (1969) ‘Utterer’s Meaning and Intentions’, in Grice (1989). —— (1989) Studies in the Way of Words (Cambridge, Mass.: Harvard University Press). Harman, G. (1974) ‘Stephen Schiffer: Meaning’, Journal of Philosophy, 71(7), 224–29. Horwich, P. (2005) Reflections on Meaning (Oxford: Clarendon Press).

Foundations of Meaning

249

Lewis, D. (1975) ‘Languages and Language’, 3–35. Reprinted in D. Lewis, Philosophical Papers Volume I (Oxford: Oxford University Press). —— (1979) ‘Counterfactual Dependence and Time’s Arrow’, Nous, 13, 455–76. Loar, B. (1981) Mind and Meaning (Cambridge: Cambridge University Press). McDowell, J. (1994) Mind and World (Cambridge, Mass.: Harvard University Press). McLaughlin, B. (1984) ‘Perception, Causation, and Supervenience’, Midwest Studies in Philosophy, 9, 569–92. Neale, S. (1992) ‘Paul Grice and the Philosophy of Language’, Linguistics & Philosophy, 15, 5, 509–59. Ramsey, F.P. (1927) ‘Facts and Propositions’, in F.P. Ramsey, Philosophical Papers (Oxford: Basil Blackwell). Schaffer, J. (2009) ‘On What Grounds What’, in D. Manley, D. Chalmers, and R. Wasserman (eds) Metametaphysics: New Essays on the Foundations of Ontology (Oxford: Oxford University Press). Schiffer, S. (1972) Meaning (Oxford: Oxford University Press). Soames, S. (2005) Reference and Description: The Case against Two-Dimensionalism (Princeton, N.J.: Princeton University Press). Speaks, J. (forthcoming) ‘Explaining the Disquotational Principle’, Canadian Journal of Philosophy. Stalnaker, R. (1996) ‘Varieties of Supervenience’, Philosophical Perspectives, 11, 221–41.

12 Linguistic Puzzles and Semantic Pretence James A. Woodbridge and Bradley Armour-Garb

A central task of philosophy of language is to explain and elucidate the notions of meaning, reference, and truth – put loosely, the problems of intentionality and extensionality. That said, for the most part philosophers of language have approached these issues indirectly, by attempting to resolve certain linguistic (and, in some cases, logical, metaphysical, or epistemological) puzzles. As Bertrand Russell notes: A logical theory may be tested by its capacity for dealing with puzzles, and it is a wholesome plan, in thinking about logic, to stock the mind with as many puzzles as possible, since these serve much the same purpose as is served by experiments in physical science. (Russell 1905, pp. 484–85) The same is true of philosophical theories. Indeed, philosophical puzzles arising from particular linguistic cases were the launching pad of modern philosophy of language at the end of the 19th and beginning of the 20th centuries, and they remain the life-blood of its aims and methods. We think, for example, of a range of familiar problems, e.g., negative existential claims, ‘empty’ denoting expressions generally, the informativeness of certain identity claims, the non-substitutivity, salva veritate, of expressions with the same semantic-value in certain contexts, the semantic paradoxes, etc. Still further dilemmas emerge in the philosophy of language once theorizing begins. Tensions arise between our conceptual and our ontological commitments, a clash between a kind of theoretical indispensability and a problematic metaphysics. For example, there are fairly good reasons for thinking that our talk about language and thought commits us to propositions, which are supposed to be the contents of the sentences that we assert and the objects of our mental attitudes. But there is a pressing question as to what propositions could be and, given some possible candidates, how they could do the jobs they are supposed to do.1 From Gottlob Frege and Bertrand Russell to the present day, philosophers of language have, for the most part, 250

Linguistic Puzzles and Semantic Pretence

251

attempted to resolve such puzzles by appeal to semantic, logical, pragmatic or re-interpretive innovation. In this chapter, we set out what we see as a novel and very promising approach to resolving a number of the puzzles that provide philosophy of language with much of its subject matter. We will begin by briefly cataloguing a few of the relevant puzzles.

1 The puzzles Consider the puzzle of how a true identity claim, such as: (1) Hesperus is identical to Phosphorus, could be informative, since prima facie it appears to say that some object is identical to itself, which is something we know a priori (since everything is self-identical). Or consider our seeming ability to talk ‘about’ things that do not exist – in particular our ability to state (truly) that they do no exist, as in: (2) Santa Claus does not exist. A naïve referentialist view of language that takes it simply to present and relate objects and properties will yield an epistemological puzzle for identity claims and a metaphysical puzzle for negative existentials. It will also creates a semantic puzzle about how, since semantic values are compositional, substituting expressions that stand for the same object (e.g., ‘George Eliot’ and ‘Mary Anne Evans’) can yield a change in truth-value in certain sorts of sentences, for instance, the shift from true to false in belief attributions such as: (3) Bob believes that George Eliot is a great writer. (4) Bob believe that Mary Anne Evans is a great writer. A naïve take on our discourse about language itself – specifically the putative semantic relations of truth, reference, and predicate-satisfaction, which appear to make out the word-world relation – appears to yield logical and semantic puzzles in certain cases that are properly judged to be pathological, such as: (6) This sentence is false. (Liar Paradox) (7) This sentence is true. (Truth-Teller) (8) The expression ‘the least number not denotable in less than 18 syllables’ denotes the least number not denotable in less than 18 syllables. (Berry’s Paradox)

252

James A. Woodbridge and Bradley Armour-Garb

(9) The predicate ‘is not true of itself’ is true of itself if, and only if, it is not true of itself. (Heterological Paradox) Logical and semantic (‘soritical’) puzzles can also arise from certain ordinary-seeming descriptions, such as: (10) Borderline Bob is bald. when speakers employ a purportedly vague predicate to characterize a supposed borderline case. (This sort of case seems potentially neither true nor false and generates questions about whether our language (or the world) is indeterminate.) Philosophers of language have attempted to deal with linguistic puzzles like those described before through a variety of theoretical innovations that we can broadly classify according to four approaches: semantic, logical, pragmatic and re-interpretive. While we have learned much from these advances, it is fair to say that few of the original puzzles have been resolved in a fully satisfactory way. Be that as it may, rather than dwelling on the negative, we would like to highlight one feature common to all of these approaches, a feature that we think should be retained, even if none of the four approaches should be. The feature: Each of these approaches maintains in some sense that what is being said (e.g., by a given sentence or expression) cannot be read off of the surface. That is, in one way or another, each of the extant approaches denies a face-value reading of some aspects of linguistic discourse. For what follows, having identified these four approaches, we go on to provide a new kind of view, which, while possessing the aforenoted common feature, is, in important ways, different from the traditional approaches. What is central to the new approach is the postulation of semantic pretence at work where these puzzles arise. As we will show, this new approach offers certain advantages to resolving the linguistic puzzles catalogued. 1.1

Linguistic puzzles and theoretical innovations

In this subsection we summarize and review the more orthodox approaches to dealing with the linguistic puzzles that have driven philosophy of language. The aim here is to locate the approach that we want to promote, relative to the better known standard approaches, highlighting the feature common to all such approaches. 1.1.1

Semantic approaches

The first theoretical approach for dealing with various linguistic puzzles involves what we will call semantic innovation. The basic idea behind semantic innovation is to postulate that there is more to what linguistic expressions mean than the dimension of their semantic-values that is already

Linguistic Puzzles and Semantic Pretence

253

recognized from logical inquiry. A prime example of semantic innovation comes from Frege’s views, as presented in the article ‘On Sense and Reference’ (Frege 1892). There Frege attempts to deal with the linguistic puzzles that propelled his inquiry – informative identity claims, sentences employing empty singular terms, non-substitutability salva veritate in certain contexts of expressions that pick out the same objects – via the postulation of another dimension of meaning, that of sense, beyond the dimension of reference (or more generally, semantic-value). The dimension of sense is supposed to provide an aspect of meaning that can differ between co-referring expressions found in an informative identity claim. Sense could also serve as a component of meaning that a singular term still contributed to the meaning of a sentence, even when it lacked any semantic-value to contribute. Frege further explains that a difference in the sense of expressions can account for their non-substitutability salva veritate in certain contexts, even when those expressions agree in semantic-value in contexts of primary occurrence. The curious embedded contexts, he maintains, produce a shift in the expressions’ semantic-values – from what they are in contexts of primary occurrence to their customary senses. Substituting expressions in the embedded context thus yields a change in the semanticvalues of a sentence’s component expressions, and thus explains any change in the truth-value of the whole. 1.1.2

Logical approaches

The second theoretical approach involves what we will call logical innovation. On this sort of approach, theorists attempt to deal with various linguistic puzzles by postulating a kind of hidden complexity in the form or underlying structure of the expressions in questions. The paradigm example here is Russell’s work involving his theory of descriptions (Russell 1905). Russell rejects Frege’s semantic innovation involving the postulation of sense and instead postulates a difference in most sentences between their surface grammatical form and their underlying logical form. On his analysis, most singular terms are actually either indefinite or definite descriptions (perhaps ‘abbreviated’ as common names). Supposedly complete nominal expressions get cashed in for logically complex but incomplete expressions involving quantifiers and predicates. These quantificational expressions essentially have a gap that must be filled in with a predicate to form a complete expression. Russell holds that once we recognized the complex underlying logical forms of most sentences, we will see that there was no need to postulate a dimension of meaning beyond that of semantic-value. A different form of logical innovation may be found in the work of Saul Kripke (1975), who, in effect, aims to provide a definition of truth by equating the property of being true with possessing the semantic value one.2 The theory, if successful, would enable us to define truth by appeal to a

254 James A. Woodbridge and Bradley Armour-Garb

non-classical model theory, the central features of which includes a particular non-classical logic (viz., K3) and certain inferential rules (Kripke 1975, p. 701), while, at the same time, offering a solution to the liar paradox (and the truth-teller). 1.1.3

Pragmatic approaches

The third theoretical approach involves what we call pragmatic innovation. There are a number of familiar pragmatic approaches, from P. F. Strawson’s (1950) reliance on use and contexts of use in accounting for linguistic meaning, understood as a rule for the proper use of a sentence or expression, to Keith Donnellan’s (1966) distinction between the referential and attributive uses of definite descriptions, through H. P. Grice’s (1989) important distinction between the truth of a statement and its conversational acceptability. What is common to these approaches (we think, more specifically, of the latter two cases) is the difference between what is meant, in an utterance, on a particular occasion of use (or what is conveyed or communicated, by an utterance, on a particular occasion of use) and what is, strictly speaking, said by the sentence uttered. Once the distinction is made, certain philosophical puzzles are said to be resolved.3 1.1.4

Re-interpretive approaches

The fourth approach to resolving linguistic puzzles involves what we call re-interpretive innovation, it is in some ways an expansion or extension of the logical approach as employed by Russell. This is true at least for the application of the re-interpretive approach by its most thorough going proponents in the heyday of philosophy of language, the Logical Empiricists. In fact, A.J. Ayer, one of this movement’s most celebrated champions, in philosophy of language in particular, is happy to accept Russell’s logical innovations as the solution to the puzzles Russell emphasizes. Inspired by what they (or, more specifically, Ayer 1952) saw as Russell’s success on those fronts, the Logical Empiricists extended the sort of moves that he made to resolve a number of further puzzles about language (or puzzles that became about language), when various traditional philosophical issues get re-cast in linguistic form. This approach leads to a variety of ‘re-interpretations’ of various fragments of language, designed to dissolve what seem to be philosophical issues about the subject matter of these ways of talking.4 While the problems that confront the Logical Empiricists seem to be insurmountable, we maintain (and will make evident in ensuing sections) that there are certain merits to a re-interpretive approach, so long as one is careful not to (i) wed it to other problematic positions or (ii) apply it in a blunt fashion, which takes it to be a brute fact that the sentences from some fragment of discourse express something other than what they appear to express on the surface, with no explanation as to how this might work or why it is so.

Linguistic Puzzles and Semantic Pretence

255

In more current philosophy, we find the re-interpretive approach mainly in metaethics, under the guise of various versions of expressivism, put forward (e.g.) by Allan Gibbard (1990) and Simon Blackburn (1993). On Gibbard’s account of so-called normative judgments, despite its surface appearances, a declarative utterance of this sort does not make an assertion; instead it expresses acceptance of a system of norms that either requires, forbids, or permits the action picked out in the utterance. Blackburn offers a projectivist, ‘quasi-realist’ account of moral discourse, according to which the seemingly assertoric utterances from the discourse do not actually state putative moral facts (there are none) but instead express approval or disapproval for an action.5 Expressivist views face a number of familiar problems raised for noncognitivist accounts, including the Frege-Geach (or embedding) problem of how such non-factual claims function logically in seemingly valid arguments when they are embedded in conditionals. (For a useful discussion of the problem, see Kalderon 2005, chapter 2). Another version of applying the re-interpretive approach in metaethics is exemplified by the error-theoretic account John Mackie offers. On Mackie’s (1977) view, as for any error theory of any discourse, the re-interpretive move involves interpretation in the propositional logic sense of an assignment of truth-values, rather than the postulation of a different content being expressed or a different speech act being performed. An error-theoretic view of some discourse takes the informational content of the relevant sentences to be just what it seems, while also assigning all the affirmative atomic sentences from the fragment in question the truth-value ‘false’. Error theories raise the following puzzle, which we call The Problem of Error (henceforth, PE): If some fragment of discourse is actually, and systematically, false, why do competent language users employ that way of talking, uttering sentences that do not express what the speakers aim to express, in asserting what they do? What explains the widespread, systematic error being attributed to such (presumably rational) speakers? We should note that these are puzzles that error theories raise; they are not, at least not on their own, objections to error-theoretic accounts. That said, since, in general, one should not take a widespread attribution of error lightly, we maintain that if some non-error-theoretic account could offer all of the practical and commitment-avoiding advantages of an error-theoretic account, without generating new problems of its own, we should prefer the non-error-theoretic account. As we will show, there is a way of retaining a re-interpretive approach without taking on the commitments of the Logical Empiricists, non-cognitivism, or embracing a Mackie-style error theory. Before moving on, we return, briefly, to the Logical Empiricists. Despite the problems with Ayer and other Logical Empiricists’ brute and blunt application of the re-interpretive idea, we think that this can be an important insight about a way of talking, provided it is separated from a positivistic,

256 James A. Woodbridge and Bradley Armour-Garb

reductionist, verificationist, or otherwise problematic framework. And, while the Logical Empiricists applied the re-interpretive approach mainly to resolve what they took to be further linguistic puzzles (resulting from their linguistic re-casting of what had standardly been taken to be philosophical issues concerning the subject matter of certain ways of talking), we think that the approach’s basic idea will actually turn out to be useful for dealing with the linguistic puzzles of interest here. In a pretence approach it is (unsurprisingly) the operation of an element of pretence – a kind of fiction – in an utterance so understood that is held to generate the distinction between what is really being said and what appears to be said. The approach is thus a variety of fictionalism (the relationship between a pretence account and fictionalism is one of species to genus) about some way of talking, but one that is importantly different from what is probably still the more common understanding of fictionalism about some discourse. In order to see this, we turn to fictionalism proper.

2

Fictionalism

Fictionalism about some discourse is primarily a linguistic thesis. It is a response to worries about a supposedly problematic discourse (e.g., one that seems to traffic in metaphysically problematic objects or properties) that aims to resolve the apparent problems there by reanalysis of at least some aspect of the discourse itself. In so doing, a fictionalist account of some discourse does not directly address any metaphysical or epistemological problems to which the content of the discourse gives rise, but it is intended to strip the discourse of various problematic commitments, both ontological and epistemic, while retaining certain expressive advantages the discourse might provide in its current form. Even so, fictionalism is thus compatible with both semantic realism and semantic anti-realism about a given discourse on a given subject matter. 2.1 Kinds of fictionalism The particular approach involving semantic pretence that we aim to explain and motivate here is a form of fictionalism, one that shares some central aspects with the re-interpretive approach to resolving linguistic puzzles. As in the re-interpretive approach, the postulation of pretence at work in a fragment of discourse involves maintaining that the content of an utterance so understood (specifically what we call an utterance’s serious content – what, if anything, the utterance says about the real world) cannot be read off of the surface, i.e., on a face-value reading of the utterance. A face-value reading will yield only the pretend content of a pretence-involving utterance – what the utterance says when it is understood from inside the pretence it involves. We will have more to say about this distinction presently. For now, we note that this is similar to the aspect of a re-interpretive approach,

Linguistic Puzzles and Semantic Pretence

257

according to which it turns out that what is really being said (or, sometimes, what is really being done; cf. expressivist accounts of ethical discourse) is different from what appears is being said (done) on the surface. To locate the kind of fictionalism we favour within the general approach, it will help to lay out some distinctions between different kinds of fictionalism. A major distinction frequently drawn within fictionalism at present is that between revolutionary fictionalism and hermeneutic fictionalism (Stanley 2001). The former sort of view claims that people using some apparently problematic discourse intend to make certain representational claims with the talk, but the discourse (or, at least, the way speakers employ it) is genuinely problematic. However, the revolutionary fictionalist claims, the discourse can be rendered unproblematic and can even be vindicated, by coming to understand it as involving a fiction. (Cf. Field 1980 and 1989 on mathematical discourse.6) Revolutionary fictionalism thus offers a reform proposal about how we should come to understand a way of talking. By contrast, hermeneutic fictionalism claims that some apparently problematic discourse is just that – merely apparently problematic. Theories of this type maintain that a proper understanding of how the discourse functions will show that it already involves fiction in a way that renders it unproblematic (maintaining focus on mathematical discourse, Yablo 2005 is a paradigm example here7). Hermeneutic fictionalism is thus a descriptive account of how a discourse works, rather than a normative account of how we ought to come to understand it. The revolutionary/hermeneutic distinction is not particularly important for our purposes here, save for the fact that they enable us to highlight two variants of fictionalism, which are sometimes tied to revolutionary fictionalism and hermeneutic fictionalism, respectively. We will call these variants Error-Theoretic Fictionalism (henceforth, ETF) and Pretence-Involving Fictionalism (henceforth, PIF). The variant that interests us, for resolving the linguistic puzzles, is PIF, but to be clear about what a view of this sort does and does not claim, and in order to perceive its merits properly, it is crucial to distinguish it from the other variant, ETF. 2.1.1 Error-theoretic fictionalism ETF is probably the most common form of fictionalism, as the best-known fictionalist accounts of various ways of talking amount to error theories of the discourse. A prime example of this sort of view is, again, Hartry Field’s account of mathematical discourse (Field 1980 and 1989). On Field’s view, claims involving ordinary number-terms, such as: (11) Two plus two equals four. are uniformly false, since there are no abstract objects of the sort that these terms (e.g., ‘two’ and ‘four’) purport to denote.8 Field holds that

258

James A. Woodbridge and Bradley Armour-Garb

mathematicians and other competent language users go about asserting things that, in light of his nominalism, are actually false. (Of course, they believe that what they are asserting is straightforwardly true, but Field claims that they are in error about this.) He does allow that we can take claims like (11) to be true in a sense similar to how a claim like: (12) There is a brilliant detective at 221b Baker Street. can be true, namely in the sense of being true ‘according to a certain wellknown story’. A claim from mathematical discourse like (11) can be considered true ‘only in that it is true according to standard mathematics’ (Field 1989, p. 3). Strictly speaking, however, any such claim is false, hence Field’s fictionalism about mathematics is an example of ETF. Gideon Rosen’s (1990) account of possible worlds discourse can also be taken as a case of ETF. According to the view, any claim apparently about possible worlds, such as: (13) There is a (non-actual) possible world at which there are blue swans. is false if taken as a straightforward existential claim about reality, just as a claim like (12) is false if taken as a straightforward existential claim. (Rosen 1990, pp. 331–32) However, as for Field, Rosen allows that if the sentence is understood as involving a (possibly elided) ‘story prefix’ of the form ‘In the fiction, F, ...’ or ‘According to such and such a story ...’, then it could turn out to be true. Following David Lewis (1978), then, both Rosen and Field allow that, while the existence-implications sentences from a fiction-involving way of talking are all, uniformly false, strictly speaking, some sentences belonging to the discourse might turn out to be ‘true in the fiction’. This can happen if (and only if) they are offered and understood as elliptical presentations of claims prefixed with some variation on the (sentential) operator ‘In the fiction ...’ On this explanation, we can understand the sentences only as being about the content of the fiction they belong to, and not about the real world outside of the fiction. If they get the details of the story right, they can turn out to be true (as is the case with (12), understood as a claim about the Sherlock Holmes stories). However, if we take the sentences to be about the real world outside the fiction, rather than the fiction itself, then they all end up being false. By yielding error theories in this way, ETF accounts face the following modified version of PE. While speakers can use sentences from the discourse to make true assertions, these claims can only be about the content of some fiction. This amounts to attributing massive error to speakers regarding the status of their talk, since typically speakers take themselves to

Linguistic Puzzles and Semantic Pretence

259

be making true assertions about the world when they employ the discourse that they do. This is in addition to taking the ‘problematic’ sentences of the discourse itself to all be in error, since they are uniformly false, if taken to make claims about the world. Since such attributions of massive and widespread error are not to be taken lightly – in particular when attributed to presumably rational language users – we take these consequences to be serious enough to motivate the search for a variety of fictionalism that avoids these worries. That is what PIF purports to offer. 2.1.2 Pretence-involving fictionalism Until fairly recently, ETF was the only form of fictionalism on the theoretical table. Theorists typically assumed that anytime one gave a fictionalist account of some discourse, he or she was giving an error-theoretic account of the talk.9 But, again, postulating an error theory of any sort raises the issues belonging to PE, so if a fictionalist account could be put forward that was not error-theoretic, though with all of the expressive and commitmentavoiding advantages of an ETF account, it is to be preferred over its errortheoretic sibling. Exactly this possibility is what an appeal to pretence forms of fiction provides. In supposedly allowing for the possibility that certain utterances can still be used to make genuinely true assertions even though they involve a kind of fiction, PIF is the usual strategy for a theorist developing a hermeneutic fictionalist account. The notion of pretence is the key factor in such a view because an appeal to pretence – specifically the sort related to make-believe – can block the error-theoretic conclusions that ETF draws. This is because make-believe involves systematic dependencies between the pretence and the real world. PIF, as an approach in the philosophy of language, has received substantially increased attention over the past 15 years. The central ideas stem largely from Kendall Walton’s pioneering work in aesthetics, specifically the role of make-believe in the representational arts (Walton 1990). The sort of PIF account that comes most directly from Walton’s work (including Walton’s own account of our talk putatively about what does and does not exist, henceforth, existence-talk) is specifically semantic pretence-involving fictionalism (henceforth, SPIF). On this sort of view, pretence is a factor in the semantic functioning of various locutions, at least in certain contexts (e.g., embedded contexts such as belief attributions). A SPIF view is to be distinguished from an account that postulates pretence as a factor in certain aspects of the pragmatics of utterances, rather than in the semantic functioning of certain utterances. We discuss this other variety of PIF, and the problems that it faces, in the sections that follow. For now, we focus on SPIF accounts, with the aim of explaining the role of make-believe in them and the advantages this gives them over ETF accounts.

260 James A. Woodbridge and Bradley Armour-Garb

2.2 The workings of SPIF accounts The notion of make-believe relevant to SPIF accounts of ways of talking takes off from the sort that is familiar from children’s games of makebelieve. On Walton’s analysis, this kind of pretence typically involves several elements: i) props of some sort, ii) stipulated pretences, and iii) principles of generation. These work together to generate further pretences of a different sort – pretences that depend in part on how thing are in the real world. The principles of generation are rules about how to take real world facts about the things serving as props in the game, along with stipulated background pretences of the game that are expressly made-believe, to determine what else is to be pretended (or as we will say here, what further pretences are prescribed). These further pretences are thus generated from reality (Crimmins 1998, p. 5). The systematic dependency this produces, between whether some pretence is prescribed and whether certain real-world conditions obtain outside of the game, allows for the possibility of exploiting the pretence for serious linguistic purposes. As an example of how this might work, consider the following case, in which some of the details of the particular make-believe involved are explicit. Say that Dex and Zev are playing a Star Trek game of make-believe, where the game involves certain stipulated pretences about various props, including the pretences that Dex is Captain Kirk, Zev is Mr. Spock, cell phones are communicators, flashlights are phaser pistols, the kitchen is the bridge of the Starship Enterprise, the bathroom is the transporter station, etc. Given these stipulated pretences (along with other background pretences belonging to the Star Trek story or stipulated by the pretence theorists), principles of generation establish further pretences as to be pretended, that is, they are prescribed by the game based on what real-world conditions hold outside the game. For example, if Dex drops his only flashlight, it is to be pretended that Captain Kirk is unarmed, and if Zev pinches someone on the shoulder it is to be pretended that he has rendered that person unconscious. We can take advantage of the dependencies the game established in order to make a serious claim indirectly, by making an utterance that belongs to the make-believe. This amounts to making a ‘partially pretend’ claim.10 For example, someone might say: (14) Captain Kirk stole Mr. Spock’s phaser pistol and hid it in the transporter station. This utterance involves pretences from the make-believe. In using it a speaker is not seriously talking about a kind of laser gun and a station that ‘beams’ people (and things) to remote locations. Using (14) to make an assertion puts forward the pretences displayed in the utterance as being prescribed. Given

Linguistic Puzzles and Semantic Pretence

261

the principles of generation and the stipulated pretences of the game, the pretences displayed in (14) are prescribed if and only if certain real-world conditions obtain. By asserting (14), one expresses the obtaining of those conditions – in effect, one thereby asserts what she would assert by directly stating those conditions, as in an utterance of: (15) Dex took a flashlight that belongs to Zev away from him and hid it in the bathroom. By presenting certain pretences as prescribed – by putting forward what we call the pretend content that an utterance presents on the surface – an assertoric use of (14) says indirectly what an asseretoric utterance of (15) says directly. (15) directly presents the serious content put forward indirectly through an utterance of (14). Thus, someone who utters (14) makes a genuinely true claim exactly when (15) is true, which is when the serious content of (14) is true, i.e., when the pretences displayed in (14) are prescribed. In general, then, speakers can use sentences that belong to a game of make-believe to make genuinely true, serious assertions indirectly, exactly when the pretences they display are prescribed.11 Pretence-involving utterances can thus provide a way of engaging in ‘indirectly serious discourse’ (p. 32). This distinguishes PIF from ETF: While an utterance understood in terms of PIF might not be true, when taken literally (if it is possible to do this, and we were to take it seriously at face-value), nevertheless, in virtue of the systematic dependencies make-believe has on the real-world, such an utterance can still be genuinely true, viz., by indirectly expressing a genuinely true serious content. As a result, a game of make-believe can provide a mechanism through which a speaker can, by making as if to say one thing, succeed in making quite a different, albeit still serious assertion about the world. The forgoing explains how pretence-involving claims can be used for serious purposes and can avoid any attribution of error to the discourse or the speakers using it. Returning to our toy example, one might offer (14) as an explanation for why Zev cannot find his flashlight. An appeal to makebelieve thus allows for, rather than undermines, the serious purposes served by a way of talking. And if some talk is problematic when taken at face value, an appeal to pretence might explain how it serves any serious purposes at all. Certain linguistic puzzles might thus be solved by recognizing makebelieve at work in ways of talking where it has not been noticed before. 2.3 Distinctions among types of pretence So far, we have considered an act of overt, or explicit, make-believe. To clarify the specific way that SPIF can provide solutions to our linguistic puzzles, it will help to make some distinctions between different ways that pretence can operate in utterances. The first distinction is between what Walton calls ‘prop-oriented make-believe’ and what he calls ‘content-oriented

262

James A. Woodbridge and Bradley Armour-Garb

make-believe’ (Walton 1993). Prop-oriented make believe is exemplified by (14); it is the sort of pretence that is most relevant for SPIF accounts of ways of talking, as it serves mainly to talk about the real features of the props employed in the game of make-believe. It relies on the game’s principles of generation, which yield some of what is to be pretended in that game as a function of what occurs in the real world outside of the game. Speakers can then exploit the systematic dependencies, to talk about the real world indirectly (e.g., by making claims about ‘phaser pistols’ to talk about the properties of certain flashlights). In a SPIF account of some fragment of discourse (e.g., existence-talk or identity-talk) the props are locutions or pieces of language employed in the relevant utterances. The serious content of these utterances thus often ends up in a certain sense being about locutions used in the utterance from the discourse. Content-oriented make-believe is less relevant for our purposes, as it serves more to talk about what the ‘world of the game’ includes, i.e., what pretences are part of the overall story of the make-believe, and, thus, what is more a matter of stipulation than dependent on real-world conditions. This is the sort of pretence involved in our interactions with novels, plays, films, and other representational works of art. (Walton 1990) We might also use this label to categorize the fictionalism involved in ETF accounts, since that approach restricts fiction-involving claims to being about the content of the relevant fiction, at least in so far as an ETF account recognizes any utterances it covers as usable for making a true assertion. Recall that ETF acknowledges fiction-involving utterances as true only when they are understood to be about the details of the fiction in which the talk is embedded (e.g., as involving a ‘story-prefix’), rather than about the world outside of the fiction. That said, some claims from a fragment of discourse for which we provide a SPIF account will involve content-oriented makebelieve. What we might call pretence-framework claims will express some of the details of the stipulated background story for the make-believe behind the discourse, and thus involve content-oriented make-believe. Even so, the more interesting cases, and the real gain of PIF over ETF, are those where we use utterances from that fragment of discourse to talk about the world, and this is made possible via PIF’s introduction of the invocation of proporiented make-believe. The second distinction we need to make here is that between cases that involve what we can call extrinsic pretence and cases that involve what we can call intrinsic pretence (Woodbridge 2005). The central difference here has to do with whether pretence attaches to the utterance ‘from the outside’, or whether pretence is integral to the operation of utterance in its saying anything at all. In the basic cases of extrinsic pretence (first-order extrinsic pretence12), we could take the utterance that was made literally. What we mean by this is that a face-value reading of the utterance gives us something that we could also, in some circumstances, take seriously – in

Linguistic Puzzles and Semantic Pretence

263

the case of an assertoric utterance, as a genuine, direct statement about the real world. Most metaphors of the form ‘A is B’ involve extrinsic pretence in this way.13 Intrinsic pretence is really what is important for our purposes here because that is what we take most of the cases presenting our linguistic puzzles to involve. In cases of intrinsic pretence, the pretend statement an utterance makes is not something someone could offer as a serious statement in any actual circumstances. A face-value reading of the utterance provides something that could only be a pretend statement. We pretend that the utterance is meaningful, when we take it at face value (i.e., without the operation of some pretence), but the only serious content there is to associate with it is the content the utterance puts forward indirectly, in virtue of its role in the pretence. So, unlike sentences that can have both a literal meaning and a figurative meaning (e.g., many metaphors), utterances that involve intrinsic pretence cannot be taken literally. So here it makes no sense to talk about whether the utterance is literally true or false. However, it still might be genuinely true, provided the serious content it expresses indirectly says something true, but an intrinsic pretence-involving utterance has no status on the axis of literal interpretation. Typically, the reason an utterance invokes pretence intrinsically is because there is no way to take some part of it seriously, at face value. In other words, an utterance’s lack of literal content as a whole usually results from the failure of at least one of its components to have any literal content. This, we maintain, is what is going on in several of the linguistic puzzles under consideration here. To explain what we mean, we now turn to explaining how a SPIF account of the relevant way of talking can resolve the linguistic puzzles it generates. 2.4 SPIF accounts for some linguistic puzzles For space considerations we confine ourselves here to discussing a particular sort of linguistic puzzle and explaining how a SPIF account of the talk that generates it will resolve the problems that are thought to arise. The puzzle on which we will focus regards our apparent ability to talk ‘about’ things that do not exist – most interestingly, our ability to state truly that they do not exist (i.e., to make true negative existential assertions). The account of existence-talk we consider here is modelled on the one developed by Walton (1990, chapter 11) and augmented by Mark Crimmins (1998).14 Certain aspects of this account have been further applied in the development of PIF resolutions of other linguistic puzzles, including informative identity claims and non-substitutivity in attitude ascription (Crimmins 1998), semantic paradox (Woodbridge 2005), worries about possible-worldtalk and number-talk (Yablo 1996 and 2005), and metaphysical and epistemological worries about propositions (Woodbridge 2006; Armour-Garb and Woodbridge 2009b).

264 James A. Woodbridge and Bradley Armour-Garb

2.4.1 Empty names in general Consider first, the use of empty names outside of existence claims. The puzzle that can arise here is how to understand claims that employ empty names, or names that have no bearers, since, ex hypothesis, there is nothing that such claims are about. Speakers typically proceed as if names have bearers. If they do not, there are two further sets of circumstances that are relevant. Do the speakers believe the name refers, or do they believe it is empty? In the former case we have what we will call the knowing use of empty names. Probably the most common context of the knowing use of empty names involves the special case of fictional names (understood as such). These are names that come from, and are linked to, some work of fiction, such as the name ‘Sherlock Holmes’. There are a number of different sorts of claims this name can appear in, even with the assumption that it is known to be empty in place. The first sort of claim we might make with a fictional name is what Walton calls an ‘ordinary statement’ concerning a work of fiction. In such cases, a claim like: (16) Sherlock Holmes lives at 221b Baker Street. makes a serious assertion about the story that supplies the name. (Walton 1990, p. 403) The games of make-believe that ground our interactions with written works of fiction involve treating the story provided as a prop. In this particular case, call the relevant story that provides the name the ‘Holmes’stories. The make-believe involved in interacting with the relevant work of fiction also include a principle of generation that amounts to treating the story presented as a veridical record of real events and states of affairs. As a result, the serious content expressed indirectly through an assertoric utterance of (16) is something like what would be expressed directly by an assertoric utterance of: (16*) The ‘Holmes’-stories are such that they portray someone named ‘Sherlock Holmes’ who lives at 221b Baker Street. (Cf. Crimmins 1998) Another sort of claim that employs a fictional name involves extending its use beyond relaying the content of the story that serves as a prop for the make-believe the story generates. For example, one might want to make a serious point by uttering: (17) Sherlock Holmes is smarter than Sam Spade. This claim counts as a move in a combined make-believe, involving both the ‘Holmes’-stories and the ‘Sam Spade’-stories as props. The rules of the expanded game prescribe pretending that these two collections of stories are

Linguistic Puzzles and Semantic Pretence

265

both accurate records of real event, involving real people, in a single world. This allows us to assess what the collections of stories say and make comparative assessments of various things by relating them to the relevant stories. Following Crimmins, we might analyze the serious content expressed indirectly by an utterance of (17) as what would be expressed directly by uttering: (17*) The ‘Holmes’-stories are such that they portray a level of intelligence as possessed by someone named ‘Sherlock Holmes’ and the ‘Sam Spade’-stories are such that they portray a level of intelligence as possessed by someone named ‘Sam Spade’, and the first level of intelligence is higher than the second. (Cf. Crimmins 1998, p. 3) The last context of use we will consider for fictional names involves what is sometimes called a meta-fictional claim. This sort of claim invokes pretence to assert something about the real world that is not drawn out of the content of a story. In fact, such a claim typically would be incorrect if understood as expressing part of that content. The paradigm example is an utterance of a sentence like: (18) Sherlock Holmes was created by Arthur Conan Doyle. In this sort of case, the pretence surrounding the role of certain stories as a prop gets extended to include a principle of generation that makes it to be pretended that someone has created a person the story portrays as part of the world if, and only if, the someone is the original author of the stories that serve as a prop in the make-believe. This yields the intuitive result that the serious content of an utterance of (18) is that Conan Doyle is the original author of the ‘Holmes’-stories and its use of the name ‘Sherlock Holmes’. Moving from fictional names to non-fictional, empty names, we can model an account of the serious content that attaches to utterances employing the latter on the account for the former. Still assuming that the name employed is known to be empty, we can say that just as the serious content that attaches to an utterance of (16) is about the ‘Holmes’-stories, the serious content that attaches to an utterance of a sentence like: (19) Vulcan is a planet between Mercury and the Sun. employing an empty name (here, ‘Vulcan’) is about the (mini-)theory or ‘lore’ surrounding the standard use of that name (here, the ‘Vulcan’-lore) (Woodbridge 2005, p. 175, n. 82). Thus the serious content that attaches to an utterance of (19) is what would be expressed directly by an utterance of: (19*) The ‘Vulcan’-lore is such that it describes a planet called ‘Vulcan’ as located between Mercury and the Sun.

266

James A. Woodbridge and Bradley Armour-Garb

All of the cases considered thus far have assumed contexts involving the knowing use of empty names. However, it is clear that not all uses of empty names will occur in this context. Still, we think that the account offered so far can be extended to unknowing uses of empty names, or at least to the most common ones. These we take to be situations in which a speaker intends to use the relevant name in the same way that her linguistic community does. This deference to her linguistic community, in particular to experts about the use of the name, will make it the case that the name functions in her utterances the same way it does in those of the experts. It functions as a known-to-be-empty name in her speech, although she does not know this, so, via deference of use, the utterances that she makes employing this name inherits the serious content that attaches to the utterances made by the experts who use the name. In the bizarre and presumably extremely rare (because seemingly pointless) possible circumstances in which a speaker employs a name that is in fact empty, as used by her linguistic community, but where she intentionally rejects deference to expert use and instead stipulates that the name has a bearer, but without any knowledge of an intended referent, then the name functions in her utterances as an arbitrary name. This deviant use leaves it open as to what, if any, serious content attaches to her utterances that employ the name. However, this situation is not peculiar to the SPIF account of empty names we offer here; the same would hold on any account of how names (empty or referring) function in utterances, whether the account involved pretence or not. 2.4.2 Existence-talk Return to one of the explicit puzzle cases canvassed before, that presented by negative existential claims, such as: (2) Santa Claus does not exist. Postulating make-believe at work in claims like (2) explains how uses of them can produce true assertions. The idea is to recognize that existence-talk is based on a game of make-believe governed by rules like the following: 1. It is to be pretended that every name or singular term has a bearer. 2. It is to be pretended that ‘exists’ is used to attribute a property (‘existence’) that some things have and some things lack (a discriminating property). 3. The pretence that something has the property of existence is prescribed if and only if attempts to refer with the name ‘providing’ the ‘something’ to the game (by 1.) are successful. Otherwise the pretence that the denoted entity lacks the property of existence is prescribed. 4. The pretences displayed in an utterance of N exists are prescribed iff N as used in that utterance refers to an object.

Linguistic Puzzles and Semantic Pretence

267

So the serious content expressed indirectly via uttering a sentence like (2) assertorically is that the name ‘Santa Claus’ as used in that utterance of (2) fails to refer to any object.15 This Walton-inspired account of existence-talk (along with the preceding account of talk employing empty names) amounts specifically to a SPIF account involving semantic pretence because the pretence is about the meanings of the expressions involved in the talk (names, singular terms, and the predicate ‘exists’). The expressions used in the utterances are themselves props in the relevant game of make-believe, so the prop-oriented make-believe that becomes available for exploitation makes it possible for claims that use the expressions in specific ways to express indirectly serious content about those very uses of the expressions. The pretences that back the discourse thus effect a collapse between use and mention, providing a means of performing a kind of deferred ostension that lets us pick out and describe particular kinds of uses of expressions by making utterances that employ (and thus display) the kinds of use in question. SPIF accounts like these thus put familiar kinds of linguistic resources to useful, new purposes, extending the expressive capacity of the language in a logico-syntactically conservative way. At least in the case of existence-talk, some aspects of the prop provided by the locution ‘exists’ result in the utterances that employ this locution involving pretence intrinsically. The main reason for this is that surface appearances have this locution operating like a genuine descriptive predicate that serves to characterize objects. But this role must be one that this expression has only within some pretence, because being a genuine predicate (as opposed to being a grammatical predicate) requires that the expression comes with criteria for something to be in its extension and criteria for something to be in its anti-extension. But while there is a pretence to the effect that ‘exists’ comes with these criteria and that negative existential claims place particular objects in its anti-extension, all of this must involve pretence. The reason is that ‘exists’ cannot have an anti-extension. By this we do not mean there are criteria for something being in its anti-extension, but as a matter of fact, nothing satisfies those criteria. The expression does not have a contingently empty anti-extension; it does not have an antiextension at all. Nothing could possibly be in its anti-extension, since it would first have to be, i.e., to exist, and would thus be in the expression’s extension. As a result, there are no pretence-free uses of ‘exists’ – they all involve a pretence that the expression functions as a predicate. This is particularly apparent in negative existential claims, but also holds for positive existentials as well. The intrinsic pretence involved in existence-talk is what prevents any version of PE from applying to this account. We have seen that instances of existence-talk express indirectly serious content about kinds of uses of what Russell calls denoting expressions. There is no bar to the instances of existencetalk being genuinely true in virtue of this serious content they express being

268 James A. Woodbridge and Bradley Armour-Garb

true. But the intrinsic nature of the pretence the talk involves makes it the case that there is no sense in asking for its status on a literal interpretation. Instances of existence talk cannot be taken seriously at face-value because there are no pretence-free uses of the expression ‘exists’; so these utterances cannot be taken literally, and it makes no sense to say that they are literally false. As a result, PE gets no grip on this SPIF account of existence-talk. 2.5

Objections to PIF generally

We hope we have laid out the difference between PIF accounts and ETF accounts in enough detail to show how the newer variety of fictionalism avoids the modified PE puzzles generated by the error-theoretic aspect of the older variety of fictionalism. But the move from ETF to PIF generates a new worry: what we will call The Engagement Problem (henceforth, EP). EP is usually presented as the claim that a given pretence-involving solution to a philosophical puzzle is implausible, since, in general, ordinary speakers, who assert the likes of (1) and (2), do not seem to be aware of, much less actively participating in, any pretence or game of makebelieve. (Crimmins 1998, pp. 14–15; Richard 2000, pp. 211–12; Stanley 2001, pp. 46–7). While we agree that the objection would be serious if it applied, we deny that it creates a problem for SPIF accounts because this approach does not provide an account of speakers’ attitudes or activities. A speaker who has uttered (14) (or even (12) or (16)) would most likely think of herself as somewhat engaged in (or at least intentionally alluding to) the make-believe in which the utterance counts as a move, and the same might be true of speakers using metaphors, hyperbole or other figurative modes of speech. But it is not (and certainly need not be) true that people making existence claims think of themselves as pretending anything. The sorts of SPIF analyses we consider helpful in dealing with the linguistic puzzles of interest here do not assume that speakers or hearers are engaged in, or even aware of, such a pretence. On our understanding of this approach, pretence comes in as part of the account of how the talk functions semantically; it does not enter as part of an account of what speakers intend to do or what hearers take them to be doing. While we can describe a speaker’s use of a pretence-involving way of talking as like the use of a figure of speech that is best understood in terms of a possible game of make-believe, that does not mean that the speakers are using language figuratively in the usual intentional sense. A speaker need not engage in the game behind the talk in order to use that talk. Moreover, she does not have to be aware of how or whether the talk’s functioning involves pretence. Speakers typically do not (although they might) take any attitude towards their talk; they simply use it as a tool to make claims. They proceed on generic assumptions that names and singular terms pick out objects and that predicates serve to characterize objects. They do not endorse universal principles to these effects; they do not feel forced to stop

Linguistic Puzzles and Semantic Pretence

269

talking certain ways when confronted with counter-examples (such as negative existential claims like (2)). Although speakers employing a pretence-involving way of talking need not engage in the pretence, SPIF theorists offering an account of that fragment of discourse, will mention pretence, in order to explain what serious claims about the world its instances make (and how they do this). But no one using that discourse needs to engage in, or even be aware of, the games of make-believe that figure in the explanation of how pretence-involving ways of talking function. SPIF accounts thus avoid the EP by keeping all reference to pretence within the theorist’s explanation of the claim’s semantics – of how it ends up with the serious content it has. What a speaker seriously asserts via some pretence-involving discourse makes no mention of pretence at all, and no awareness of the pretence (let alone any engagement in it) needs to be any component of the speaker’s attitudes or mentioned in any explanation of what the speaker is doing. Another objection that some opponents of SPIF accounts make is that pretence theorists are engaged in ‘bad old paraphrasing’ and, thus, that SPIF accounts suffer the ills that that method endures (Stanley 2001). In drawing an analogy to paraphrase, this objection focuses too much on a single aspect of a SPIF account, namely that it assigns truth conditions to sentences other than the ones those sentences seem, on the surface, to have. But the important difference between the paraphraser and the pretence theorist is that, unlike the former, the latter provides an account of how sentences get connected with the truth conditions the theorist assigns. Since the paraphraser’s inability to do that is what is really behind the objection, someone giving a SPIF account of some discourse can resist this complaint.

3 Pragmatic PIF and its problems In a series of papers, Fred Kroon (2000, 2001, 2004) has developed and defended a different sort of pretence account of some putatively problematic fragments of discourse. He defends a pragmatic pretence account, which is so described because he contends that speakers exploit the semantic content of certain expressions so as to assert or convey what is not, strictly speaking, semantically expressed through the utterance of sentences. Kroon holds that his pragmatic pretence account explains both the reason why certain parts of various sorts of utterances are problematic and why, and how, speakers manage to convey serious content by, in effect, exploiting those problematic features. We contend that there are problems with Kroon’s proposed account, problems that arise for any pretence account that is suitably characterized as pragmatic. Our aim, in this section, is to lay out Kroon’s account, to assess it, critically, and set the grounds for embracing a semantic pretence account of the sort that we favour over Kroon’s pragmatic version of PIF. We will largely

270

James A. Woodbridge and Bradley Armour-Garb

focus on the objections his account faces with the EP, but we begin our critique with a brief discussion of how PE comes back to challenge Kroon’s account. But first we turn to the account itself. 3.1 Kroon’s PIF account One of Kroon’s primary concerns is accounting for negative existential claims, such as: (20) Hamlet does not exist. He claims that the pragmatic features of (20) and its ilk manage to convey serious content, by exploiting the semantic content of ‘exists’, in order to assert what is not semantically expressed by an utterance of that sentence. Kroon claims that an utterance of (20) (or of (2)) is false, as it is a ‘quasicontradiction’ that is roughly of the form: (20’) The individual who is Hamlet and who exists, does not exist.16 On his view, in uttering (20) a speaker: ... adopts the pretence that the reference determiner underlying his use of the name ‘Hamlet’ secures reference to some individual, and hence an individual who exists, and uses the resulting interpretative tension to assert that: [(20*)] Outside of the pretence that the underlying reference determiner (for my use of ‘Hamlet’) secures reference to an individual, it fails to secure reference to any individual. (Kroon 2004, p. 19) He further claims that, through an assertion of (20), a speaker may say something true, which a hearer can then come to understand and know, even though what has been uttered is, strictly speaking, false. In more precise terms, Kroon says that we should understand this asserted content and the way in which the audience works out this content as follows. As he notes, ‘There is a striking sense in which the speaker does as if the description correctly describes the intended referent, and that he achieves his communicative purpose partly through knowing that his audience knows that he is doing as if the description is apt’ (pp. 12–13). Notice that if Kroon is right, then ordinary speakers and, for the point of such an assertion to succeed, ordinary hearers must (i) be aware of the pretence; (ii) be conversant enough with the notion of a reference determiner to use, or grasp, such a notion; and (iii) be aware of when a speaker is employing it. Accordingly, Kroon maintains that there is a shared pretence between speakers and hearers (and explicit knowledge of this pretence), when the former utter negative existentials, the serious content of which would be conveyed by an assertion of (20*).

Linguistic Puzzles and Semantic Pretence

271

Here is the picture that Kroon offers of how pretence factors into the sorts of claims presenting the linguistic puzzles that are of interest here:17 there is an ‘interpretative tension’ generated by a speaker’s utterance of a claim like (20) that tells her audience that she is only pretending. This interpretive tension arises from the fact that, if her audience takes her literally, they will have to assume that she is trying to assert a contradiction (Kroon 2001, p. 210). Kroon claims that since the speaker cannot realistically be claiming the truth of a contradiction (we assume that the speaker is not a dialetheist!), her audience should recognize that she is engaged in pretence and that, in fact, it is from within that pretence that the statements made by claims like (20) are put forward. As he notes, ‘it is the speaker’s use of the device of a blatant contradiction ... that now allows the speaker’s audience to understand that she is disavowing the thought that the world is, in relevant respects, the way her pretence depicts it to be’ (p. 211). 3.2 Problems with Kroon’s pretence account As mentioned before, we shall focus on the problems Kroon’s account faces with PE and EP. We will begin with the PE and then turn to EP, or, as will become clear, the engagement problems. 3.2.1 Kroon’s PIF account and the problem of error As we previously noted, there are a number of puzzles that arise for an ETF account of some fragment of discourse. Kroon’s view on negative existentials is that when a speaker utters a sentence like (20), what she utters is blatantly contradictory – a pragmatic contradiction.18 So, now we have it that the utterance the speaker makes is (pragmatically) contradictory. But why would a speaker go in for talking that way? That is, why utter what is a pragmatic contradiction, if what you wish the hearer to understand is something true (and, hence, non-contradictory)? Although Kroon does not answer these questions directly, we can infer what he might say in response, given what he wishes to say about the puzzle posed by informative identity sentences. In the case of informative identity sentences (e.g., (1)), Kroon (2001) proposes that ordinary speakers are making a mistake – that they could have uttered other sentences that are not plagued by the problem possessed by utterances of sentences like (1). Moreover, he seems to suggest that such a way of talking is so entrenched that even if speakers were aware of the problem, they would continue uttering these contradictory sentences. When we turn to negative existentials like (20), Kroon’s claim must be that speakers just cannot but utter sentences, like (20), that result in a pragmatic contradiction, like (20’). This is supposed to explain why, though what they aim to convey is non-contradictory, what they utter is contradictory, pragmatically speaking. But to attribute to speakers an utterance of a pragmatic contradiction (or even to take ordinary speakers to utter sentences that are

272 James A. Woodbridge and Bradley Armour-Garb

of a contradictory form), in the interest of expressing something that is true (and, hence, non-contradictory), is to attribute error, even if it is not error in the usual sense. We are not persuaded, for we are not convinced that such utterances (or assertions) must be construed as pragmatically contradictory and, thus, in an error-theoretic fashion. Such attributions of global error, whether semantic or pragmatic, are to be avoided if possible. Thus, if a non-error-theoretic approach is on offer, one that promises to do what a given error-theoretic account purports to do, it is to be preferred. We have already presented such an account in Section 2, so we take this point as one reason for preferring our SPIF account over Kroon’s pragmatic PIF. 3.2.2 Kroon’s PIF account and the engagement problems The engagement problem is usually presented as the claim that a given pretence-involving solution to a philosophical puzzle is implausible, since, in general, ordinary speakers, who assert the likes of (1) and (2), do not seem to be aware of, much less involved in, a game of pretence. Call this version of the engagement problem, EP1. At base, we take EP1 to arise because it seems quite implausible that ordinary speakers, through asserting the likes of sentences like (20), are even aware of, much less actively participating in, such pretences. That, however, is how Kroon understands the role of pretence in such claims. He acknowledges speakers’ awareness of pretence, when he describes them as ‘opportunistically engaged in a pretence’, and assimilates the relevant cases to ones where he acknowledges that a speaker ‘achieves his communicative purpose partly through knowing that his audience knows that he is doing as if the description is apt’. (Kroon 2004, p. 13) As mentioned, while this might be plausible when speakers are intentionally speaking figuratively, it is problematic for any account of a discourse that speakers do not typically consider figurative, e.g., those of interest here. The quote in the previous paragraph shows that we can extend EP1 as directed at Kroon’s pragmatic pretence account. He not only contends that speakers are aware of and actively exploiting, a pretence, he likewise holds that hearers are aware of the pretence as well. Indeed, it is through understanding that the speaker is actively engaged in a pretence that the hearer is able to extract the serious (and true) content that the speaker manages to convey. But, as noted, such pretence awareness is simply implausible, whether it is attributed to speakers or to hearers.19 As hinted at before, Kroon’s response (2001 and 2004) to this (familiar) problem is that, in at least some cases, speakers just have to be pretending and hearers just have to be aware of the pretence, in order to be in a position to grasp the serious content of what speaker’s have managed successfully to convey. That is, in at least some relevant cases, it is implausible in the extreme to contend that speakers (and hearers) are not both aware of, and engaged in, such a pretence.

Linguistic Puzzles and Semantic Pretence

273

It strikes us as implausible that speakers intentionally and consciously engage in the pretences that Kroon takes to underwrite these communicative acts, or that they take hearers to be aware of, much less to process, such pretences. But we shall leave that concern aside. What we wish to point out now is that if there is a pretence account that does not require this of speakers and hearers while, at the same time, providing an account of the relevant problematic discourse then Kroon’s claim would be falsified, in which case, in light of EP1, we would have a reason for favouring this alternative pretence account. We have already explained how SPIF accounts can avoid EP, so this point provides further reason for preferring the SPIF approach over Kroon’s pragmatic version of PIF. One might also claim that EP arises when what is to be pretended is outside of the ken of ordinary language users.20 This leads to a further problem, which Kroon does not address directly – what we might call the sophistication complaint. This complaint drives the second engagement problem, what we will call EP2. To see EP2, notice that when Kroon provides the serious content of what is asserted by an utterance of (20), not only does he require that speakers be aware of the hypothesized pretence, he also takes the content of such an assertion to explicitly mention (and, in some cases, quantify over) the pretence. Given the serious content of an assertion of a negative existential, the serious content itself includes the notion of pretence. In fact, understanding the serious content that gets conveyed by such utterances (and assertions) requires understanding the used term ‘pretence’. But, EP2 concludes, this attributes too much to ordinary speakers and hearers. This problem is not restricted to the concept of a pretence, for it is also the case that the serious content of such sentences includes the notion of a ‘reference determiner’, along with other terms normally restricted to semantics. We are reluctant to grant that ordinary speakers and hearers have knowledge of, much less attitudes towards, the notion (much less the nature!) of reference determiners, whether reference is construed causally, deflationistically, or descriptively, as Kroon would have it. And, even if they did have such explicit knowledge, we find it unlikely that that is what they intend to convey through utterances of (20). At the very least, Kroon owes us an explanation as to why the sophistication complaint is not warranted; thus far, no such explanation has been forthcoming (hence, we worry). We take both the standard engagement problem, EP1, and the sophistication complaint, EP2, to present serious challenges to a pragmatic account of pretence of the sort that Kroon advocates. Our third variant of the problem – EP3 – raises the question of whether Kroon is entitled to put forward a pretence account which, if successful, would serve to resolve certain philosophical puzzles or, at least, some putatively problematic discourse. Here is the problem. As we have noted, Kroon claims that ordinary speakers and hearers are aware of, and are engaged in, the sort of pragmatic pretence that he proposes.

274 James A. Woodbridge and Bradley Armour-Garb

He argues for this by considering cases in which the existence of a fiction is overt, as might occur in a negative existential like (21) Superman does not exist, as said by a speaker who is talking about the Superman story. Having defended his pragmatic version of PIF for this kind of fiction-based negative existential, he goes on to claim that ‘... there is no principled difference between fiction-based negative existentials like ‘Holmes doesn’t exist’ and ordinary negative existential like ‘Homer doesn’t exist.’ (Kroon 2001, p. 216) As such, he proposes that his fiction-based solution to the problems that (21) poses should carry over, without remainder, to those for (22), below, which is not fiction-based. One might conclude from this that, because EP seems less of a worry when we consider discourse that is itself fictionally construed, the engagement problems for ‘ordinary’ cases do not seem as pressing. One might thus take Kroon’s success at resolving the problems posed by fiction-based cases to be grist for a solution to the engagement problems. Whether Kroon’s pragmatic pretence account of putatively problematic fictional discourse is successful or not – we are inclined to think that it is not – what we wish to point out is that there is a disparallel between fiction-based and ordinary negative existential. The issue, we think, turns exactly on the final version of the engagement problem, what we will call EP3. As noted, Kroon’s pragmatic pretence account assumes that both speakers and hearers are aware of the pretence. For Kroon, this is a constant, through fiction-based and ordinary cases. The problem we find with this is that, even if it is plausible that fiction-based cases require some awareness of the presence of a pretence – manifest through the awareness that the terms have fictional uses – this awareness is not, and certainly need not be, obvious, in the ‘ordinary’ cases. To make this point, we will stick with negative existential statements. Contrast (21) with: (22) Vulcan does not exist, (where, by ‘Vulcan’, we mean the planet hypothesized in nineteenth century astronomy, not the fictional home of Mr. Spock or any of its inhabitants), as they are expanded into: (21’) Superman does not really exist (22’) Vulcan does not really exist. Notice that, while (21’) seems an acceptable expansion of (21), (22’) does not seem to be an acceptable expansion of (22), at least as ‘really’ is used in (21’). The problem is that (21’) seems acceptable because: (21’’) Superman does not exist; he is just a part of the fiction

Linguistic Puzzles and Semantic Pretence

275

is likewise an acceptable expansion of (21). But (22’) does not seem to be an acceptable expansion and, we conjecture, this is because, while: (22’’’) Vulcan does not exist; there is no such thing as ‘Vulcan’. appears to be an acceptable expansion of (22), (22’’) Vulcan does not exist; it is just a part of the fiction. does not appear to be an acceptable expansion of (22). Notice, moreover, that: (21’’’) Superman does not exist; there is no such thing as ‘Superman’. likewise does not seem an acceptable expansion of (21). Of course, while there are uses of (21’’’) that one can imagine, the problem with it seems to be that, for at least some of the uses of (21’’’) (in particular, those applicable to (22’’’)), a proper rejoinder might be that there is a Superman, though it is just a fictional character. To summarize, there is an acceptable expansion from (21) to (21’). While one can expand (22) to (22’) (in the sense that there can be a reading of (22’) that coheres with a reading of (22)), this is a very special case regarding a particular reading of (22), one that, as it happens, likewise presupposes that speaker and hearer are cognizant of a given fiction.21 The conclusion we draw from this is that there is a principled difference between (21) and (22). As we would describe it, the difference regards the fact that semantic competence with the relevant use of ‘Superman’ requires awareness that the term is fictional,22 whereas this is not the case in the relevant use of ‘Vulcan’. If that is right, it explains why a speaker and a hearer would be aware that certain uses of ‘Superman’-involving sentences are uttered and understood only given the awareness of the fiction, whereas the same will not be true of ordinary uses of ‘Vulcan’-involving sentences, including a sentence like (22). But if a proper understanding of (21) requires awareness of a fictional use, though no such fictional use is supported, much less required, for a speaker (or hearer) to understand an ordinary use of (22), then it is at least not obvious that there is ‘no principled difference’ between a fictionbased negative existential and an ordinary negative existential. But if that is right then there is a difference between the two sorts of cases, in which case the engagement problem re-emerges. A slightly different version of the same argument is as follows: If a proper understanding of (21) requires awareness that ‘Superman’ is being used as a fictional term, though no such fictional use is required in order to understand ordinary ‘Vulcan’-uses as in (22), then it is false that there is no ‘principled difference’ between

276 James A. Woodbridge and Bradley Armour-Garb

fiction-based negative existentials and ordinary negative existentials, in which case EP re-emerges. If, as we should, we take the engagement problem seriously, then we see the problem as posing a challenge to any pretence theorist: Either explain why the engagement problem is not a problem or explain why, although it may be a problem for some views, it is not the theorist’s problem. One way of explaining why it is not actually a problem would be by showing that it is really the only way in which certain problematic discourse can be resolved. This is Kroon’s tack. Largely for the reasons we have provided in this section, we take EP to be an insurmountable problem for any pragmatic PIF account because of where it locates the pretence – in the pragmatics of the discourse, i.e., in factors pertaining to its use, specifically, those involving what speakers are doing with the sentences that they are uttering. As such, any pragmatic pretence approach will have to bite the bullet with respect to EP, at least to some extent, and therefore follow Kroon’s tack. Our riposte is the provision of a SPIF account that resolves the relevant linguistic puzzles without requiring pretence awareness.

4 SPIF accounts and further objections The foregoing section on the difficulties pragmatic PIF accounts face with PE and EP provide reason for preferring SPIF accounts, since the latter avoid these objections. To recap, SPIF accounts are not error theories in any problematic sense because the role of make-believe in the functioning of the discourse makes it possible for speakers to use the discourse to make genuinely true claims about the world outside of the pretence at work in the discourse. By the operation of the principles of generation governing the relevant game of make-believe, SPIF accounts can exploit prop-oriented make-believe to make indirectly serious assertions about the real world. It is therefore not restricted in its capacity for making true assertions just to talking about the content of the fiction. And because our understanding of SPIF accounts places all awareness of the pretence involved in the discourse into the theorist’s explanation of the talk (and not in any attitude speakers have towards the talk they employ), EP does not pose a problem for accounts of this sort. Admittedly not all proponents of SPIF accounts have been sufficiently clear on this latter issue. In particular, we find that the three most prominent semantic pretence theorists – Walton, Yablo, and Crimmins – each fall victim in some way to a variant of EP. For what follows, we briefly explain the problems these theorists face. After doing so, we return to some further challenges to proponents of SPIF, recently mounted by Kroon. 4.1 Prior versions of SPIF and the EP We are not claiming that all advocates of SPIF fall victim to the EP (as we mentioned, previously, we take it that our account avoids the problem).

Linguistic Puzzles and Semantic Pretence

277

Indeed, there is a sense in which each of the relevant fictionalists – including Kroon, Walton, Yablo, and Crimmins – do not have the problem that their accounts explicitly commit them to attributing active participation in any game of make-believe to language users who employ fragments of discourse they claim involve pretence. Rather, as we will suggest, our worry is that they have not completely avoided some aspect of EP, or they have, but at the expense of limiting the application of their proposed pretence account. 4.1.1

Walton and the EP

Walton has dealt mainly with figurative discourse (or: discourse that makes up or is about works of fiction), so some element of pretence awareness or engagement is to be expected. That said, he is adamant that speakers need not be actively engaged in make-believe (Walton 1990, pp. 406–11). But his pretence account of existence-talk is where he most needs to fend off accusations, not only of engagement but of even any pretence awareness. For one thing, it seems that his only recourse against the charge that he falls victim to an EP is by noting that he rejects introspective psychology, and also that speakers can be pretending in some sense without being aware of it. As we will explain presently, we find even this concession of Walton’s to be problematic. In our formulation of SPIF, we get away from engagement entirely by locating the pretence at the level of the theorist, who is theorizing about the relevant discourse (the ‘talk’). The theorist may talk about pretence linking up utterances with their serious content, but at no point do speakers have to be engaged in any such pretence. Indeed, just as speakers need not be aware of defence to experts securing reference for natural kind terms, they do not even have to be aware of such a pretence, nor do they even have to be disposed to acknowledge that they are engaged in a sort of pretence if asked. The problem that we find for Walton is that, even though speakers are not actively engaged in the pretence, it does seem that, on his account, they have to be pretence-aware – aware of and alluding to the pretence. So, pretence awareness becomes the real problem for him. In particular, the worry is that he has not adequately answered opponents of the approach who charge that his pretence account ultimately attributes pretence awareness to speakers engaged in the talk at issue. 4.1.2 Yablo and the EP Yablo (2001 and 2005) is very clear about the EP: He wants to have nothing to do with engagement. In response to the objection that ordinary speakers are not pretending, he notes, we are not just pretending to assert, when we say that the number of planets is 9. ... [i]f pretending is making believe, where ‘making’ signifies

278

James A. Woodbridge and Bradley Armour-Garb

an act deliberately undertaken, then the objection seems right. (Yablo 2001, p. 90) Rather than attributing such pretending to speakers, Yablo contends that they are really ‘simulating’, which he describes as being in relevant respects as if one believed, while not believing, save per accidens. More specifically, he holds that: [s]omeone is simulating belief that S if although things are in relevant respects as if they believe that S, when they reflect on the matter they find that they do not believe it; or at least are agnostic on the matter; or at least do feel the propriety of their stance to depend on their belief that S if they have one. (Ibid.) Let us call this sort of simulation dispositional engagement. While we are sympathetic to the idea behind this notion of simulation, we worry that it will only really cover certain cases – cases where it would be clear to the speaker (upon reflection) that she is making as if certain objects exist and being described in her utterance, whether or not she chooses to believe in their existence and that her utterance offers a description of them. We (2009b) have argued elsewhere for a pretence account of proposition-talk. But, short of claiming that speakers are in relevant respects as if they believe that S, we claim that it is compatible with our SPIF account that such speakers take no attitude towards (their talk about) propositions, nor towards whether anything in proposition-talk commits them to any sort of as ifness (nor to whether they feel – or even can feel – that the propriety of their stance depends on anything approaching a belief that S). 4.1.3

Crimmins and the EP

Crimmins tries to avoid engagement problems by talking about ‘shallow pretence’ (Crimmins 1998, pp. 10, 14–15). While we agree with what he says about the level of a speaker’s engagement – specifically, her lack thereof – with any pretence a way of talking involves, we worry that characterizing speakers as pretending in any sense (even ‘shallowly’) runs the risk of setting up what can become an impassable stumbling block for certain theorists suspicious of the approach. Crimmins’s discussion still suggests the speakers’ engagement in – or at least awareness of – pretence, despite his intention for it not to do so. Hence, while he is on the right track, he has not done enough to convince the sceptics that EP is not an issue that presses. We are therefore willing to give the detractors the word ‘pretend’ in this context: speakers (other than a rather small minority, including the present authors) are not pretending when they use proposition-talk (or existence-talk, identity-talk, or truth-talk). Nor need speakers intentionally allude to or even be aware of any pretence at work in any talk for which we

Linguistic Puzzles and Semantic Pretence

279

want to give a SPIF account. Still, at the level of theorizing abut how the talk functions, we should understand these fragments of discourse to be ‘as if’ ways of talking that involve a systematic dependency on how things actually are. And we should explain the kind of ‘as ifness’ involved in terms of pretence, specifically, games of make-believe. 4.2

SPIF and the modal objection

One further objection recently launched against a semantic pretence account of the sort that Crimmins puts forward has to do with whether such an account (or any SPIF account) can adequately cover the modal content of a putatively pretence-involving utterance, that is, ‘what possible situations are described by [such an] utterance’ (Kroon 2004, p. 9). Our SPIF account of existence-talk, for example, appears to yield only the actual truth-conditions of its instances, namely whether the relevant denoting expression as used in the utterance succeeds in referring to something. The objection is that this account of the serious content of existence claims seems to generate modal intuitions that conflict with our modal intu itions regarding existence-talk. This indicates a difference in modal content between the instance of existence-talk and the postulated direct expression of its putative serious content. So, the objector claims, the SPIF account of existence-talk cannot be right. The conflict in modal intuitions is supposed to be that assuming (e.g.) that Hamlet does not actually exist, then (following Kripke) necessarily Hamlet does not exist, while, assuming that ‘Hamlet’ as used in an utterance of (20) does not actually refer to anything, nevertheless, it could have referred to something. (p.10) We do not agree that modal intuitions conflict in this way, when the issues are properly understood. Given that ‘Hamlet’ is a name, and it actually does not refer, then, given the orthodoxy that names are rigid designators, that name does not refer in any possible world. This might jibe against an intuition that the name could successfully refer in some non-actual possible world, but we have to remember that the issue is whether the name as used by us here in the actual world, to talk from here about other possible worlds, refers to anything there. So, being a rigid designator, a name that does not actually refer to anything is empty when used by us here to talk about any possible world. Intuitions to the contrary are most likely conflating the possibility of a homophonic name existing in another possible world and successfully referring there. But really that is a different name (or, at minimum, a different use of a name), and so not relevant. If the denoting expression is a non-rigid designator, then it seems right to say that even if it does not actually refer to anything, it could have (unless there are other reasons for rejecting this, as there is for, e.g., ‘the least rapidly converging series’). Thus a SPIF account like ours can avoid the putative modal content problem, if we note that names are rigid designators.

280 James A. Woodbridge and Bradley Armour-Garb

4.3 SPIF and the problem of meaning Another possible objection to a SPIF account of existence-talk in particular is that it violates a constraint on ‘the universality of ‘exists’’ (p. 11). This objection can be called the problem of meaning(s). The initial objection is that our SPIF account does not give the expression ‘exists’ a meaning that adheres to the universality constraint. As a result, the objection goes, an account like ours is postulating a second, pretend meaning for the expression, in addition to a usual meaning it must have that does adhere to the constraint. Worse still, in some contexts the same expression will have to be used with both its pretend meaning and its usual meaning, to avoid contradiction.23 ‘But, contradiction or not, the idea that two different meanings are involved surely can’t be right.’ (Ibid.) We agree that two meanings cannot be involved in the use of ‘exists’, but that is because we reject the claim that ‘exists’ has a ‘usual’ meaning that does not involve pretence. Part of our SPIF account of existence-talk includes reasons for recognizing that this fragment of discourse involves pretence intrinsically because there are no pretence-free uses of its central locution, ‘exists’. Since we still recognize that the expression cannot really have an anti-extension, our account does not really violate the universality constraint, even though it holds that every use of ‘exists’ involves a pretence that the expression functions as a predicate. Nevertheless, there is still only one meaning for ‘exists’; it is just the meaning this locution gets from the pretences at work in its linguistic functioning. Thus, we deny that the problem of meaning(s) present a real objection to our account.

5

Conclusion

After setting out what we see as the dominant attempts at offering theoretical innovations in the philosophy of language, we landed on a fictionalist account, which bore some relation to what we have called the re-interpretive approach. Having briefly discussed the varieties of fictionalism currently on offer, we focused on our favoured SPIF account, which we presented and defended, as an approach that can be applied to various problematic fragments of discourse – in particular, those fragments that generate various linguistic puzzles that both launched and continue to propel work done in the philosophy of language. Starting with the particular SPIF account that we provided in Section 2.4, for uses of empty names and of existence-talk, we showed how the account might resolve the litany of linguistic puzzles that these ways of talking seem to generate, and we suggested how the approach might be extended to resolve many of the other puzzles catalogued at the beginning of this chapter. After considering some objections for our SPIF account, we showed the approach to be resistant to various objections, both those that apply to PIF, in general, and some that apply to SPIF accounts, in particular. We take

Linguistic Puzzles and Semantic Pretence

281

the approach developed here to offer a promising new form of analysis in philosophy of language, one with a broad range of application and a potential for significant illumination of the way various portions of our language manage to express content, despite certain apparent puzzles that attempting to account for this might generate.

Notes 1. For more on this issue, see Armour-Garb and Woodbridge (2009b). 2. This approach is by no means restricted to Kripke. For a more recent version, see (Field 2008). For a lucid discussion of Kripke’s approach, see (Field 2008, pp. 58–65). 3. In addition to puzzles that elicit the referential/attributive distinction, there is work, by Charles Parsons (1974), Keith Simmons (1993), and Michael Glanzberg (2001), on contextual solutions to the liar paradoxes and, by Diana Raffman (1994), Stewart Shapiro (2003), and others, on contextual solutions to the sorites. Contextual solutions to the liar (and the sorites) constitute a sort of pragmatic account, since what gets expressed by the utterance of a sentence is contextsensitive. But the very idea of context sensitivity requires the distinction between what’s conveyed and what is, strictly speaking, said (by the sentence uttered). 4. Thus Ayer’s view includes the following positions on various fragments of discourse: ‘analyticism’ about mathematics and logic – where the claims have no factual content but instead are really just definitions in use, or stipulations of use principles for certain vocabulary; phenomenalism about external object talk; emotivism about ethics; instrumentalism about scientific unobservables. 5. For a different sort of expressivism in the philosophy of language, consider Robert Brandom’s (1994) expressivist semantics. 6. Other examples of revolutionary fictionalism might include Rosen (1990) on modal discourse (although we could understand his account as hermeneutic) and Joyce (2001) on moral discourse. See also Hussain (2004). 7. Other examples might include Walton (1990) on existence-talk, Crimmins (1998) on intentional-attitude ascriptions, Yablo (2000) on talk about abstract or ‘platonic’ objects. 8. Field goes on to argue that there are nominalistic reformulations of physical theory that do not require the mathematical claims to be correct. Hence, at least that part of science could be done without numbers (contra the Quine-Putnam indispensability arguments). One of Field’s important insights is that a theory does not have to be true to be good. He claims that mathematical theories, which are not true, can still be good, in virtue of the fact that they amount to conservative extensions of non-mathematical discourse. For worries about whether Field has in fact succeeded, see (Shapiro 1983). 9. Field is an exception here. (Field 1989, p. 2) describes an approach that amounts to ETF, but he also points out that this is not the only way to understand fictionalism. 10. The contrast is with what we call purely pretend claims, claims that involve pretence but in a way that makes it the case that they express no serious content. We maintain that a diagnosis of the liar paradox, along with the other cases of semantic pathology, will centre on providing pretence accounts of the traditional semantic notions and recognizing the pathological cases as involving pure pretence. For a brief start on the idea see (Woodbridge 2005).

282

James A. Woodbridge and Bradley Armour-Garb

11. This aspect of pretence-involving utterances is sometimes called ‘piggy-backing’ (Richard 2000). 12. Higher-order levels of extrinsic pretence are possible, e.g., second-order extrinsic pretence involves merely pretending that it is to be pretended that a is F, etc. Second-order extrinsic pretence involves a change in how we regard the subject in the pretence (from being F to having the features required to be fictionally F in a first-order pretence). Third- and higher-orders of pretence involve a change in subject (from a to games of make-believe themselves) as well. 13. To see this about metaphors, consider the sentence (I) Nevada Gov. Jim Gibbons is the headliner of a bad lounge act. One could take this sentence to make a serious statement about the world directly, that is, one could take it at face value. Taking or offering (I) metaphorically involves placing the face-value reading of the utterance in the context of a pretence. Specifically, (I) invokes a pretence consisting of a game of makebelieve that prescribes pretending someone is the headliner of a bad lounge act whenever that person actually possesses certain features, features that really have nothing to do with headlining a lounge act. See (Walton, 1993) for the details of the role of make-believe in (much) metaphor. What we are adding here is a specification of the type of pretence many cases involve as extrinsic, in particular, first-order extrinsic. The utterance’s non-literal content, the serious claim it makes indirectly (namely, that Gibbons has the pretence-prescribing features), depends on an antecedent literal content that attaches to the whole utterance and the principles of generation that link prescribed pretences to realworld conditions. 14. A related but slightly different account, also based on Walton’s work on makebelieve, is developed in (Evans 1982, chapter 10). 15. Although we cannot adequately address this point here, it bears noting that the serious content here is expressed in terms of reference, which is a way of talking that will itself get a pretence account. The serious content of instances of reference-talk will, in turn, get expressed in terms of identity-talk, yet another fragment of discourse we hold gets a pretence account. The serious content of the instance of this talk will get expressed in terms of reference (and possibly existence-talk). This indicates a circle of pretence, in which the claims all express serious content indirectly, where it turns out that there is no direct, pretence-free way to express that serious content. 16. By contrast, an utterance of (20#) Hamlet exists has the force of a quasi-truism, as is roughly of the form (20#’) The individual who is Hamlet and who exists, exists. 17. Kroon’s actual focus in his (2001) presentation of this picture is plural identity claims, but he holds it for negative existentials as well. See Kroon (2004). 18. If we allow that utterances are truth-apt, then his treatment of negative existentials is properly characterized as error-theoretic. But even if not, given that (20’) arises from an utterance of (20) in virtue of the presupposition that Hamlet exists, the speaker is still, by Kroon’s account, committed to have uttered a pragmatic contradiction. 19. That Kroon is committed to pretence awareness is clear from what he says in his works (2001 and 2004). 20. So, e.g., one might, as we (2009b) have, claim that the serious content of ‘proposition-talk’ is best construed through a pretence. The detractor might

Linguistic Puzzles and Semantic Pretence

283

object if we attributed awareness of the pretence to ordinary speakers and hearers, since, she might argue, it is an open question whether, in a certain sense, they even have the concept of a proposition. 21. Put roughly, one can imagine a circumstance in which ‘Vulcan’ names a posited planet that is understood to be nothing more than an instrumental calculating device (like using geocentric astronomy for nautical navigation). But notice that in order for one to be competent with that use of ‘Vulcan’ (at least in order to be in a position to expand from (22) to (22’), like one can from (21) to (21’), but unlike a standard reading of (22)), semantic competence demands that she be aware of the fact that this use of ‘Vulcan’ is stipulated to be part of a fiction. Hence, rather than challenging our claim, this case serves to support it. 22. Indeed, consider someone who memorized Anna Karenina but who mistakenly believed that it was a book of non-fiction. Although the person could report lots of ‘Anna Karenina events’, it is evident that the speaker is not a competent user of that name. 23. Kroon (2004, p. 11) actually makes this claim about an example involving an instance of identity-talk, but it carries over to existence-talk.

Bibliography Armour-Garb, B. and Woodbridge, J. (2009a) ‘Why Deflationists Should Be PretenseTheorists (and perhaps Already are)’, in N. Pedersen, C. Wright (eds.) New Waves in Truth (Basingstoke and New York: Palgrave Macmillan, forthcoming). —— (2009b) ‘The Story about Propositions’, in preparation. Ayer, A.J. (1952) Language, Truth and Logic, 2nd edn (New York: Dover Publications). Blackburn, S. (1993) Essays in Quasi-Realism (Oxford: Oxford University Press). Brandom, R. (1994) Making It Explicit (Cambridge, Mass.: Harvard University Press). Crimmins, M. (1998) ‘Hesperus and Phosphorus: Sense, Pretense, and Reference’, Philosophical Review, 107, 1–47. Donnellan, K. (1966) ‘Reference and Definite Descriptions’, Philosophical Review, 77, 281–304. Evans, G. (1982) The Varieties of Reference, J. McDowell (ed.) (Oxford: Oxford University Press). Field, H. (1980) Science Without Numbers: A Defense of Nominalism (Princeton, N.J.: Princeton University Press). —— (1989) Realism, Mathematics and Modality (Oxford: Basil Blackwell). —— (2008) Saving Truth from Paradox (Oxford: Oxford University Press). Frege, G. (1892) ‘On Sense and Reference’, in P. Geach and M. Black (eds) Translations from the Philosophical Writings of Gottlob Frege, 2nd edn (Oxford: Basil Blackwell, 1960), 56–78. Gibbard, A. (1990) Wise Choices, Apt Feelings: A Theory of Normative Judgment (Cambridge, Mass.: Harvard University Press). Glanzberg, M. (2001) ‘The Liar in Context’, Philosophical Studies, 103, 217–51. Grice, H.P. (1989) Studies in the Ways of Words (Cambridge, Mass.: Harvard University Press). Hussain, N. (2004) ‘The Return of Moral Fictionalism’, Philosophical Perspectives, 18: Ethics, 149–87. Joyce, R. (2001) The Myth of Morality (New York: Cambridge University Press). Kalderon, M. (2005) Moral Fictionalism (Oxford: Oxford University Press).

284

James A. Woodbridge and Bradley Armour-Garb

Kripke, S. (1975) ‘Outline of a Theory of Truth’, Journal of Philosophy, 72, 690–716. Kroon, F. (2000) ‘ “Disavowal Through Commitment” Theories of Negative Existentials’, in A. Everett and T. Hofweber (eds.) Empty Names, Fiction, and the Puzzles of Non-Existence (Stanford: CSLI Publications, 2000), 95–116. —— (2001) ‘Fictionalism and the Informativeness of Identity’, Philosophical Studies, 106, 197–225. —— (2004) ‘Descriptivism, Pretense, and the Frege-Russell Problems’, Philosophical Review, 113, 1–30. Mackie, J. (1977) Ethics: Inventing Right and Wrong (London: Penguin Press). Parsons, C. (1974) ‘The Liar Paradox’, Journal of Philosophical Logic, 3, 381–412. Raffman, D. (1994) ‘Vagueness Without Paradox’, Philosophical Review, 103, 41–74. Richard, M. (2000) ‘Semantic Pretense’, in A. Everett and T. Hofweber (eds.) Empty Names, Fiction, and the Puzzles of Non-Existence (Stanford: CSLI Publications), 205–32. Rosen, G. (1990) ‘Modal Fictionalism’, Mind, 99, 327–54. Russell, B. (1905) ‘On Denoting’, Mind, 14, 479–93. Shapiro, S. (1983) ‘Conservativeness and Incompleteness’, Journal of Philosophy, 80, 521–31. —— (2003) ‘Vagueness and Conversation’, in J.C. Beall (ed.) Liars and Heaps (Oxford: Oxford University Press), 39–72. Simmons, K. (1993) Universality and the Liar: An Essay on Truth and the Diagonal Argument (New York: Cambridge University Press). Stanley, J. (2001) ‘Hermeneutic Fictionalism’, Midwest Studies in Philosophy, 25: Figurative Language, 36–71. Strawson, P. (1950) ‘On Referring’, Mind, 59, 320–44. Walton, K. (1990) Mimesis as Make-Believe (Cambridge, Mass.: Harvard University Press). —— (1993) ‘Metaphor and Prop-Oriented Make-Believe’, European Journal of Philosophy, 1, 39–56. Woodbridge, J. (2005) ‘Truth as a Pretense’, in M. Kalderon (ed.) Fictionalism in Metaphysics (Oxford: Oxford University Press), 134–77. —— (2006) ‘Propositions as Semantic Pretense’, Language & Communication, 26, 343–55. Yablo, S. (1996) ‘How in the World?’, Philosophical Topics, 24, 255–86. —— (2000) ‘A Paradox of Existence’, in A. Everett and T. Hofweber (eds) Empty Names, Fiction, and the Puzzles of Non-Existence (Stanford: CSLI Publications, 2000), 275–312. —— (2001) ‘Go Figure: A Path Through Fictionalism’, Midwest Studies in Philosophy, 25: Figurative Language, 72–102. —— (2005) ‘The Myth of the Seven’, in M. Kalderon (ed.) Fictionalism in Metaphysics (Oxford: Oxford University Press), 88–115.

Index ambiguity, 1, 51–3, 89–95, 211–14, 234 analyticity, 2, 186–204, 227–42 aposteriority, 61–2, 79, 145, 190 apriority, 61–2, 77–9, 144–5, 157, 188–90, 196–8, 251 assertion, 4, 29–31, 68, 108–33, 153–4, 255, 258–64, 269–73, 267 Ayer, A.J., 254–5

Donnellan, Keith, 254 Dummett, Michael, 2, 4, 27, 108–10, 115, 129–30, 138–9, 149–50

Bach, Kent, 33, 90, 98 Blackburn, Simon, 91, 255 Boer, Steven, 213 Burge, Tyler, 5, 47, 143–4, 190, 206–9, 212–15

Gibbard, Alan, 91, 255 Grice, Paul, 2, 3, 6, 24–37, 89, 110, 115–8, 122–4, 227–32

Caplan, Ben, 63, 77–80 character, 5, 12, 17, 89, 165–8, 177–8, 193–7 Chisholm, Roderick, 226 Chomsky, Noam, 33, 42–6 communication, 28–31, 37, 67–73, 90, 101, 108–13, 117–20 communicative intention, 2, 94–106, 108–19, 124, 227, 232 communicative success, see communication contextualism, see also vagueness, 2, 4, 8, 12–14, 24, 87–90, 94–8, 104–5 conversational implicature, see implicature Crimmins, Richard, 64, 260–8, 276–9

idiolect, 3, 47, 123, 212 implicature, 1, 2, 12, 30, 33, 105, 127, 179 indexicals, 1, 2, 5, 8–19, 25, 37, 89–90, 94–6, 127, 139, 150, 165–6, 191–9, 206 intention, communicative, see communicative intention

Davidson, Donald, 2, 24–6, 87, 108–9, 114–28, 149–53 de re attitudes, 219–22, 241 demonstratives, 5, 25, 37, 61, 176, 206–22 descriptivist, see Russell, Bertrand direct reference, 2, 3, 4, 5, 60–5, 74–80, 186, 192, 199, 251

empty terms, 213, 250, 253, 264–7, 279–80 Evans, Gareth, 27, 219 Field, Hartry, 257–8 Frege, Gottlob, 1, 2, 3, 4, 62–8, 76–8, 178, 187, 200, 250, 253–5

Higginbotham, James, 214–19

Kaplan, David, 12–13, 16–17, 61, 87, 89, 165 Keefe, Rosanna, 8, 17–18 King, Jeff, 217–19 Kripke, Saul, 2–4, 6, 61–8, 75–80, 87, 91, 186, 190, 221–3, 243–4, 253–4, 279 Kroon, Fred, 269–77 Lewis, David, 9, 110, 113, 122, 124, 187, 258 linguistic communication, see communication linguistic pragmatism, see pragmatism MacFarlane, John, 13 Mackie, John, 255 McDowell, John, 91, 150, 219 285

286

Index

McKinsey, Michael, 63–4, 68 Millar, Alan, 91, 102–4 Millian, see direct reference minimal semantics, 2, 3, 14, 24–37, 46, 90, 96–9, 190–92, 199 Neale, Stephen, 65–71, 229 normative, 4, 87–106, 111–13, 121–8, 257 Peacocke, Christopher, 145–6, 151 pragmatism, 3–4, 60, 65–9, 77–80 Putnam, Hilary, 2, 186, 190, 195–6 Quine, W.V.O., 1–2, 186, 190, 200, 209 Recanati, François, 18, 24, 32, 37, 90, 98 referentialist, see direct reference Richard, Mark, 268 Rosen, Gideon, 258 rule-governed activity, see normative Rumfitt, Ian, 153–7

Russell, Bertrand, 1, 2, 61, 186, 189, 199, 209–13, 220, 250, 253–4, 267 Schiffer, Stephen, 55 Segal, Gabriel, 212, 214–16 singular thought, see de re attitudes Soames, Scott, 12–14, 25, 60–5, 68, 73, 91, 190, 199 sorites, see vagueness speaker intention, see communicative intention Stainton, Robert, 30, 109–10, 129–33 Stanley, Jason, 30, 90, 94–5, 129, 257, 268–9 Strawson, P.F., 1, 2, 114–5, 254 vacuous terms, see empty terms vagueness, 1, 2, 8–19, 94, 252 Walton, Kendall, 259–67, 277 Williamson, Timothy, 47, 113, 121, 179, 186, 190–1 Yablo, Stephen, 257, 263, 276, 277–8