Quine and Davidson on Language, Thought and Reality

This page intentionally left blank

Quine and Davidson on Language, Thought and Reality

Quine and Davidson are among the leading thinkers of the twentieth century. Their influence on contemporary philosophy is second to none, and their impact is also strongly felt in disciplines such as linguistics and psychology. This is the first book devoted to both of them, but also the first to question some of their basic assumptions. Hans-Johann Glock critically scrutinizes their ideas on ontology, analyticity, apriority and necessity, truth, meaning and interpretation, thought and language, and shows that their attempts to accommodate meaning and thought within a naturalistic framework, either by impugning intensional notions as unclear and non-factual or by extracting them from physical facts, are ultimately unsuccessful. His discussion includes interesting comparisons of Quine and Davidson with other philosophers, particularly Wittgenstein. Glock also offers detailed accounts of central issues in contemporary analytic philosophy, such as whether philosophy is continuous with science, whether truth is a feature of sentences or propositions, whether meaning and reference are indeterminate, whether language requires conventions, and whether animals are capable of having beliefs and desires. h a n s - j o h a n n g l o c k is Reader in Philosophy at the University of Reading. He is the author of A Wittgenstein Dictionary (1996), which has been translated into German, French, Portuguese and Polish. He is editor of The Rise of Analytic Philosophy (1997), Wittgenstein: a Critical Reader (2001), and Strawson and Kant (2003), and co-editor (with Robert L. Arrington) of Wittgenstein’s Philosophical Investigations (1991), and Wittgenstein and Quine (1996).

Quine and Davidson on Language, Thought and Reality hans-johann glock University of Reading

   Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge  , United Kingdom Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521821803 © Hans-Johann Glock 2003 This book is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2003 - isbn-13 978-0-511-07039-6 eBook (EBL) -  eBook (EBL) isbn-10 0-511-07039-X - isbn-13 978-0-521-82180-3 hardback - isbn-10 0-521-82180-0 hardback

Cambridge University Press has no responsibility for the persistence or accuracy of s for external or third-party internet websites referred to in this book, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Fur ¨ Gabi

It is the profession of philosophers to question platitudes that others accept without thinking twice. A dangerous profession, since philosophers are more easily discredited than platitudes, but a useful one. For when a good philosopher challenges a platitude, it usually turns out that the platitude was essentially right; but the philosopher has noticed trouble that one who did not think twice could not have met. In the end, the challenge is answered and the platitude survives, more often than not. But the philosopher has done the adherents of the platitude a service: he has made them think twice. ( d av i d l e w i s 1 9 6 9 : 1 )

Contents

Acknowledgements page xi List of abbreviations xiii Introduction 1 1 1 2 3 4 5

Logical pragmatism 13 The impact of logical positivism 13 The legacy of American pragmatism 18 Quine’s naturalism 23 Davidson on reason’s place in nature 30 A philosophical anthropology 36

2 1 2 3 4 5 6

Ontology 40 Ontological commitment and ontological parsimony 41 Intensions and criteria of identity 47 Ontology and singular terms 52 Ontology and quantification 56 Ontological deflationism 60 Linguistic frameworks and logical paraphrase 63

3 1 2 3 4 5 6

Analyticity, apriority and necessity 71 The circularity charge 72 Holism and truth by virtue of meaning 77 A linguistic account of analyticity 81 Holism and the a priori 86 The deep need for the convention 92 Modal logic and essentialism 95

4 Truth 102 1 The legacy of Tarski 102 [ix]

x

Contents

2 3 4 5

Has Tarski defined truth? 110 Unbearable truth-bearers 118 Epistemic and correspondence theories 126 Deflationism and minimalism 131

5 1 2 3 4 5 6 7

Meaning and truth 137 Quine on reference, meaning and use 137 Davidson’s truth-conditional theory of meaning 142 Meaning and truth-conditions 151 The immanence of meaning 154 The notion of a truth-condition 155 The individuation problem 156 Truth-conditional semantics and natural languages 157

6 1 2 3 4

Radical translation and radical interpretation 166 Translation runs amok 170 Quine’s closet hermeneutics 175 On safari with Davidson 182 Charity is not enough 194

7 1 2 3 4 5

Indeterminacies 200 Does radical translation start at home? 201 Unsuccessful objections to indeterminacy 207 Scrutinizing reference I: gavagai 214 Scrutinizing reference II: proxy functions 219 Indeterminacy and underdetermination 224

8 1 2 3 4 5

Meaning and understanding 233 The extensionality problem 234 The privations of modesty 239 Psychological modesty 244 Language, norms and conventions 250 Thought, causation, and first-person authority 259

9 1 2 3 4

Thought and language 268 The intensional nature of thought 270 Animals, thoughts and concepts 274 The holistic nature of thought 281 Belief and the concept of belief 286 References 294 Index 301

Acknowledgements

In writing this book, I have incurred many debts. I am grateful for permission to use material from the following articles of mine: ‘Necessity and Normativity’, in H. Sluga and D. Stern (eds.), The Cambridge Companion to Wittgenstein (Cambridge University Press), 198–225; ‘On Safari with Wittgenstein, Quine and Davidson’, in R. L. Arrington and H. J. Glock (eds.), Wittgenstein and Quine (London: Routledge, 1996), 144–72; ‘Truth without People?’, Philosophy 72 (1997), 85–104; ‘Animals, Thoughts and Concepts’, Synthese 123 (2000), 35–64; ‘Does Ontology Exist?’, Philosophy 77 (2002), 235–60. As always, I owe thanks to my colleagues and graduate students at Reading for their forbearance and support. I am especially indebted to Javier Kalhat for his assistance at the final stages. Hilary Gaskin of CUP has been very patient and efficient throughout. Two anonymous readers for the Press improved the book through their fair and constructive reports. Over the years, I have learnt more than I can say from the meetings of the St John’s College discussion group at Oxford. Maria Alvarez, Bob Arrington, Jonathan Dancy, Kathrin Gluer, Anthony Kenny, John Preston, ¨ Bede Rundle, Stuart Shanker, Ralph Stoecker and Peter Strawson have greatly improved my philosophical understanding through conversations and comments on material contained in this book. Peter Strawson’s philosophy of language has also been a general source of inspiration to me. I am grateful to the participants of a graduate seminar on Davidson at Queen’s University (Ontario) in 1997, that I conducted jointly with David Bakhurst. David was an exemplary host, and has been an esteemed philosophical companion ever since my first baffling encounters with the writings of Quine and Davidson at Oxford in 1983. He also commented on parts of the final draft.

[xi]

xii

Acknowledgements

The bulk of this book was written at the University of Bielefeld between 1998 and 2000. I am very grateful to the Alexander von Humboldt Foundation for awarding me a research fellowship. I am also indebted to my host Eike von Savigny, for his immensely generous support over the years, and for constantly keeping me on my toes philosophically speaking. Ansgar Beckermann honoured and helped me by devoting a Ph.D. seminar to my drafts, as well as through his astute and informative comments. Christian Nimtz is not just a flesh-and-blood encyclopaedia of analytic philosophy. He was also brave enough to confront my obtuseness in dealing with computers and my obstinacy in resisting the allure of naturalism. My discussions of formal semantics have benefited greatly from his suggestions. Back at Reading, I was privileged to run a seminar on truth together with Wolfgang Kunne. I have profited immeasurably from his writings, ¨ from our conversations, and from his exacting scrutiny of several chapters, which saved me from many a blunder. For many years now, my friend John Hyman has tried his best to bring out the best in me, not least by reading most of the chapters, sometimes in several versions. I wish I could have done justice to his probing questions, his acute criticisms and his ingenious proposals. As things stand, I can only thank him. I would not even know how to thank Peter Hacker. He set me on the thorny but rewarding path of analytic philosophy. His intellectual and moral support have been unstinting, and his comments on virtually the whole of the book have been invaluable. My greatest debt, finally, is to Gabi, Sonja and Helen. They had to put up with yet another ordeal, not to mention many journeys, and still managed to comfort and inspire me through their love, kindness and good humour.

Abbreviations

Works by Quine

Books From a Logical Point of View (Cambridge/Mass.: Harvard University Press, 1980; 1. edn. 1953). ML Methods of Logic (London: Routledge & Kegan Paul, 1966; 1. edn. 1952). OR Ontological Relativity and Other Essays (New York: Columbia University Press, 1969). PL Philosophy of Logic (Englewood Cliffs/N.J.: Prentice-Hall, 1970). PT Pursuit of Truth (Cambridge/Mass.: Harvard University Press, 1992; 1. edn. 1990). Q Quiddities: An Intermittently Philosophical Dictionary (London: Penguin, 1987). RR The Roots of Reference (La Salle: Open Court, 1974). SLP Selected Logic Papers (New York: Random House, 1966). SS From Stimulus to Science (Cambridge/Mass.: Harvard University Press, 1995). The Time of My Life: An Autobiography (Cambridge/Mass.: MIT Press, TML 1984). TT Theories and Things (Cambridge/Mass.: Harvard University Press, 1981). WB The Web of Belief, with J. S. Ullian (New York: Random House, 1970). WO Word and Object (Cambridge/Mass.: MIT Press, 1960). WP Ways of Paradox and Other Essays (Cambridge/Mass.: Harvard University Press, 1976; 1. edn. 1966). FLPV

[xiii]

xiv

List of abbreviations

CBG EEW ER FM IPO ITA LAP LoC MR

MVD N NNK PLSP

PPE

PPLT PTF RDH RHS RI RIT RLS

Other items ‘Comments’, in Barrett and Gibson (eds.), 1990. ‘On Empirically Equivalent Systems of the World’, Erkenntnis 9 (1975), 313–28. ‘Events and Reification’, in LePore and McLaughlin (eds.), 1985, 162–71. ‘Facts of the Matter’, in R. W. Shahan and C. Swoyer (eds.), Essays on the Philosophy of W. V. Quine (Hassocks: Harvester, 1979), 155–69. ‘In Praise of Observation Sentences’, Journal of Philosophy 90 (1993), 107–16. ‘Indeterminacy of Translation Again’, Journal of Philosophy 84 (1987), 5–10. ‘Let Me Accentuate the Positive’, in A. Malachowski (ed.), Reading Rorty (Oxford: Blackwell, 1990), 117–19. ‘Lectures on Carnap (Harvard University, November 1934)’, in Creath (ed.), 1990, 47–103. ‘Methodological Reflections on Current Linguistic Theory’, in G. Harman and D. Davidson (eds.), Semantics of Natural Language (Dordrecht: Reidel, 1972), 442–54. ‘Mind and Verbal Dispositions’, in Guttenplan (ed.), 1975, 83–95. ‘Naturalism – or Living Within One’s Means’, Dialectica 49 (1995), 251–61. ‘The Nature of Natural Knowledge’, in Guttenplan (ed.), 1975, 67–81. ‘W. V. Quine: Perspectives on Logic, Science and Philosophy: Interview with B. Edminster and M. O’Shea’, Harvard Review of Philosophy 4 (1994), 47–57. ‘The Pragmatists’ Place in Empiricism’, in R. J. Mulvaney and P. J. Zeltner (eds.), Pragmatism (Columbia: University of South Carolina Press, 1981), 23–39. ‘Philosophical Progress in Language Theory’, Metaphilosophy 1 (1970), 1–19. ‘Progress on Two Fronts’, Journal of Philosophy 43 (1996), 159–63. ‘Replies’, in Davidson and Hintikka (eds.), 1969. ‘Replies to Critics’, in Hahn and Schilpp (eds.), 1986. ‘Responses’, Inquiry 37 (1994), 495–505. ‘On the Reasons for Indeterminacy of Translation’, Journal of Philosophy 67 (1970), 178–83. ‘Reactions’, in Leonardi and Santambrogio (eds.), 1995, 345–61.

List of abbreviations

SAO SN TI WPO WWD

‘Semantics and Abstract Objects’, Proceedings of the American Academy of Arts and Sciences 80 (1951), 90–6. ‘Structure and Nature’, Journal of Philosophy 89 (1992), 5–9. ‘Three Indeterminacies’, in R. Barrett and R. Gibson (eds.), 1990, 1–16. ‘Whither Physical Objects’, in R. S. Cohen et al. (eds.), Essays in Memory of Imre Lakatos (Dordrecht: Reidel, 1976), 497–504. ‘Where Do We Disagree?’, in Hahn (ed.), 1999, 73–9.

Works by Davidson

EAE ITI SIO

COT

CT DD EX FDT IA LL

MTE NDE PAV PCT

Books Essays on Actions and Events (Oxford University Press, 1980). Inquiries into Truth and Interpretation (Oxford University Press, 1984). Subjective, Intersubjective, Objective (Oxford University Press, 2001). Other items ‘The Conditions of Thought’, in J. Brandl and W. L. Gombocz (eds.), The Mind of Donald Davidson. Grazer Philosophische Studien 36 (Amsterdam: Rodopi, 1989), 193–200. ‘The Centrality of Truth’, in J. Peregrin (ed.), Truth and its Nature (if any) (Dordrecht: Kluwer, 1999), 105–15. ‘Davidson, Donald’, in Guttenplan (ed.), 1994, 231–6. Expressing Evaluations, The Lindley Lecture (monograph), University of Kansas Press, 1984. ‘The Folly of Trying to Define Truth’, Journal of Philosophy 93 (1996), 263–78. ‘Intellectual Autobiography’, in Hahn (ed.), 1999, 3–79. ‘Locating Literary Language’, in R. W. Dasenbrock (ed.), Literary Theory after Davidson (University Park: University of Pennsylvania Press, 1993), 295–308. ‘Meaning, Truth and Evidence’, in Barrett and Gibson (eds.), 1990, 68–79. ‘A Nice Derangement of Epitaphs’, in LePore (ed.), 1986, 433–46. ‘Post-Analytic Visions’, in G. Borradori (ed.), The American Philosopher (Chicago: University of Chicago Press, 1994), 40–54. ‘Pursuit of the Concept of Truth’, in Leonardi and Santambrogio (eds.), 1995, 7–21.

xv

xvi

List of abbreviations

PEA

QVT RD RH RI RQE SAL

SCT STL UT

‘Problems in the Explanation of Action’, in P. Pettit, R. Sylvan and J. Norman (eds.), Metaphysics and Morality: Essays in Honour of J. J. C. Smart (Oxford: Blackwell, 1987), 35–49. ‘What is Quine’s View of Truth?’, Inquiry 37 (1994), 437–40. ‘Replies’, in Stoecker (ed.), 1993. ‘Replies to Critics’, in Hahn (ed.), 1999. ‘Representation and Interpretation’, in M. Said et al. (eds.), Modelling the Mind (Oxford University Press, 1990), 13–26. ‘Reply to Quine on Events’, in Lepore and McLaughlin (eds.), 1985, 172–6. ‘The Social Aspect of Language’, in B. McGuinness and G. Oliveri (eds.), The Philosophy of Michael Dummett (Dordrecht: Kluwer, 1994), 1–16. ‘The Structure and Content of Truth’, Journal of Philosophy 87 (1990), 279–328. ‘Seeing Through Language’, in Preston (ed.), 1997, 15–27. ‘Towards a Unified Theory of Meaning and Action’, Grazer Philosophische Studien 11 (1980), 1–12.

Introduction

Quine and Davidson are among the leading philosophers of the twentieth century, and their current influence on analytic philosophy is second to none. The reason for this judgement is not just that many contemporary philosophers accept their findings. It is first and foremost that they have fundamentally altered the terms of debate within analytic philosophy. Even those who resolutely reject their views often define their own positions in relation to them. No philosopher can afford to ignore them, and their impact is strongly felt in other disciplines, notably linguistics and psychology. As far as I know, this is the first book devoted to both Quine and Davidson. It is an attempt to elucidate and critically assess their contributions to the philosophy of language, the philosophy of mind and metaphysics. I want to make out a case for the following claims: First, these contributions are best seen in conjunction. Quine provides the acknowledged starting-point for Davidson. Davidson rejects aspects of Quine’s position – especially his eliminativism, certain aspects of his extensionalism, his behaviourism and his empiricist invocation of neural stimulations. At the same time, he accepts many of Quine’s fundamental claims – notably the rejection of the analytic/synthetic distinction, his thesis of the indeterminacy of translation, and his suspicion of the notion of meaning. He also develops other Quinean ideas in powerful and illuminating ways, such as the thought experiment of radical translation, the connection between meaning and communication, and the attack on linguistic conventions. Secondly, Quine and Davidson can profitably be seen as logical pragmatists. They have been influenced by logical positivism on the one hand, and by American pragmatism on the other. What holds together the

[1]

2

Quine and Davidson on Language, Thought and Reality

apparently diverse strands of their work is a philosophical anthropology, a powerful conception of human beings and of human action. Both hold that human thought and human rationality are embodied in human practice, and especially in linguistic behaviour. For this reason, their philosophical anthropology in turn revolves around a conception of language. Furthermore, this conception of language combines the formal semantics developed by Tarski and Carnap with the pragmatist idea that language is essentially a form of human behaviour. They regard language as a process of communication and interaction, but insist that it can nevertheless be elucidated with the help of abstract logical calculi. Thirdly, Quine’s and Davidson’s picture of human practice in general and of linguistic behaviour in particular combines profound insights with serious distortions. The distortions arise partly because of their acceptance of certain orthodoxies of formal semantics, partly because of their pragmatist tendency to reject any distinction between conceptual or philosophical issues and factual or scientific issues, and partly because Davidson occasionally veers from the Scylla of Quine’s empiricism to the Charybdis of an excessive rationalism, for example in his principle of charity and in his denial of non-linguistic thought. Finally, both Quine and Davidson try to accommodate higher phenomena such as meaning and thought within a naturalistic framework. Quine does so by impugning intensional concepts as unclear and nonfactual, and by eliminating them from the ‘canonical notation’ of scientific philosophy. Davidson, because he acknowledges the legitimacy of intensional discourse, perforce does so in a more roundabout manner. He tries to extract ascriptions of thought and meaning from the physical facts he considers to be more fundamental, but in a way that stops short of downright reduction. Intensional notions are not part of the basic vocabulary by which we describe nature, but we are entitled to apply them to sufficiently complex patterns of physical phenomena, such as the sounds and movements of speakers. Though attractive and prima facie plausible, both these naturalistic projects are misguided in my view. There is nothing problematic about higher phenomena. They are just as real and factual as those described by the natural sciences, even though they presuppose creatures with a distinctive range of cognitive and linguistic capacities. My primary concern is with the philosophy of language and its implications for the philosophy of mind, metaphysics and the nature of philosophy. This means that important parts of Quine’s and Davidson’s work

Introduction

will only be discussed in so far as they bear on these central topics. Quine’s seminal contributions to formal logic are mentioned only to the extent that they impinge on his philosophical logic. Equally, Davidson’s anomalous monism features mainly as part of the background to his brand of naturalism, his discussion of radical interpretation and his reflections on thought and language. Limitations of space have also prevented me from dwelling on the issue of scepticism. Furthermore, I shall ignore certain details. Thus I shall not attempt to cover all of Davidson’s attempts to bring recalcitrant idioms of natural languages within the purview of a Tarskian truth-theory, leaving aside, for example, his account of metaphors. I shall also be brief about Quine’s behaviourist theory of the onto- and phylogenesis of thought and language. In my view this is the least interesting and durable part of his work. It is largely speculative, these speculations have been forcefully contested by nativists, and to establish the truth of the matter one would have to draw on empirical research beyond the scope of this book. In recent years there has been a veritable flood of secondary literature on Quine and Davidson, a flood that bears witness to the enormous importance and fecundity of their work. Any attempt to survey all of it would be futile. There are several collections of critical essays, which shed valuable light on Quine’s and Davidson’s work, especially when they include their responses. However, most of these essays concentrate on specific topics, rather than on general features of their philosophical systems. There have also been several book-length studies, especially of Quine. Some of them are devoted to specific issues, but others try to present either Quine’s or Davidson’s work in its entirety, and to bring out the connections between the various parts (Gibson 1982; Hookway 1988 and Orenstein 2002 provide good introductions to Quine; Evnine 1991 does the same for Davidson). However, there is no book on both Quine and Davidson. Furthermore, the bulk of the secondary literature does not challenge certain basic assumptions of their work, assumptions that have become orthodoxies in post-positivist analytic philosophy. The objections raised in critical articles are sometimes convincing, but they are often scholastic and tend to concern details rather than the overall outlook. Understandably, most of the books have been written by followers with the aim of vindicating either Quine or Davidson. In my view, by contrast, Quine’s and Davidson’s arguments are often uncompelling and their conclusions partly mistaken. Some contemporary practitioners believe that the hallmark of analytic philosophy is its

3

4

Quine and Davidson on Language, Thought and Reality

preoccupation with knockdown arguments. This picture is lopsided. Like Frege, Russell and Wittgenstein, Quine and Davidson are great analytic philosophers, not because they have provided a series of indisputable demonstrations, but because they have questioned deep-seated assumptions in a way that is striking, innovative and illuminating. At the same time, I regard some of the assumptions questioned by Quine and Davidson as both sound and important. In sober philosophizing, we should not abandon intensional concepts, or the distinction between conceptual and factual issues, or the idea of linguistic rules, for instance, without compelling reasons. Quine and Davidson have not provided such reasons, or so I shall argue. My own thinking on the matters under discussion has been influenced by Wittgenstein. This is no coincidence. At a tactical level, there are interesting similarities between Wittgenstein on the one hand, and our protagonists on the other, for example, between Wittgenstein’s discussion of rule-following and their discussion of radical translation/interpretation, or in their reactions to the analytic/synthetic distinction. There are also parallels at the strategic level. In strikingly different ways, all three are part of the so-called linguistic turn of analytic philosophy: they ascribe a central philosophical role to language, albeit for different reasons. Furthermore, all three propound conceptions of language and the mind that shun both Platonism and Cartesianism. Linguistic expressions acquire meaning not by being associated with either private mental processes or abstract entities, but by having a certain role within communication. Finally, in all of them there are important pragmatist themes, first and foremost the stress on the philosophical importance of human action. From a bird’seye view of contemporary philosophy in the English-speaking world, the three may appear comparatively close, at least relative to the revival of traditional metaphysics on the one hand, and the dissolution of philosophy into research programmes in cognitive science and AI on the other. The differences are equally important. Wittgenstein doggedly repudiates the assimilation of philosophy and science that Quine preaches, even though he does not practise it. He also develops a picture of human language and behaviour that eschews the austere naturalistic and behaviourist tendencies characteristic of Quine and, to a lesser extent, of Davidson. Finally, both of them set great store by the power of formal logic. The later Wittgenstein, by contrast, was sceptical about the philosophical value of logical analysis, notwithstanding the fact that his own Tractatus was one of the major inspirations behind formal philosophizing.

Introduction

In my view, Wittgenstein’s work contains the seeds of powerful objections to important claims in Quine and Davidson. However, I shall not presuppose Wittgensteinian ideas in this book. Even when I use some of them, I try to develop them ab novo, out of a discussion of Quine and Davidson. Moreover, other thinkers are equally important to my case. There is an anti-naturalist tradition in analytic philosophy, which refuses to collapse philosophy (logic, semantics, epistemology) into natural science. Its godfathers are Kant, Frege and Wittgenstein, and it includes Carnap, Ryle, Grice, Strawson, Dummett, the later Putnam and contemporary Wittgensteinians. Quine and Davidson pose a powerful challenge to this tradition. In trying to meet that challenge, I also hope to develop afresh some insights of that anti-naturalist tradition. Questions of inspiration aside, the aim must of course be to criticize Quine and Davidson in a way which is rationally compelling. Ideally, such a critique should be immanent: rather than confront opponents with dogmatic assumptions of one’s own, one should point only to internal inconsistencies in their positions. There are limits to such a procedure, especially when one is dealing with extremely sophisticated and resourceful thinkers like Quine and Davidson. Sometimes there is no alternative to resting one’s case with assumptions that strike one as plausible. But I shall try to work my way towards fundamental objections by starting out from internal tensions and unclarities. In this way, the book should be of interest even to those who remain unconvinced by my more sweeping criticisms, or to those who would attack Quine and Davidson from a perspective that is more rather than less naturalistic than their own. On some points, I shall argue, Quine and Davidson are simply wrong. This creates a problem. Being simply wrong is the fate of lesser mortals. Great philosophers instead suffer the indignity of being constantly misunderstood. In this time-honoured tradition, Quine, Davidson and their followers occasionally seem to think that any radical criticism must be based on misunderstanding. The risk of misinterpretation is real. But part of the blame must lie with our protagonists. They have many philosophical virtues. Yet excessive sensitivity to tensions in their own work, whether they be synchronic inconsistencies or diachronic changes of mind, is not one of them. Nor is this to be expected in such original and prolific authors. Commentators who believe that one must never ascribe inconsistent views to an author have, I suspect, never bothered to reread their own writings.

5

6

Quine and Davidson on Language, Thought and Reality

Austin once remarked: ‘There’s the bit where you say it, and there’s the bit where you take it back’ (1962: 2). In the case of our protagonists, the latter bit is often temporally remote from the former, and not advertised as a recantation. Both of them command an attractive philosophical style; indeed, Quine’s prose is among the chief glories of analytic philosophy. Nevertheless, their manner of presentation can be confusing. It is often unclear how the various pieces of the jigsaw puzzle are meant to fit together. They also have a habit of approaching the same topic from slightly divergent angles, which may leave readers in the lurch over whether any substantial change of mind has taken place. In such cases it is preferable to focus initially on the original statements, which are straightforward and provocative, and to postpone discussion of subsequent modifications. Philosophers can be assessed not just for what they meant to write or should have written, but also for what they actually did write. Quine and Davidson are justly famous or infamous for views which are both extreme and extremely interesting, for example that the unit of justification is the whole of science, that words have meaning only in the context of a sentence, that there is no difference in principle between logic and physics, that intensional vocabulary has no place in the description of reality, that there is no fact of the matter as to what we mean or refer to by our words, that there is no such thing as language, etc. While it is important that they have modified some of these claims, it is equally important to subject the original claims to critical scrutiny. For these claims have exerted a tremendous influence within analytic philosophy and beyond. At the same time, I have tried to do justice to the complexity and development of their positions, and I have provided copious quotes and references, so that readers can judge for themselves. My ambition is to present and criticize Quine and Davidson in an unassuming and uncluttered manner. The book is aimed mainly at graduate students and professionals. Yet it also aspires to be accessible to undergraduates in their final year who have no prior acquaintance with Quine or Davidson, provided that they are familiar with the predicate calculus. I am mainly interested in the relation between Quine and Davidson, and in the substantive merits of their positions. For this reason, the chapters are divided not according to author, but according to topic. Some of these topics turn out to be mainly Quinean (ontology, analyticity, indeterminacy), others predominantly Davidsonian (meaning and truth, meaning and understanding, thought and language), and for yet others they both are equally important (truth, radical translation/interpretation).

Introduction

The topics are presented in rough chronological order. But there is also a natural progression. After a partly historical introduction (ch. 1), we start out with the ontological and metaphilosophical concerns that first brought Quine to philosophical fame (chs. 2 and 3), and then tackle the concept of truth (ch. 4). This provides the essential background for Quine’s and Davidson’s account of meaning (ch. 5), which in turn is intimately connected to their discussion of radical translation/interpretation (chs. 6–7). Finally, we consider the implications of that discussion for the nature of linguistic communication (ch. 8) and the relation of thought and language (ch. 9). Chapter 1, ‘Logical pragmatism’, sketches the two main sources of Quine’s and Davidson’s work, logical positivism and American pragmatism, and indicates how they combine to form what I call logical pragmatism. It provides an initial sketch of Quine’s naturalism, distinguishing epistemological naturalism, the claim that there is no knowledge outside of natural science, from ontological naturalism, the claim that there is no reality other than that investigated by natural science. This position is contrasted with Davidson’s combination of an ontological monism with a conceptual dualism, and with his attempt to extract higher from lower level phenomena. I end by suggesting that both positions revolve around a philosophical anthropology that involves a striking conception of human behaviour in general and of linguistic behaviour in particular. Chapter 2, entitled ‘Ontology’, turns to Quine’s naturalistic conception of ontology as spelling out and reducing the ontological commitments of science. It defends a version of his standard of ontological admissibility – ‘No entity without identity’ – while raising various objections to his criterion of ontological commitment, according to which we are committed to all and only those things that feature as values of our objectual quantifications. This criterion of ontological commitment is irredeemably intensional, contrary to Quine’s preference for a purely extensional ontology; it wrongly sets aside singular terms and the existential implications of predicates and sentences; it is also wrong in holding that we treat everything that we refer to or quantify over as existent. Such findings lend support to a deflationary approach to ontology, according to which reference to and quantification over abstract objects, ‘intensions’ included, is harmless, calling for clarification rather than elimination. Finally, I shall argue that Quine’s method of logical paraphrase is at best a contribution to conceptual clarification, not to the scientific investigation of reality he seeks to emulate.

7

8

Quine and Davidson on Language, Thought and Reality

Chapter 3, ‘Analyticity, apriority and necessity’, tackles the source of Quine’s naturalistic conception of ontology, namely his attack on the analytic/synthetic distinction, and his more general claim that from a logical point of view there is no qualitative difference between the empirical propositions of natural science and the allegedly a priori propositions of logic, mathematics and philosophy. Quine undermines many received versions of this distinction, including that of the logical positivists. At the same time, his famous circularity charge against analyticity, namely that it can only be explained through other intensional notions, is itself circular; it assumes what it purports to establish, namely that intensional notions are obscure. Furthermore, one can distinguish between conceptual and factual propositions without subscribing to the dubious idea of ‘truth by virtue of meaning’ or running foul of Quine’s epistemic holism, in so far as the latter is warranted. Ultimately, Quine’s attack on analyticity rests on an austere behaviourism which denies that language is structured by rules or conventions; but this rejection of normativity cannot account for logical connections or meaningful discourse. The final section underwrites, if not Quine’s specific attacks on modal logic, then at any rate his more general rejection of essentialist metaphysics. The next chapter is devoted to the topic of truth. Tarski’s semantic theory of truth provides the basis not just for Quine’s and Davidson’s accounts of truth, but also for Davidson’s whole philosophy of language, and with it for the formal semantics that has come to dominate analytic philosophy. It is customary to regard Davidson’s discussion of truth as a mere prolegomenon to his theory of meaning. But while meaning is Davidson’s ultimate concern, he has recently devoted a lot of space to truth in its own right. Together with Quine’s treatment, these remarks form an important topic, and one which I hope to elucidate more thoroughly than previous secondary literature. I first introduce Tarski’s semantic theory and its reception by Quine and Davidson. I further argue that as an account of truth the semantic theory suffers from even greater limitations than they have been willing to admit, yet without subscribing to Davidson’s recent claim that the concept of truth is so basic as to be indefinable. The sequel attacks an idea shared by all three, namely that the bearers of truth and falsehood are sentences, while at the same time defending Davidson’s neo-pragmatist claim that the notion of truth is essentially linked to human activity. Next, I applaud their twofold – although by no means unequivocal – repudiation of both correspondence and epistemic theories. I end by arguing against Davidson

Introduction

that the proper moral is a minimalist account of truth, though not the one implicit in Quine’s disquotational theory. Chapter 5, ‘Meaning and truth’, briefly discusses Quine’s approach to reference, meaning and use, as well as his contextualist claim that sentences rather than words are the primary vehicles of meaning. It then moves on to Davidson’s project of a theory of meaning for natural languages, and his thesis that such a theory is furnished by a Tarskian truththeory. I acknowledge that there is a connection between meaning and truth-conditions, while also raising several qualms about the way in which Davidson develops this connection. The final section discusses the obstacles to applying a truth-theory to natural languages. I remain sceptical about the attempt to read the predicate calculus into natural languages, and in particular about Davidson’s attempt to bring non-declarative sentences within the range of a truth-conditional theory through his paratactic theory of mood. The next two chapters turn to the topic for which our protagonists are currently best known, namely radical translation (Quine) or radical interpretation (Davidson), that is, the interpretation of a completely unknown language. Quine and Davidson use this heuristic device to ensure that we approach linguistic behaviour from an austere perspective, one which does not presuppose those semantic notions which they regard as problematic. In Quine, this approach is behaviouristic. It is supposed to clinch his attack on intensional notions like analyticity, because it leads to the thesis that translation is indeterminate – incompatible translations can be equally compatible with the semantically relevant facts – and hence to the conclusion that there are no criteria of identity for meanings or intensions. In Davidson, radical interpretation serves the purpose of showing that one can construct a truth-conditional theory of meaning for a natural language without assuming any prior understanding of its speakers. Chapter 6 ‘Radical translation and radical interpretation’ is devoted to the hermeneutic credentials of these two projects. It argues that neither radical translation nor radical interpretation are feasible methods for translating from scratch. They are either unsuitable to the task, or trade on assumptions that prevent them from being truly radical. I start by introducing radical translation and Quine’s famous thesis that it is indeterminate. Next I argue that Quine’s behaviourist method cannot even reach the meagre results he countenances, without tacitly relying on hermeneutic methods and intensional notions he officially disowns, not least because the pivotal notion of assent must itself be both intensional and intentional.

9

10

Quine and Davidson on Language, Thought and Reality

Davidson tends to accept this point, but he holds on to the idea that we can identify assent without any prior assumptions about the native interpretees (e.g. concerning their perceptual capacities, desires and forms of communication). Instead of assuming that the natives share most of our beliefs, as his principle of charity bids us, we should assume that they share with us certain basic human capacities and propensities. The need to make specific anthropological assumptions counts against the Davidsonian project of deriving interpretations from basic physical data. It can also be invoked against the Quinean indeterminacy thesis. To do so is the purpose of chapter 7, ‘Indeterminacies’. I begin by denying Quine’s notorious claim that radical translation ‘starts at home’, which implies that even utterances of one’s own language are subject to indeterminacy. Next I defend the indeterminacy thesis against some prima facie plausible objections, namely that it is self-defeating, that it confuses the epistemological question of whether we can identify meanings with the ontological question of whether they exist, and that it rests on a misguided third-person perspective according to which semantic properties must be manifest in behaviour. I for my part attack the indeterminacy thesis by adopting a more realistic approach to translation from scratch, and by questioning its semantic and methodological assumptions. The so-called ‘argument from below’ for the inscrutability of reference can be blocked by taking into account the common human nature that unites us with the natives, and by abandoning the contextualist dogma that only the properties of whole sentences but not those of their components are semantically relevant. Quine’s so-called ‘argument from above’ can be reduced to a harmless instance of the underdetermination of theory by evidence. Instead of establishing that there is no fact of the matter as to what our expressions mean, the indeterminacy thesis presupposes that all facts are physical. This physicalist assumption renders Quine’s attack on intensional discourse circular once more, and it is inherently dubious. Chapter 8 looks at the way in which Davidson uses radical interpretation to shed light on ‘Meaning and understanding’. It starts out by arguing that the appeal to radical interpretation and its corollaries cannot solve the frequently diagnosed ‘extensionality problem’, namely of transforming an extensional theory of truth-conditions into a theory of sentence meaning. The next section is devoted to the debate between Dummett and Davidson over whether a theory of meaning should be ‘modest’. I side with Dummett in renouncing ‘explanatory modesty’: the theorems of a theory of meaning must not be disquotational trivialities, but should explain

Introduction

the concepts involved. I side with Davidson in defending ‘psychological modesty’: a theory of meaning should not appeal to tacit knowledge and inferences of a kind that competent speakers cannot even become aware of. But I shall also complain that Davidson is not consistent on this issue. The third section defends the idea of a common language and of shared linguistic conventions against Davidson’s attack in ‘A Nice Derangement of Epitaphs’. Finally, I turn to Davidson’s recent idea that the ‘triangulation’ between speaker, hearer and an objective world furnishes both a causal theory of meaning or content and an explanation of the first-person authority we have concerning our own thoughts and utterances. Chapter 9 is devoted to ‘Thought and language’. Like Quine, Davidson maintains that thoughts must be capable of being manifested in behaviour. Unlike Quine, he insists that such behaviour must include linguistic behaviour. As a result, he denies that non-linguistic creatures are capable of having beliefs and desires, and hence of being rational. Davidson rests his case on four ingenious considerations: the intensional nature of thought; the idea that thoughts involve concepts; the holistic nature of thought; and the claim that belief requires the concept of belief. The last argument is unconvincing, but the first three shed valuable light on the extent to which thought requires language. We can only ascribe simple thoughts to animals and neonates, and even that ascription is incongruous, in that the rich idiom we employ has conceptual connections that go beyond the phenomena to which it is applied. The main impetus of this book is critical. One excuse is the tremendous scope and influence of Quine’s and Davidson’s accounts of language, thought and reality. If they suffer from the shortcomings I diagnose, this has far-reaching methodological and substantive implications. It rehabilitates the idea that there is a qualitative difference between science, which is concerned with empirical facts, and philosophy, which is concerned with conceptual issues and hence a priori. It also counts against those numerous and highly influential forms of naturalism which insist that human behaviour consists au fond of nothing but sounds and movements, or that all facts are (reducible to) physical facts. I do not purport to refute this kind of naturalism. But I hope to show that it finds no succour in Quine and Davidson. Rightly understood, some of their claims even militate against such a position. And though they subscribe to it in many respects, they do not provide sound arguments in its favour. In particular, the flight from intensions, the indeterminacy thesis included, presupposes rather than demonstrates that intensional discourse is obscure and non-factual.

11

12

Quine and Davidson on Language, Thought and Reality

Another excuse is that my project is not exclusively negative, since there are significant positive results to be gained. Although logical pragmatism is ultimately inadequate, there is much to be learnt from it. Not only does it contain important insights, even its errors are profound and important. Furthermore, the failure of the naturalisms of Quine and Davidson points to the need for a less austere position, one which does justice to the unique nature of human beings, human behaviour and language, without therefore embracing either Platonism or Cartesianism. There are forms of knowledge that cannot be reduced to physical science (contra epistemological naturalism), and the phenomena they describe – for example, numbers, concepts, thoughts, meaning – are no less real or factual than those of natural science (contra ontological naturalism). But the factuality and reality of these phenomena is not that of an abstract or mental realm that is just like the physical, only more ethereal. Rather, it depends on distinctively human capacities. These capacities do not require a supernatural explanation. They are perfectly intelligible features of animals of a special – social, linguistic, and cultural – kind; and their causal prerequisites and evolutionary emergence can be explained by science. Through a proper understanding of these capacities we may eventually be able to avoid naturalism, without lapsing into supernaturalism.

1

Logical pragmatism

This chapter provides a general introduction to Quine and Davidson. The first two sections are devoted to the historical context. I shall argue that the philosophical roots of their work are two-fold, namely formal logic and logical positivism on the one hand, American pragmatism and behaviourism on the other. It is not my ambition, however, to trace the details of actual historical influence. For my purposes it suffices to show that Quine and Davidson combine problems and methods bequeathed by logical positivism with important pragmatist themes, and that they can profitably be labelled as logical pragmatists. Sections 3 and 4 sketch the main features, respectively, of Quine’s and Davidson’s work, and highlight some important similarities and differences. The final section argues that their philosophies culminate in a philosophical anthropology, a conception of human behaviour in general, and of linguistic behaviour in particular. 1

The impact of logical positivism

Quine was born in 1908. He took his B.A. from Oberlin College and entered Harvard as a graduate in 1930. With the exception of several prolonged visits to other universities, he stayed there until his death in 2000. Davidson, born in 1917, was Quine’s pupil at Harvard; he later held posts at Stanford, Rockefeller, Chicago and Berkeley. Quine initially majored in mathematics. At Harvard, under the guidance of A. N. Whitehead, C. I. Lewis and Samuel Sheffer, he specialized in the function-theoretic logic invented by Frege and Russell. In 1933 he visited Vienna, Prague and Warsaw. This European tour had a lasting impact on him (TML 92–108). In Warsaw he became acquainted with the

[13]

14

Quine and Davidson on Language, Thought and Reality

Polish school of logic, and in particular with Alfred Tarski, whose work in formal semantics has exerted a tremendous influence, not least because of its impact on Quine and Davidson. In two books and numerous articles since 1933, Quine has made important contributions to the technical side of formal logic, notably by developing an alternative to the standard Zermelo-Fraenkel formulation of set-theory (see FLPV ch. V). Quine’s philosophical reputation rests mainly on the way in which he brings formal logic to bear on non-technical questions. Unlike many other analytic philosophers, he uses formal logic not just as a tool for the dissection of concepts and the paraphrase of propositions, but as the startingpoint for substantial doctrines, as is evident from the title of his first collection of philosophical essays: From a Logical Point of View (1953). In this he was influenced by Russell and by the logical positivists, whom he encountered in Vienna and Prague. The logical positivists were philosopher-scientists who aimed to develop a ‘consistent empiricism’. They agreed with British empiricism and Mach that all of human knowledge is based on experience, but tried to defend this idea in a more cogent way, with the help of modern logic. They employed logical rather than psychological analysis to identify the elements of experience. Moreover, they invoked Wittgenstein’s Tractatus to account for the necessity of logic and mathematics, without reducing it to empirical generality (Mill), lapsing into Platonism (Frege) or admitting synthetic a priori truths (Kant). All necessary propositions, they argued, are analytic, that is, true solely in virtue of the meanings of their constituent words. The logical positivists condemned metaphysics as meaningless, because it consists neither of a posteriori statements of fact – like empirical science – nor of analytic propositions that explicate the meaning of words – like logic and mathematics. Legitimate philosophy boils down to ‘the logic of science’ (Carnap 1937: 279). Its task is the linguistic analysis of those propositions which alone are strictly speaking meaningful, namely those of science. The ultimate aim is to vindicate empiricism by means of reductive analysis. The theoretical terms of science are defined through a more primitive observational vocabulary, and this makes it possible to translate all scientific theories into statements about what is given in experience. Quine first came to fame in 1951 through his article ‘Two Dogmas of Empiricism’, an off-spring of discussions with Carnap and Tarski in 1940 (see WO 67n; Creath 1990: 294–300). ‘Two Dogmas’ vigorously attacked the two pillars of the logical positivists’ conception of philosophy, namely the

Logical pragmatism

distinction between analytic and synthetic propositions and the project of reductive analysis. As a result, Quine is often seen as the most sophisticated and relentless critic of logical positivism. Nevertheless, he wrote of Carnap, the most eminent logical positivist: ‘I was very much his disciple for six years. In later years his views went on evolving, and so did mine, in divergent ways. But even where we disagreed, he was still setting the theme; the line of my thought was largely determined by problems that I felt his position presented’ (WP 41). Furthermore, logical positivism was not just a spring-board for Quine, he also agreed with many of its fundamental tenets (Hacker 1996b: 7–8): I. Like the logical positivists, Quine is an empiricist in both epistemology and semantics: sensory experience provides not just the evidence on which our beliefs rest, it also endows our language with its meaning: ‘whatever evidence there is for science is sensory evidence’, and ‘all inculcation of meaning of words must rest ultimately on sensory evidence’ (OR 75; see TT ch. 7). However, just as the logical positivists had tried to improve on Hume and Mach, Quine tried to improve on their version of empiricism. II. Quine does not accept the positivists’ principle of verification, according to which the meaning of a sentence is determined by the method of confirming or infirming it. However, this is only because of his epistemic holism, according to which confirming or infirming evidence can be specified not for individual statements, but only for larger ‘blocks of theory’. Subject to this holistic caveat, Quine accepts the verificationist doctrine that meaning is determined by empirical evidence (see chapter 6, section 1 below). III. Like the philosopher-scientists of the Vienna Circle, Quine espouses a form of scientism. Although scientism has wider cultural implications, it is in the first instance an epistemological thesis. It holds that science is the final arbiter of all knowledge claims. More specifically, the ‘hard’ sciences – the mathematical and natural sciences – not only yield our best explanation of physical reality; they comprise all of human knowledge. At the very least, they constitute the paradigm of human knowledge, and hence should be emulated by all other cognitive disciplines, including philosophy. IV. The Vienna Circle was committed to the ‘unity of science’, the idea that all scientific disciplines, including the social sciences, can be unified in a single system, the foundations of which are provided by physics. This ideal went hand in hand with reductionism, the idea that all cognitively significant propositions are analysable into an array of

15

16

Quine and Davidson on Language, Thought and Reality

basic propositions or ‘protocol-sentences’ about what is ‘given’ in experience. Quine repudiates reductionism, the idea that individual propositions are synonymous with and hence translatable into constructions from such basic propositions. But he holds on to the idea that physics is the master science: it gives us the fundamental description of reality to which all other sciences must approximate. He also retains the idea that simple empirical statements, he calls them ‘observation sentences’, provide the foundations of both knowledge and linguistic meaning (OR 87–8; PL 8; PT 4–5). Unlike the phenomenalists in the Circle, led by Schlick, he denies that these sentences are about sense-data; but in this he sides with the physicalists, led by Neurath. Quine’s observation sentences concern physical rather than mental phenomena; although, as we shall see, their content is given by neural stimulations rather than macroscopic physical objects and events. V. Quine shared the Circle’s anti-Platonist distaste for ‘abstract entities’ and the nominalist preference for austere ‘desert landscapes’ (FLPV 4). Although he came to accept the existence of certain abstract entities, his philosophy is shaped by a preference for nominalism. Abstract entities are admitted into one’s ontology only if they are absolutely indispensable for respectable science.

As a result of the rise of Nazism, most members of the Vienna Circle emigrated to the USA. By the forties, their views had achieved the status of orthodoxy. It is probably no more than mild hyperbole, therefore, when Davidson states that he got through graduate school by reading Feigl’s and Sellars’ anthology of positivist writings (EAE 261). Logical positivism provides the essential background to Davidson’s philosophical endeavours, but by way of opposition as much as by way of inspiration. While his style of philosophizing and the analytical tools he employs are influenced by logical positivism, he does not subscribe to any of the positivistic articles of faith mentioned above. Like Quine, Davidson follows the example of the logical positivists in his method of philosophical analysis. Whereas Wittgenstein and so-called ordinary language philosophy sought to clarify philosophically troublesome expressions by describing their ordinary use, Davidson relies heavily, though not exclusively, on analysing them in the idiom of formal logic. Nevertheless, his attitude towards formal logic is significantly different. Along with Frege, Russell, Tarski and the logical positivists, Quine is an exponent of what has come to be known as ideal language philosophy. This tradition holds that natural languages suffer

Logical pragmatism

from various logical defects (ambiguity, vagueness, referential failure, category-confusions), and that they must therefore be replaced by an ideal language – an interpreted logical calculus – at least for the purpose of philosophical and scientific inquiry. Davidson, by contrast, has been the most eminent champion of a theory of meaning for natural languages. The immediate roots of this project lie in the logical semantics of Tarski and Carnap. Under the influence of Tarski’s work on truth, Carnap came to conclude that the notion of meaning could be elucidated through the idea of truth-conditions (1956: 10). Unlike Tarski and Carnap, however, the languages Davidson is interested in are natural rather than artificial. In this respect, he stands in the tradition not of ideal language philosophy, but of the Tractatus (see Baker and Hacker 1984, ch. 1, 140–53; Smart 1986). Unlike its inventors, the early Wittgenstein regarded Frege’s and Russell’s new logic not as an ideal language which avoids the shortcomings of ordinary language, but as indicating the underlying logical form that sentences in the vernacular possessed all along. Just as the Tractatus maintains that the depth-structure of ordinary language is given by Russellian logic, Davidson maintains that it is given by a Tarskian truth-theory. Contrary to received wisdom, therefore, neither the early Wittgenstein nor Davidson are strictly speaking ideal language philosophers. In explicit opposition to Quine, Davidson aims to bring out the ‘metaphysics implicit in natural language’. He is interested not in ‘improving on natural language, but in understanding it’. Alluding to a simile of Wittgenstein, he describes ‘the language of science not as a substitute for our present language, but as a suburb of it’ (ITI 203; RQE 172, 176). Formal logic is philosophically important because it reveals the underlying structure of natural languages. As regards matters of content, Davidson has distanced himself not just from logical positivism but also from empiricism altogether. His formal semantics is not based on verificationism, nor does it rely on sensory evidence. He also rejects scientism, mainly because his philosophy of mind is inimical to the unity of science. The logical positivists fervently opposed the idea, held by the hermeneutic tradition, that there is a methodological difference between the natural sciences on the one hand, the psychological, social and historical sciences (Geisteswissenschaften) on the other. Moreover, the physicalists among them held that mental statements could be reduced to, that is, translated into, physical statements either about overt behaviour or about neurophysiological events. Both of these claims

17

18

Quine and Davidson on Language, Thought and Reality

are direct implications of the unity of science, and both are vigorously attacked in Davidson’s philosophy of mind (PAV 54). On the other hand, some doctrinal points of contact remain. While Davidson does not subscribe to nominalism, he follows logical positivism and Quine in his predilection for purely extensional languages like the first-order predicate calculus. Furthermore, Davidson’s philosophy of mind combines a conceptual dualism with a monistic physicalist ontology. Although we talk about mental events and human actions in terms that cannot be reduced to physics, these events and actions are ultimately physical events.

2

The legacy of American pragmatism

The impact on Quine and Davidson of American pragmatism is more diffuse than that of logical positivism. Pragmatism was founded by C. S. Peirce, popularized by William James, and further developed by John Dewey and G. H. Mead. The demise of German idealism in the middle of the nineteenth century sparked off various intellectual trends that tried to overcome religious and metaphysical mystery-mongering by stressing the importance of human practice. Pragmatism was the Anglo-Saxon version of this move from the Absolute to action. It differed from its continental cousins – Marxism, Existentialism, hermeneutics – in its empiricist and utilitarian tendencies, and in its association with natural science in general, and with Darwinism in particular. As regards Quine and Davidson, four pragmatist ideas are important: Anti-foundationalism: The Cartesian ideal of certainty is misguided. Our theories should aspire not to indubitable foundations, but to fallible conjectures with maximum explanatory power. Instrumentalism: Knowledge is not the passive mirroring of a mind-independent reality, but the result of an active process of inquiry. Our concepts and beliefs are instruments for the explanation and prediction of experience. Like human practice in general, the process of inquiry is essentially one of communication. Verificationism: The content of a concept or belief is determined by the experiential consequences we would expect our actions to have if the concept applied or the belief were true. For example, to claim of the liquid in a flask that it is an acid is to claim, inter alia, that the action of placing a blue litmus paper in the flask would have a certain result, namely that of turning the litmus paper red. By this

Logical pragmatism

token, the meaning of a word like ‘acid’ consists of the ‘conceivable experimental phenomena’ implied by its affirmation or denial (Peirce 1934: 273). Non-realist accounts of truth: According to Peirce, truth is ‘the opinion which is fated to be ultimately agreed to by all who investigate’ (1934: 268). According to James, a belief is true if it is expedient for us to believe it (1978: 106). Both accounts make truth partly dependent not on how things are, but on the beliefs and requirements of human beings.

Quine described his position as pragmatist in From a Logical Point of View (16–19, 20, 44–6, 79), but the term all but disappears in his subsequent writings. There is some reason to believe that this is no coincidence, but marks a shift from a radically pragmatist or instrumentalist to a more realist account of knowledge. Moreover, Quine later characterized pragmatism simply as a strand of empiricism, and states that he had not read widely in pragmatism before giving his Dewey Lectures in 1968 (PPE 23; CBG 292). On the other hand, Quine did not have to read widely in pragmatism to be influenced by ideas which are distinctly pragmatist rather than generally empiricist, since such ideas were transmitted by his teacher C. I. Lewis. Quine continues to stress even in his later writings that the formation of our beliefs and theories is shaped primarily not by brute facts or experience, but by ‘pragmatic’ considerations, that is, considerations of predictive power and cognitive efficacy. Some pragmatists insisted that our theories about the world should be subservient to the aims and requirements of our activities. Quine’s conception of knowledge is not utilitarian in this way (LAP 119). But it is instrumentalist. He holds that our theories ‘are almost completely a matter of human creativity – creativity to the purpose, however, of matching up with the neural input’ (PLSP 50–1; see TT 2; WO 17–20; PT 14–15). ‘Owing something to pragmatism is not one of my obsessions’, Davidson avers. At the same time, he recognizes a direct debt to the pragmatism of Quine and Lewis, and he subscribes to many ideas associated with pragmatism (PAV 49, see also 43). James (1978: 238), in particular, anticipated Quine’s and Davidson’s holistic view that our beliefs cannot be assessed individually, but only as part of a web of other beliefs. Furthermore, Quine advocates a version of naturalism, an idea which he traces to Dewey and Peirce (OR 26–9; PPE 35–7). This historical claim is problematic, since both the idea and the label go back at least to

19

20

Quine and Davidson on Language, Thought and Reality

nineteenth-century physiological naturalists like Czolbe. But it is justified in one important respect. From its very inception, pragmatism rejected the Cartesian idea that the findings of natural science require a more certain foundation, notably in our infallible knowledge of a private mental realm. This anti-foundationalism unites classical pragmatism not just with Quine and Davidson, but also with other contemporary proponents of pragmatist ideas such as Hilary Putnam, Susan Haack and Richard Rorty. Rorty (1990) has tried to associate Quine and even more so Davidson with other pragmatist ideas. One of them is what Rorty calls ‘antirepresentationalism’, the denial that our beliefs represent reality and that notions like reality or truth have an absolute sense independent of human purposes and interests. Quine is an instrumentalist about knowledge, since he stresses that our theories are tools for the purpose of organizing and predicting experiences. Nevertheless, it is problematic to align Quine or Davidson with pragmatism on the topic of truth. For they both reject pragmatist theories of truth (whether Peircean or Jamesian) in favour of Tarskian theories (see ITI xviii). Davidson has also rejected, albeit politely, Rorty’s suggestion (1986) that he shares with pragmatism not a theory of truth, but a deflationary attitude towards truth according to which there is no problem about truth because ‘true’ has no explanatory uses. Indeed, this suggestion does injustice to both sides. As Rorty himself acknowledges, James himself had a constructive (and highly controversial) theory of truth. And Davidson has explicitly rejected deflationary accounts of truth. It is precisely because neither the pragmatists nor Davidson are deflationists about truth, that Rorty’s claim of parallels between the two contains a kernel of truth. Davidson concedes that, like pragmatism, he rejects the correspondence theory of truth, along with its epistemological and metaphysical corollaries. Moreover, he rejects deflationism on grounds which he explicitly links to Dewey, namely that ‘access to truth could not be a special prerogative of philosophy, and that truth must have essential connections with human interests’ (SCT 279; see chapter 5, section 1 below). This agreement is part of a more fundamental consensus. Like pragmatism, the philosophies of Quine and Davidson revolve around a striking conception of human beings and of human behaviour. Quine’s view of human behaviour is directly rooted in behaviourism. His theory of language-learning is based on the work of his Harvard colleague

Logical pragmatism

B. F. Skinner, while his general behaviourist outlook goes back to Watson (TML 110). But a strong connection with pragmatism remains. Dewey’s and Mead’s attack on introspectionist psychology was one source of behaviourism, and their theory of language was in turn influenced by Watson, although their position was far less crude. Moreover, all pragmatists share with our protagonists a third-person perspective on mind and meaning. They also share with Davidson an emphasis on the rational aspects of human behaviour. Davidson’s third-person perspective and his emphasis on rationality are a direct result of his work in decision theory during the fifties. But as he himself recognizes, the idea that rationality and with it language are essential to human action aligns his work with pragmatism, notably on the question of animal minds. Unlike Quine, whose epistemology starts out from the experiences of individuals, Davidson also follows pragmatism by regarding knowledge as an essentially social phenomenon arising out of linguistic communication. A final and equally general point of contact has been highlighted by Rorty (1986: 333, 339). Like Quine and Davidson, the American pragmatists are famous for their ‘debunking of dualisms’. Because of their naturalism and their behaviourism, they rejected the Cartesian dualism of mind and matter. Because of their Hegelian heritage and their general dislike for academic compartmentalizations, they revolted against Kantian dichotomies, especially the differentiation of concepts and intuition, of theoretical and practical reason, and of philosophy and empirical science. In the first respect they have been followed by Quine’s and Davidson’s antidualistic conceptions of the mind. In the second respect they have been followed and superseded by Quine’s attacks on the positivists’ distinction between analytic and synthetic statements and by Davidson’s rejection of the Kantian distinction between conceptual scheme and empirical content. Both attacks are holistic, in that they regard philosophical and scientific statements as inseparable parts of a single web of beliefs. Some interpreters have suggested that Quine should not be seen as part of the pragmatist tradition (Hookway 1988: 1–3; Koppelberg 1987: 313–14). One objection is that Quine picked up holism not from the pragmatists, but from Neurath. However, Quine himself states that his position owed no more to reading Neurath than it did to reading the pragmatists (POQ 212). In any event, there were acknowledged pragmatist tendencies in logical positivism over and above holism (see Carnap 1963: 860–2). Quine himself notes the proximity of Peirce’s theory of meaning to the logical positivists’ verificationism (PPE 30). Furthermore, both Carnap and Neurath

21

22

Quine and Davidson on Language, Thought and Reality

stressed the importance of pragmatic considerations in the construction of scientific theories and logical systems, appealed to behaviourism in order to incorporate psychology into unified science, and invoked the pragmatists to stress the social nature of language. Davidson’s work has been viewed not so much as a form of pragmatism than as a new kind of hermeneutics (Ramberg 1989; Malpas 1992) or a novel rationalist metaphysics (Evnine 1991). However, these characterizations are not incompatible. Pragmatism shares many features with hermeneutics, notably its stress on communication and its attempt to steer a middle ground between dualist and materialist conceptions of the mind. It also shares features with rationalism, such as the stress on the importance of reason and a rejection of the empiricist myth of the given, the idea that conceptually unmediated sensory stimulations provide the foundations of knowledge and meaning. The way in which Davidson uses the ideas of Quine, the arch-empiricist, to undermine certain tenets of empiricism is an impressive execution of James’ admirable ambition (1978: 13–24, 127– 9): to defend the values of the tender-hearted rationalists by the methods of the tough-minded empiricists. This is not to deny that our protagonists diverge from pragmatism on important issues. For one thing, their philosophy is not infused with moral and political ambitions. Indeed, Quine has explicitly denied that such aspirations have a place in philosophy proper (TT ch. 23; RHS 493). For another, Quine’s outlook is more austere than that of the pragmatists, and his allegiance to empiricism far more pronounced. Most importantly, Quine and Davidson did not pick up ready-made pragmatist ideas, since they had to develop those ideas independently out of a critique of logical positivism, and especially of Carnap. These points do not alter the fact that Quine and Davidson are inspired by pragmatist themes and attitudes, whatever their actual historical origin. It is equally wrong, however, to portray their work as a pragmatist attack on logical positivism, particularly in Quine’s case (pace Murphy 1990: ch. 7). Their philosophical styles, their problems and their methods owe much more to analytic philosophy in general and logical positivism in particular than to pragmatism. Both roots are important. Quine and Davidson can be characterized as logical pragmatists, because their relation to American pragmatism is in important ways analogous to that of the logical empiricists to classical empiricism. They develop some, but not all, fundamental ideas of pragmatism; and they do so in a clearer and more cogent way, thanks to their magisterial use of logico-linguistic analysis

Logical pragmatism

and to their greater sophistication in semantics. Their philosophy of language combines the formal approach developed by Frege, Russell, the early Wittgenstein and the logical positivists, with the pragmatist idea that language is a form of human behaviour, an idea that is also important to the later Wittgenstein. As a result, Quine’s and Davidson’s philosophy of language is often more sophisticated than that of the original pragmatists. With the exception of Peirce, the pragmatists lacked the dialectical acuity and semantic know-how of the analytic tradition. To take one important example, Quine’s and Davidson’s attack on Kantian dualisms (analytic/synthetic, philosophy/science, scheme/content) is driven not by the Hegelian sentiments of the pragmatists, but by powerful and elaborate arguments in philosophical logic. At the same time, however, this formal approach can also be a source of errors that require correction by a more pragmatist perspective. The regimented languages of formal logic are neither superior to the dynamic and multifaceted practice that constitutes a natural language, nor do they reveal the hidden structure of the latter (see chapter 5, section 4 and chapter 8, section 3). 3

Quine’ s naturalism

Quine marks a decisive watershed in the development of analytic philosophy, because he challenged its very conception of philosophy. From Plato onwards, it has often been maintained that philosophy, like logic and mathematics, is a priori, independent of sensory experience. Its problems cannot be solved, its propositions cannot be supported or refuted, by either everyday observation or scientific experiments. This idea has been opposed by radical empiricists like Mill and, on occasion, Russell, who maintained that even the purportedly a priori disciplines – mathematics, logic and philosophy – are ultimately based on experience. The main attraction of this prima facie implausible insistence lies in the fact that it is very hard to provide a satisfactory explanation of the special status of these disciplines. Their propositions are a priori according to Platonists because they are about abstract entities beyond space and time, according to Aristotelians because they describe the most general features of reality, and according to Kantians because they express non-empirical preconditions of experience. None of these explanations is compelling or even reasonably clear. By contrast, the logical positivists’ linguistic doctrine of necessity promised to provide a distinctive role for philosophy, without dubious appeals

23

24

Quine and Davidson on Language, Thought and Reality

to a Platonic realm of abstract entities, Aristotelian essences or Kantian pure reason. While science results in empirical propositions that describe reality – and are hence synthetic – philosophy is concerned with analytic propositions which unfold the meaning of the terms employed by science and/or common sense. This linguistic turn was ostensibly directed against Kant’s suggestion that philosophical propositions are synthetic a priori. Nevertheless, it is a linguistic transformation of a Kantian idea. Unlike science or common sense, philosophy does not itself describe objects of any kind, not even the abstract entities or essences postulated by Platonism and Aristotelianism. Instead, it is a second-order discipline which reflects on the conceptual scheme that science and common sense employ in their empirical descriptions and explanations of empirical reality. This kind of line has been taken not just by logical positivists such as Schlick and Carnap, but also by other exponents of the linguistic turn, such as ordinary language philosophy and Wittgenstein. In spite of their considerable differences, these philosophers tended to accept that there is a qualitative difference between science, which is concerned with factual issues and hence a posteriori, and philosophy or logic, which is concerned with conceptual issues, and hence a priori. For the philosophy of mind, for example, this means that one must distinguish between the empirical discoveries of psychologists or neurophysiologists, which concern phenomena with which mental phenomena are contingently related, such as the firing of neurons, and the philosophical analysis of mental concepts, which specifies necessary features of mental phenomena. Quine has completely overturned this picture by vigorously denying that there is a significant difference between philosophy and science. Many who claim that philosophical problems can simply be solved by science, notably neurophilosophers and members of the artificial intelligentsia, simply appeal to his authority. In this vein, Paul Churchland (1986: 2–3) commends Quine for overcoming a priori ‘armchair’ philosophy by demonstrating that ‘philosophy at its best and properly conceived is continuous with the empirical sciences’. Ironically, in practice Quine is himself of the armchair variety. He did not claim that philosophy is empirical because he had discovered scientific evidence which solves philosophical problems. Instead, his line of argument was purely a priori. He maintained that from a logical point of view there is no qualitative difference between empirical propositions and the allegedly necessary propositions of logic, mathematics and philosophy. And in support of this claim he employed thought-experiments – notably in his argument for the

Logical pragmatism

indeterminacy of translation – which are totally removed from empirical evidence. In fact, his arguments are much closer to Descartes’ hyperbolical doubts than to scientific research. In his writings on ontology, Quine argued that philosophical questions concerning the existence of, for example, numbers, are no different in principle from scientific questions concerning the existence of, for example, neutrinos. More importantly still, Quine rejected the analytic/ synthetic distinction as an unfounded ‘dogma of empiricism’. He thereby challenged the idea that there is a distinct type of proposition which articulates conceptual connections rather than empirical facts, and reinvigorated the position of radical empiricism, according to which philosophy is continuous with the sciences and ultimately based on experience. As a result of his work, this conception of philosophy as science has achieved the status of orthodoxy, especially in the USA. According to Quine, proper or ‘scientific philosophy’ does not just emulate the methods of the deductive-nomological sciences; it is itself ‘continuous with science’, and in fact part of science. Quine wants to ‘rub out or at least blur the distinction between philosophy and various sciences’ (PPLT 2; PLSP 57, 47, 51). But while he could not be more explicit about this basic point, he provides diverse (though not necessarily incompatible) accounts of the role philosophy is to play within science. In some places he follows Locke’s famous image of philosophy as an underlabourer: philosophy is a ‘handmaiden to science’ with the task of ‘tying up loose ends’ such as paradoxes and questions of evidence, problems that working scientists tend to ignore. In others he is closer to the more flattering Aristotelian image of philosophy as the queen of the sciences. It deals with the ‘general, basic concepts of science’ such as truth, existence and necessity (PLSP 57, 47–8). In more typical passages, he follows Russell and expresses the same image by reference to reality rather than concepts. Philosophy is concerned with ‘a limning of the most general traits of reality’. It investigates the fundamental ‘furniture of our universe’, and differs from science only quantitatively, in the generality and breadth of its questions and categories (WO 161, 254, 228–9, 275–6). Quine’s work is not just of the utmost importance to the selfunderstanding of philosophy, it also raises a host of novel substantive issues in philosophical logic, epistemology, and metaphysics. Like other philosophical revolutionaries, he has shown how the nature of philosophy is linked to other topics, such as logical necessity, linguistic meaning, synonymy, knowledge, scientific method and ontology. Thus Quine’s attack

25

26

Quine and Davidson on Language, Thought and Reality

on the analytic/synthetic distinction involved two lines of reasoning – one concerning epistemology and scientific method, the other concerning semantics and ontology. The impetus of the first line is that the analytic/synthetic distinction presupposes a second dogma of empiricism, namely ‘reductionism’, the view that every meaningful statement is translatable into a statement about the immediate experiences that confirm it. Reductionism would allow one to define analytic statements as those which are confirmed come what experience may (FLPV 38, 41). However, Quine argues, it is at odds with the holistic nature of scientific beliefformation, the fact that our beliefs form a ‘web’ in which each belief is linked to all others, and ultimately to experience. This means that it is impossible to specify confirming evidence for individual statements. It also means that any belief can be abandoned for the sake of preserving other parts of the web, and hence that there are no a priori statements immune to empirical revision. Quine’s second line of reasoning evolves around the distinction between intension and extension. The distinction is Carnap’s (1956), but it goes back to Frege’s distinction between sense and meaning, and it is central to both Quine and Davidson. The notion of extension is a generalization of the notion of reference. The extension of an expression is what the expression stands for or what it applies to. The extension of a singular term is the object it refers to; the extension of a predicate is the set of objects of which it is true; the extension of a declarative sentence is its truth-value (its truth or falsity). By contrast, the intension of an expression is an aspect of what the expression means, standardly that aspect which determines its extension. The intension of a singular term could be a description which singles out its referent, the intension of a predicate the attribute or property which the objects falling under it must possess, the intension of a declarative sentence the proposition or thought it expresses. Linguistic contexts or whole languages are extensional or transparent if and only if (henceforth ‘iff ’) substitution of a part with the same extension does not change the extension of the whole. In an extensional context one can substitute co-referential terms salva veritate, that is, without altering the truth-value of the sentence. By contrast, intensional contexts or languages are ‘referentially opaque’ (FLPV ch. VIII; WP ch. 17; WO §30). Substitution of co-extensional sentence parts (singular terms, predicates) need not preserve the extension of the whole sentence. Furthermore, intensional contexts are not truth-functional: the truth-value of complex propositions is not simply a function of the truth-value of the component

Logical pragmatism

propositions. Such intensional contexts are created, inter alia, by modal notions like ‘necessarily’, by so-called propositional attitudes like belief and desire, and by quotation. For example, even though ‘Tully’ and ‘Cicero’ refer to the same person, we cannot necessarily substitute the one for the other salva veritate in ‘Susan believes that Cicero was Roman’. Quine’s semantic argument against the analytic/synthetic distinction is that analyticity is part of a circle of intensional notions that cannot be reduced to purely extensional notions like truth or reference. But, he argues, all these notions are obscure, because there are no criteria of identity for ‘intensions’: while we can establish whether two expressions have the same extension, we cannot establish whether they are synonymous, that is, have the same intension or meaning. Quine subsequently defended this contention by focusing on ‘radical translation’, the translation of a completely foreign language from scratch. Because such translation cannot assume any prior understanding, it helps us to appreciate that translation is ‘indeterminate’: there are no objective standards for whether two linguistic expressions are synonymous, and hence no criteria of identity for intensions. As a result, scientific philosophy should not accept intensions into its ontology. Indeed, as regards the components of sentences, even reference is ‘inscrutable’. It is impossible to tell what precisely the components of sentences refer to. As a result, in describing language scientific philosophy should stick to a behaviourist ersatz of meaning; ‘stimulus meaning’ is an extensional feature, and one possessed by whole sentences only, namely that of being assented to under certain stimulatory conditions. What unites Quine’s metaphilosophical views with his approach to particular philosophical topics is his naturalism. All naturalists are hostile to explanations that invoke phenomena beyond nature, such as God, abstract entities in a Platonic realm beyond space and time, or Cartesian soul-substances. But beyond this consensus, naturalism comes in various shapes and sizes. For our purposes, I want to distinguish three types of naturalism.

r Metaphilosophical naturalism claims that philosophy is a branch of or continuous with natural science;

r Epistemological naturalism is nothing other than scientism as defined above: it insists that there is no genuine knowledge outside natural science; r Ontological naturalism denies that there is any realm other than the natural world of matter, energy, and spatio-temporal objects or events.

27

28

Quine and Davidson on Language, Thought and Reality

Quine comes to naturalism from a metaphilosophical angle, but this metaphilosophical naturalism is intricately linked to epistemological and ontological versions. He defines naturalism as the abandonment of ‘first philosophy’, the foundationalist project of providing science with a philosophical underpinning which is firmer than science itself (TT 72, see 21; N 257). According to Quine, there is no ‘cosmic exile’, no external Archimedean point, such as the one sought by Descartes, from which to compare our conceptual scheme or belief-system with reality (WO 275–6; OR 84–7, 126). However, Quine’s naturalism is directed not just against the idea of philosophy as a super-science that provides the foundations of science, but also against any attempt to treat philosophy as a sui generis discipline with aims distinct from those of science. Quine’s metaphilosophical naturalism is an immediate consequence of what he calls ‘methodological monism’, the rejection of the analytic/ synthetic distinction (TT 71–2). If there is no logical difference between analytic and synthetic propositions, then there is no qualitative difference of any kind between the propositions of philosophy and those of science. This metaphilosophical naturalism is reinforced by Quine’s epistemological naturalism. Philosophy is part of science because otherwise it could not be a cognitive discipline aspiring to knowledge. In earlier writings, Quine disparaged forms of discourse other than science as ‘second-rate’ (OR 24). More recently, he acknowledged that there are ‘perfectly good language-games’ other than science, e.g. fiction and poetry (PT 20). Yet he remained wedded to the idea that there is no knowledge outside science, that natural science is the only route to truth, and provides the only genuine explanation of the world. Quine is fond of Neurath’s famous ship-metaphor. ‘We are like sailors, who have to rebuild their ship on the open sea, without ever being able to put it into dry-dock and to build it afresh from best components’ (WO motto; see Neurath 1944: 47). Philosophers cannot compare our beliefsystem as a whole with reality in order to establish its truth or falsity, as foundationalism had it. Nor can they engage in a second-order reflection on our conceptual framework, as Kant and proponents of the linguistic turn had it. Science provides our overall view of the world; and like science, philosophy seeks to describe and explain reality. As a result, philosophers and scientists ‘are in the same boat’ – empirical science – ‘which stays afloat because we keep the bulk of it intact as a going concern’. Our evolving scientific doctrine, however fallible, is the ‘final arbiter’ of truth (WO 3–4, 23).

Logical pragmatism

Quine recognizes that this epistemological naturalism requires a new conception of epistemology itself. He urges us to replace traditional epistemology by ‘naturalized epistemology’. This novel discipline continues to investigate the subject of traditional epistemology, namely the relationship between theory (our beliefs) and empirical evidence; but it does so through empirical science (neurophysiology, behaviourist psychology), rather than through a priori reflection (OR ch. 3; TT ch. 2; see Stroud 1984: ch. 6; Orenstein 2002: ch. 8). Quine’s metaphilosophical and epistemological naturalism is also bound up with ontological naturalism. The task of philosophy/science is to explain reality. As part of this assignment, naturalized epistemology also explains our beliefs about reality. These explanations are not to make reference to any super-natural or immaterial phenomena. Indeed, because of Quine’s adherence to the unity of science, his ontology is not just materialist but physicalist (see WO 1–4, 264–6; IPO 108; FM 162; TT 21; RHS 430). Strictly speaking, the only things that exist are those which feature in the explanations of the most fundamental science, namely physics. At the same time, Quine’s naturalism is not reductionist in the semantic sense of the term, which applies to logical atomism and the Vienna Circle. He does not claim that ordinary statements which are prima facie about non-physical phenomena (e.g. numbers or mental states) are semantically equivalent to some logical construction using terms referring exclusively to microphysical events. Indeed, he rejects this kind of reductionism as a ‘dogma of empiricism’ (FLPV 20). Instead, Quine’s ‘canonical notation’, the formal language he recommends for the purposes of scientific philosophy, incorporates a form of eliminative naturalism. Statements which involve concepts that Quine repudiates as incompatible with naturalism are not analysed into statements of the kind he accepts; they are replaced by such statements. Quine follows Carnap (1956: 7–8) in holding that the proper method of logical analysis is ‘explication’. The objective of an explication is not to provide a synonym of the analysandum, but to furnish an alternative expression or construction which serves the cognitive purposes of the original equally well, while avoiding its scientific or philosophical drawbacks such as obscurity, philosophical puzzlement and unwanted ontological commitments (FLPV 25, 106; WO 224, §§33, 53–4; WP 151; TT 87). But while the logical positivists designed their ideal languages to put an end to metaphysics (by allowing only the formulation of scientific questions and propositions), Quine’s canonical notation, by serving science,

29

30

Quine and Davidson on Language, Thought and Reality

also serves a metaphysical purpose, namely that of limning the most pervasive and fundamental traits of reality. In Strawson’s illuminating phrase (1959: 9–10), Quine is the leading contemporary proponent of ‘revisionary metaphysics’. Unlike descriptive metaphysicians (e.g., Aristotle, Kant, Strawson), his ambition is not to elucidate the way we in fact think or speak, our actual conceptual scheme, but to replace that conceptual scheme by a new way of thinking about the world. In Quine’s case, this new conceptual scheme is dictated by modern science; his naturalism is designed to be the metaphysics of science. Quine’s aim is elimination rather than analysis. Yet Quine’s naturalism has consequences similar to those of reductionism, in spite of his indignant reactions to that label (RHS 364). It rejects as illegitimate or inferior all cognitive claims couched in any idiom other than that of natural science, and thereby discards or marginalizes many forms of thought, notably those which use intensional or intentional concepts. The ultimate explication of any legitimate form of discourse is in terms of physics, or of a science which is as close to physics as possible, for example behavioural psychology. Physics gives us the fundamental description of reality, and all deep explanations of phenomena are physical explanations, for the fundamental laws of the universe are physical laws. Explanations in less fundamental sciences, though not reducible to physics, are at best local generalizations supervenient upon physical law. Quine’s ontological austerity and his scientistic metaphilosophy have inspired most contemporary forms of naturalism. Epistemologists, semanticists, philosophers of mind and cognitive scientists alike have scrambled to lay claim to his mantle. Together with Feyerabend, Quine is the godfather of eliminative materialism, the view that our ordinary psychological statements and concepts, notably the idiom of propositional attitudes, should be replaced by a more scientific, neurophysiological jargon. At a more general level, it is mainly because of Quine that few analytic philosophers these days would dare to publish a book in the philosophy of mind without at least professing their allegiance to some form of naturalism in the preface. 4

Davidson on reason’ s place in nature

It is just conceivable, however, that Davidson might be a heretical exception to this rule. His relation to Quine is a matter of considerable interest and of some dispute. Davidson himself claims to have abandoned

Logical pragmatism

empiricism even in Quine’s ‘undogmatic’ version (ITI 189; MTE 68). Many commentators confirm that in this respect at least he has broken radically with his mentor (Evnine 1991: 4–6). In my own view, which I hope to substantiate in this book, this is only part of the truth, albeit an important part. While Davidson modifies Quine’s position in substantial and original ways, he does so without compromising the naturalistic framework or abandoning empiricism altogether. In any event, their philosophies of language are best seen in conjunction. Quine provides the acknowledged starting-point for Davidson. Inquiries into Truth and Interpretation is dedicated ‘To W. V. Quine without whom not’. Davidson presents his work as a continuation of Quine’s, and Quine views matters in much the same way (RQE 172; CTK 313; CBG 80). Like Quine, Davidson combines logical analysis and formal semantics `a la Tarski and Carnap with a pragmatist emphasis on language as a form of human behaviour. More specifically, Davidson fully endorses Quine’s third-person perspective on meaning and communication: ‘Perhaps the most important thing [Quine] taught me is that there could be no more to the communicative content of words than is conveyed by verbal behavior’ (RH 80; see chapter 7, section 2 below). In addition to this methodological agreement, Davidson accepts many of Quine’s substantial claims: that truth is a property of token-sentences, that there is no significant difference between analytic and synthetic propositions, that translation is indeterminate, and that reference is inscrutable. Furthermore, he has developed other Quinean themes – notably radical translation and the attack on the idea of linguistic conventions – in powerful ways, which often make them more palatable, at least to those with reservations about Quine’s austere naturalism. For it is equally undeniable that Davidson’s philosophical outlook is less harsh than Quine’s. Although he accepts that there is no clear line between philosophy and science on account of rejecting the analytic/ synthetic distinction, he does not advance an elaborate metaphilosophical doctrine, let alone a scientistic one. If anything, he seems to regard the blurring of this line as a licence for philosophers to put in perspective the relevance of neurophysiological findings to problems in the philosophy of mind (see PAV 44; EAE 216). On occasion, Davidson calls himself a naturalized epistemologist. But what he underwrites is only a ‘resolutely third person approach to epistemology’ (SIO 159, 194; see Gibson 1994). As the example of Peirce and Wittgenstein shows, such an approach is not tied to Quine’s transformation of epistemology into physiological psychology.

31

32

Quine and Davidson on Language, Thought and Reality

In fact, Davidson not only argues that such a move is actually inimical to a third-person perspective, he also echoes Kant and Frege in complaining that it runs together the causal explanation of our beliefs with their epistemic justification (see chapter 6, section 3). This contrast is part and parcel of Davidson’s general move away from some of Quine’s empiricist and behaviourist commitments. He develops the heuristic device of radical translation and accepts a version of the indeterminacy of translation and of the inscrutability of reference. Yet he does not opt for Quine’s nihilism about intensional semantic notions and their replacement by a behaviourist ersatz. Davidson’s hope is rather that a purely extensional semantics can deliver or at least approximate intensional notions like meaning, provided that it furnishes an adequate explanation of linguistic behaviour. Davidson uses Tarski’s semantic theory of truth to construct a systematic theory of meaning for natural languages. Such a theory must be ‘compositional’. It displays the meaning of every sentence as a function of the meanings of its components and of their arrangement, and thereby explains how a potentially infinite number of meaningful sentences can be constructed from a finite vocabulary. Davidson’s crucial step is to maintain that Tarski’s theory fits the requirements on such a compositional theory of meaning. Tarski defines the concept of truth for a formal language L through a recursive axiomatic system. The axioms lay down how the primitive signs (words) of L are to be interpreted. Together with the rules of inference, these axioms permit for each sentence s of the language the derivation of a so-called T-sentence, a sentence of the form ‘s is true iff p’, where s is the name of a sentence of L and p the translation of that sentence into the language of the theory. Whereas Tarski relied on the notion of translation to define truth, Davidson assumes an understanding of the concept of truth; he can therefore use Tarski’s theory as a Fregean theory of meaning. T-sentences state the meaning of the sentences of L by specifying the conditions under which they are true. Unlike Tarski, Davidson hopes that such a theory can be constructed not just for formal but also for natural languages. He uses a variant of radical translation, what he calls ‘radical interpretation’, to show that a Tarskian theory for a natural language is capable of empirical confirmation. It is at this point that Davidson remains within the orbit of Quine’s empiricism and naturalism. Like Quine, he suggests that the situation of radical translation/interpretation is characteristic of linguistic understanding in general, even within a linguistic community. Furthermore, he

Logical pragmatism

regards meaning as a problematic phenomenon that has to be built up or extracted from something more basic. As a result, he characterizes linguistic communication not as a form of interaction based on shared practices, but as a process of constructing explanatory theories on the basis of nonsemantic, and, more specifically non-intensional evidence. Again in line with Quine, Davidson maintains that the basic evidence for the theory construction involved in radical interpretation concerns what sentences speakers of a language would assent to or ‘hold true’ under what circumstances. By contrast to Quine’s ‘proximal theory’, however, his ‘distal theory’ describes the conditions of assent not in microscopic terms of neural stimulations, but in terms of macroscopic objects or events. Furthermore, since Davidson does not share Quine’s behaviourism, he treats assent as a form of intentional behaviour rather than a mechanical response to a stimulus. Finally, he treats the translation of linguistic utterances as an integral part of a unified theory of meaning, mind and action. We can only decide what speakers mean by their utterances if we can also make sense of their actions and mental states. Such a unified theory of interpretation presupposes certain normative principles, which are alien to Quine’s naturalism. Quine’s radical translation invokes a ‘principle of charity’ according to which we should not interpret utterances of the alien language as being obviously false. Davidson turns this heuristic maxim into a principle which is constitutive of the very ideas of language and action. Unless we can treat the natives as being by-andlarge rational, we cannot regard their utterances as meaningful or their behaviour as intentional. This is not the only respect in which Davidson’s treatment of mind and action is remote from Quine. His famous doctrine of anomalous monism (EAE ch. 11) undermines both the reductive materialism of the logical positivists and the eliminative materialism inspired by Quine. Anomalous monism is ‘ontological monism coupled with conceptual dualism’. It tries to reconcile the naturalistic (anti-Platonist and anti-Cartesian) claim that there is no realm beyond the physical with a recognition that mental and semantic discourse is neither reducible to nor replaceable by the idiom of natural science. ‘There are no such things as minds, but people have mental properties . . . These properties are constantly changing, and such changes are mental events’ (DD 231; see RH 599). The key to Davidson’s position is his ontology of events. But this ontology is itself a fallout from his philosophy of language. From his discussion of radical interpretation, Davidson concludes that both scepticism and

33

34

Quine and Davidson on Language, Thought and Reality

conceptual relativism are wrong and that our ordinary discourse must embody a fundamentally correct view of the world (SIO chs. 10–11; cf. Feldman 1998 and chapter 6, section 4 below). On this assumption, a theory of meaning can offer substantial answers to questions about reality, namely by establishing the logical form of natural languages. In particular, the existence of events can be demonstrated by showing that we can make sense of the inferential patterns of ordinary discourse only by assuming that some of its sentences quantify over events (EAE ch. 7). Events are dated occurrences with definite durations, like the Olympic Games of 1992 or Brutus’ stabbing of Caesar. They can be described in fundamentally different ways, namely through both mental and physical concepts. Some events have true descriptions in psychological terms and are hence mental. But all events, including the mental ones, also have true physical descriptions and are hence physical. Thus my coming to believe that p has a mental description, e.g. as based on my belief that q; but it also has a physical description. The two descriptions pick out the same event; indeed, every particular mental event is identical with a particular physical event. But although there is token-token identity between mental and physical events, there can be no type-type identity: there are no psychophysical laws that correlate mental events under their mental description (‘coming to believe that p’) with physical events under their physical description (‘φ-fibres firing’). Davidson’s argument in favour of this assumption rests on two ideas. First, the principal purpose of mental discourse is to explain behaviour by reference to propositional attitudes such as believing that p, desiring x and intending to . Secondly, this kind of explanation is characterized by two features which are constitutive of mental discourse but absent from physical discourse – normativity and holism (EAE chs. 11–12; PEA; DD). Mental discourse is normative. Mental states are subject to those principles of rationality which a radical interpreter guided by the principle of charity imputes to the speakers she interprets. These principles (explored in logic and decision theory) specify what it is reasonable to believe or desire. One such principle is: (P) If you prefer x to y and y to z, then you should prefer x to z.

Mental discourse is holistic in that propositional attitudes do not occur in isolation. One cannot attribute a propositional attitude to a person independently of attributing other propositional attitudes to her. This is

Logical pragmatism

not just an epistemological constraint, but due to the nature of propositional attitudes. The logical connections between propositional attitudes are partly constitutive of their identity: what makes the belief that p the belief it is, is at least in part its interconnection with other beliefs. Now suppose that there are psychophysical laws which connect preferring x to y and y to z with neural state m and preferring x to z with neural state n. This law should enable us to infer from (P) that if someone is in neural state m he should also be in neural state n. But the only kind of ‘should’ which has a place in physics is one of behaving in line with expectations supported by empirical evidence, not a normative injunction like (P). Equally, this psychophysical law, in conjunction with a physical law according to which anyone in neural state m is also in neural state n, should allow us to predict that whenever someone prefers x to y and y to z he prefers x to z. But because of the holistic nature of thought, the attribution of that last preference may run counter to normative constraints. For example, if we have independent reasons to ascribe a preference of z to x, it lumbers the subject with contradictory preferences. Accordingly, physical and mental predicates cannot feature in psychophysical laws because of ‘the disparate commitments of the mental and the physical schemes’ (EAE 222). Davidson combines the hermeneutic idea that actions are to be explained by reference to the reasons for which agents perform them (unlike merely physical events), with a causal theory, according to which actions are caused by those reasons (EAE ch. 1). The key to this combination is the idea that both actions and the reasons that explain them are among the events which have both a physical and a mental description. For example, if I raise my arm, this can be described either in biophysical terms (as a movement of muscles and bones), or as my signalling to you that I want you to be quiet. The reason for my action is a combination of desires or ‘pro-attitudes’ and beliefs, e.g. that I form the desire to silence you and come to believe that signalling will silence you. According to Davidson, each action has a physical description under which it can be seen to be caused by its reason, described in equally physical terms. At the same time, there is no causal law which relates action and reason as described in mental rather than physical terms. For what determines which mental descriptions are true of these events is subject to holistic considerations of rationality, considerations that are qualitatively different from the considerations which determine what physical descriptions are true of them. Just as there are no psychophysical laws relating

35

36

Quine and Davidson on Language, Thought and Reality

φ-fibres firing and wanting to silence you or arm-jerkings and signallings to be quiet, there are no strict psychological laws relating certain belief/desire combinations with certain actions. In both respects, therefore, our psychological discourse is sui generis. At the same time, what it describes – actions, beliefs and desires – is part of the causal physical order. ‘The mental is not an ontological but a conceptual category’ (PEA 46). This conceptual dualism includes language. Linguistic utterances are actions, and hence physical events. But they can be described as meaningful on the presumption of the same kind of rational (normative and holistic) constraints that guide the interpretation of other actions. Just like the mental, the semantic is a conceptual rather than ontological category. For Davidson, human beings, their actions and their utterances are physical phenomena; yet they cannot be properly understood by physics, but only by reference to reasons. He shares Quine’s predilection for a monistic ontology, but not for a scientistic methodology. In this way, he seeks to pay the ontological homage that philosophy owes to modern science, without denying the distinctive character of human beings. 5

A philosophical anthropology

Unlike many other analytic philosophers, Quine is a system-builder in the vein of Descartes, Kant or Hegel. His reflections on various philosophical topics (philosophy, necessity, language, knowledge and the mind) are part of a systematic and comprehensive theory. The same goes for Davidson, in spite of the fact that he has developed his system through a series of overlapping and criss-crossing articles rather than in a single book. The most fascinating feature of their oeuvres is how they link problems and concepts from various fields – metaphilosophy, semantics, epistemology, philosophy of science, philosophy of mind – and weave them into a striking whole. It can also be an exasperating feature, especially for critics. Claims from one area which are prima facie implausible are supported by ideas from another, and thus form a single, powerful and coherent perspective. Ian Hacking (1975: ch. 1) distinguishes between ‘pure’ philosophy of language, which discusses issues of meaning exclusively for their own sake, and ‘applied’ philosophy of language, which discusses them in order to understand philosophical problems in other areas. Quine and Davidson are not just applied philosophers of language, they are positively impure. Although they tackle with relish the detailed technical problems beloved by their post-positivist colleagues, they connect these problems ab initio

Logical pragmatism

to wider issues. Thus Quine has turned the apparently recherch ´e question of whether there are objective standards of translation into the linchpin of a whole new conception of philosophy. In a similar style, Davidson argues that the ‘charitable’ presuppositions of radical interpretation rule out seemingly irrefutable positions like scepticism and relativism. In some respects this rapid move from small-scale to large-scale questions is a disadvantage, because it departs from the controlled, piecemeal approach of both logical positivism and ordinary language philosophy. Consider Quine’s discussion of sentence meaning and propositions (PL ch. 1). Instead of breaking the problem down into smaller and more manageable parts, within a few pages he becomes embroiled in complex issues concerning belief-formation and scientific method, with the striking result that there is no such thing as sentence meaning and that we believe sentences rather than what they express. One may well suspect that a more circumspect approach might have led to more reliable if less iconoclastic results. On the other hand, the impure and encyclopaedic nature of Quine’s and Davidson’s approach makes it richer and more dynamic than that of most other post-positivist philosophers, roughly in the way Wittgenstein’s work is both less pure and more exciting than that of ordinary language philosophy. The immediate context in which Quine places his philosophical logic is a theory of knowledge. But there is a wider context, one which he shares with Davidson and which harks back to pragmatism. In spite of the often formal and technical nature of their writing, their systems, including their discussions of the non-human world – revolve ultimately around a philosophical anthropology, a sustained account of human behaviour and human capacities in general, and of language and linguistic capacities in particular. In Davidson’s philosophy the central place of human beings and human action is obvious. In so far as his work is guided by a single masterproblem, it is this: What is it to understand other human beings? The solution which emerges is provided by a unified theory of meaning and thought which intertwines the philosophy of language, the philosophy of mind and the theory of action (see Evnine 1991: 1–6; Stoecker 1995). Understanding what a person does, what she believes and desires, and what she means by her utterances are part of a single enterprise. The aim is to understand human behaviour by explaining it causally. Nevertheless, the picture of human beings that emerges has strong Kantian affinities. Although we are part of the causal order, we must also treat each other as

37

38

Quine and Davidson on Language, Thought and Reality

rational agents. For, following the principle of charity, the only way of making sense of the mental states of agents, their linguistic utterances and their actions is to regard them as being by and large rational. In Quine the central place of human action is less perspicuous. The ‘logical point of view’ he purports to occupy may appear remote from anthropology. However, his approach to logic itself is ultimately based not on formal considerations, but on a certain picture of human behaviour. Logical truths are ultimately those truths which are ‘well-entrenched’ within human behaviour, which means that we are particularly reluctant to abandon them. In pragmatist fashion, Quine tries to demystify one of the perennial problems of philosophy – logical necessity – through an account of human behaviour, in his case one which is uncompromisingly empiricist and behaviourist. A more general emphasis on action emerges in Quine’s naturalized epistemology. He became increasingly preoccupied with the genetic question of how human beings acquire both language and beliefs in the course of their interactions with the environment, including their interactions with other humans (RR; FTL). Quine’s instrumentalism ties knowledge to human behaviour: we actively construct our theories, and that construction is driven by the imperative of explaining the world around us. This anthropological dimension carries over from epistemology to metaphysics. Although Quine attaches great importance to ontology, he conceives of it in an instrumentalist and thereby anthropocentric way. The ontological question of what exists is transformed into the question of what kind of entities we should ‘posit’, that is to say, postulate to explain and predict our sensory experiences (see chapter 2, section 1). Quine and Davidson accord a special place to linguistic behaviour, and in this respect, if not in their metaphilosophical views, they are part of the linguistic turn of twentieth-century philosophy. Like Wittgenstein and Ryle, they resist the idea that we understand words simply through relating them to objects in reality. Instead they stress that language is itself a form of behaviour embedded in non-linguistic activities. Again in line with Wittgenstein, they are committed to the idea that human speech is not just a more or less faithful expression of human thought and rationality, but its essential medium. This is why the key to their accounts of human behaviour is their discussion of radical translation/interpretation, that is, of the problem of what it is to understand others on the basis of their actions and utterances.

Logical pragmatism

There is also a connection with pragmatism. Peirce maintained that ‘all thought whatsoever is a sign, and is mostly of the nature of language’ (1934: 281). Furthermore, like our logical pragmatists, he developed his account of language as part of a theory of human activity. The meaning of words can be determined only in the context of human action and interpretation. This idea is enshrined in Peirce’s semiotic triangle: to be meaningful, a sign must not just stand for an object, it must stand for an object to an ‘interpretant’. Davidson uses a similar model in his theory of ‘triangulation’. And he is so convinced of the close tie between thought and language that he is prepared to accept the obvious if controversial consequence, namely that non-linguistic creatures like animals and neonates are incapable of thought (see chapter 9, section 4). With these observations I pass from historical presentation to the critical discussion of substantive issues. Both Quine and Davidson subscribe to the idea that human practice is central to philosophy because it embodies the content and structure of human thought and because it holds the key to seminal problems like that of meaning and of logical necessity. This idea is controversial. Traditionally, philosophers have focused their attention on abstract forms, God and Cartesian souls. Nowadays, their focus is on brains, computers and genes. In my view, however, Quine’s and Davidson’s emphasis on human beings and human activities is the most interesting and fruitful aspect of their work. I also agree with them on the importance of linguistic behaviour. At the same time I believe that the picture of linguistic behaviour that emerges from their work combines profound insights with serious distortions. In the remainder of this book I shall try to substantiate this charge by developing what I take to be the insights and by criticizing what I regard as distortions.

39

2

Ontology

The logical positivists denounced ontology as a branch of metaphysics that is either trivial or meaningless. Proper philosophy refrains from competing with science by making claims about reality. Instead, it turns into the ‘logic of science’, a second-order investigation into the conceptual framework of the empirical sciences. When Heidegger held forth about ‘Being’ and ‘Nothing’, Carnap had no problem in showing that it involved a linguistic blunder, a case of mistaking quantifiers for names of peculiar objects. But from the fifties onward, attitudes changed. Instead of having a good laugh about Heidegger’s ‘The Nothing noths’, analytic philosophers took up ontology themselves, and with a vengeance. It is generally assumed that ontology deals with two problems which are more fundamental than those of epistemology, semantics or even logic: What kinds of things exist? What is the nature or essence of these kinds? Quine was the first and most influential force behind the rise of analytic ontology. Prima facie, this is ironical. For Quine is ‘no champion of traditional metaphysics’. Like the logical positivists, he would regard the idea that ontology studies Being as based on reification. He denies that a priori philosophical reflection can establish what kinds of things there are, and he ridicules the idea that philosophers are capable of getting essences into the cross hairs of their intellectual periscopes. Nevertheless, Quine finds a use for ‘the crusty old word’ which he regards as nuclear to its traditional employment (WP 203–4). Like traditional ontology, Quine’s naturalistic ontology seeks to establish what kinds of things there are. But it does not pursue this aspiration directly or in isolation. Instead, it attempts to help science in drawing up an inventory of the world. In this chapter, I shall touch on Quine’s own preferred ontology, but concentrate mainly on his conception of the nature of ontology.

[40]

Ontology

Section 1 provides a brief sketch of that conception, which revolves around his criterion of ontological commitment. Section 2 will uphold the common charge that this criterion is itself irreducibly intensional and therefore incompatible with Quine’s preference for a purely extensional ontology; but it also provides a qualified defence of his standard of ontological admissibility – ‘No entity without identity’. The theme of the following two sections is that Quine’s criterion is incompatible with the general idea of ontological commitment it purports to capture. It wrongly sets aside singular terms in favour of quantification (section 3), and ignores the existential implications of predicates and sentences (section 4). It is also wrong in holding that we treat everything we quantify over as existent. This lends support to a deflationary approach to ontology (section 5). Pace Quine, talking about numbers, properties and propositions does not necessarily involve the acceptance of the Platonist picture of entities beyond space and time. In the final section, I shall further contend that Quine’s method of logical paraphrase does not alter our existential assumptions, but only our way of expressing them. Consequently, the crafting of such paraphrases does not constitute a philosophical contribution to the scientific quest for the ultimate constituents of reality. The proper ‘ontological’ task of philosophy lies not in arguing for or against the existence of entities, but in clarifying what existential claims concerning different types of things amount to.

1

Ontological commitment and ontological parsimony

Ontological questions are already prominent in Quine’s technical work from the 1930s (WP chs. 18, 26). Furthermore, he first came to philosophical fame in 1948 with the article ‘On What There Is’. Its beginning is both abrupt and brilliant: A curious thing about the ontological problem is its simplicity. It can be put in three Anglo-Saxon monosyllables: ‘What is there?’ It can be answered, moreover, in a word – ‘Everything’ – and everyone will accept the answer as true. (FLPV 1)

Quine’s answer to the ontological problem presupposes that everything, that is, every thing exists. This places him in the Aristotelian tradition, which treats the terms ‘existent’, ‘being’, ‘thing’ and ‘object’ as equivalent (see OR 100–1). There is another tradition, which treats ‘existence’ as narrower than ‘being’ and ‘thing’, notably as confined to concrete

41

42

Quine and Davidson on Language, Thought and Reality

things. Thus Meinong maintained that abstract objects ‘subsist’ without existing. Quine follows tradition in yet another respect. Ontology is not concerned with the existence of individual things in particular spaces or at particular times, but with the question ‘What kinds of objects are there?’ (FLPV 1). Indeed, its primary concern is not even whether there are, for example, unicorns or black holes, but with the existence of highly general categories of things, such as material objects, mental phenomena or abstract entities. Quine’s conception of ontology rests on five main points: A. the naturalistic assimilation of philosophical and scientific questions of existence; B. the instrumentalist view that establishing the existence of objects of a certain kind is a matter of establishing that such objects should be ‘posited’ for the sake of explaining experience; C. the attempt to find a criterion of ‘ontological commitment’, a way of determining what types of entities a given theory assumes to exist; D. the method of formal paraphrase or regimentation: the ontological commitments of a theory are established and, where possible, reduced by translating it into a formal language (‘canonical notation’); E. a programme of ontological parsimony that combines a qualified nominalism with an unqualified extensionalism.

(A) In line with his naturalistic conception of philosophy, Quine denies that there is a qualitative difference between the existence questions debated in the special sciences – e.g. ‘Are there prime numbers greater than 100?’, ‘Are there quarks?’ – and the existence questions posed by philosophers, such as ‘Are there numbers?’ or ‘Are there material objects?’ The philosophical questions are special only in being concerned with the most general and fundamental ‘furniture of our universe’ (WO 254, see 271–2; WP 211). (B) Our ontological views are an integral part of our ‘overall conceptual scheme’, our most comprehensive world-view, which includes both scientific and common-sense beliefs. They serve the purpose of providing the simplest structure which fits the ‘disordered fragments of raw experience’. Hence they are subject to the same pragmatic considerations as scientific theory building in general. These criteria include explanatory power, simplicity, conservatism, modesty, precision, facility of computation and the avoidance of perplexities. Such considerations speak strongly in favour

Ontology

of physical objects, which are paradigms of what is real. Nevertheless, even physical objects are merely ‘posits’. They are not simply given as part of reality, but postulated for the sake of simplifying our account of the ‘flux of experience’ (FLPV 16–17, 44–5, 105–6; WB 54). ‘Our talk of external things, our very notion of things, is just a conceptual apparatus that helps us to foresee and control the triggering of our sensory receptors in the light of previous triggering of our sensory receptors’ (TT 1, see 15, 20). (C) Our statements and beliefs commit us to the existence of certain entities. For example, by accepting (1) There are prime numbers greater than 1010

we also accept that numbers exist, since this statement implies (2) There are numbers.

Quine’s initial question is not the ontological one: What sort of objects do in fact exist?, but rather: What sort of objects do given beliefs or theories take to exist? He is concerned with ‘ascribing’ rather than ‘evaluating’ ontologies (FLPV viii, see 15–16, 105–6, 116–17, 130). However, in Quine’s naturalism, the two questions are intimately linked. Since science embodies the whole of human knowledge, it is also our best guide to what exists. By establishing what things our best current scientific theory takes to exist, we also provide the best theory of what things actually exist. Still, as we shall see, Quine’s ontology does not simply follow current science, it is also driven by standards of ontological admissibility which are distinctly philosophical. In order to avoid commitment to objects of an unwelcome kind, we must establish what the ontological commitments of our theories are. This requires a criterion for determining what entities a theory is obliged to accept as existing. The general idea of ontological commitment is straightforward: a theory is committed to those objects which must exist if it is true (OR 93). A theory T assumes the existence of entities of type K iff T entails or presupposes that there are entities of type K. The point of Quine’s criterion of ontological commitment is to pinpoint which types of expressions occurring in T commit us in this way. (D) Quine broaches that topic by translating our theories into a formal language. Such ‘regimentation’ serves two closely connected purposes. First we reveal the ontological commitments of our present theories by translating them into a formal language. Then we reduce these commitments by reforming that language. The ultimate goal is to provide a

43

44

Quine and Davidson on Language, Thought and Reality

‘canonical notation’, that is, a logical notation that is ideal not for the purposes of practical life, but for the purposes of science and scientific philosophy. A canonical notation displays our ontological commitments and keeps them at a minimum, while being adequate for expressing all our scientific theories. Though ‘austere’, it is capable of displaying ‘all traits of reality worthy of that name’. It does not set limits to the primitive vocabulary employed by science. For example, it does not tell us whether physics should reckon with four fundamental forces rather than a single one. But it provides a framework for scientific theorizing in delineating ‘what counts as scientifically admissible construction’ from that vocabulary (WO 228–9, see 158–61, 221, 242). In his quest for a canonical notation, Quine starts out with the firstorder predicate calculus, which contains the logical constants (propositional connectives, quantifiers, identity-sign), individual variables, singular terms and general terms (predicates). Prima facie, singular terms (proper names, definite descriptions and indexicals) are paradigmatic carriers of ontological commitment, since they purport to refer to or denote objects. But that suggestion runs up against the ‘Platonic riddle of nonbeing’ (FLPV 1–3, 7). It would mean that negative existential statements like (3) Pegasus does not exist

are self-defeating. For if (3) were about Pegasus, its truth would imply that it cannot be talking about anything, and hence that it is meaningless. Quine solves the riddle by eliminating singular terms from his canonical notation (FLPV 5–8; ML §37; WO §§37–8). First proper names are converted into definite descriptions (‘the thing that pegasizes’); then definite descriptions are paraphrased according to Russell’s theory of descriptions: (3
) There is nothing which pegasizes.

(3
) lacks singular terms and hence does not give the impression of referring to a non-existing entity. The resulting notation contains only propositional connectives, predicates, the identity-sign, quantifiers and their variables. It reveals ontological commitments by showing what existential statements a given theory implies. (3
) has the form ‘∼∃xFx’; it denies rather than implies the existence of something that pegasizes. Quine concludes that the ontological responsibility of our language lies exclusively with the bound variables of quantification. For example, in order for ‘∃x (Fx & Gx)’ to be true, the domain over which the bound variable ‘x’ ranges must contain at least one object which is both F and G.

Ontology

Quine’s criterion of ontological commitment is based on the objectual interpretation of quantification rather than the substitutional one. According to the objectual interpretation, that ‘Fx’ is true in all/some cases means that it is true of all/some members of a domain of objects. (x)Fx: =o for all objects x in the domain D, Fx. ∃xFx: =o for at least one object x in the domain D, Fx.

According to the substitutional interpretation, it means that all/some of the substitution instances of ‘Fx’ are true, that is, any/some result of replacing the variable by an expression from a substitution class is true. (x)Fx: =s all substitution instances of ‘Fx’ are true. ∃xFx: =s at least one substitution instance of ‘Fx’ is true.

For Quine, substitutional quantification is irrelevant to ontology. What counts are the values of the variables, the objects in the domain of quantification over which the variables range, not the substituends for these variables, that is, the expressions that can be substituted for them (PL 91– 4). Accordingly, a theory T is ontologically committed to entities of type K iff some entities of type K must be counted among the values of the variables in order for T to be true. ‘To be is to be the value of a bound variable’ (FLPV 14–15, 103; see ML 224; OR 94–6). This famous slogan epitomizes Quine’s criterion of ontological commitment. We assume the existence of objects of a given sort by quantifying over them, that is, by employing variables of quantification that must be presumed to range over such objects, to include them among their values. However, our commonly accepted beliefs and theories refer to or quantify over a wide range of things, such as particulars, times and places, properties, relations, quantities, propositions, numbers, facts, species, actions, beliefs, possibilities, institutions, etc. Quine’s ambition is to clear out this ontological ‘slum’, which he regards as a ‘breeding ground for disorderly elements’ (FLPV 4). According to Occam’s razor, we should admit the existence of as few entities as possible. Quine is equally keen to ‘tighten our ontological belts’ wherever possible (OR 17). Therefore his paraphrase of our beliefs in canonical notation is not indiscriminate. Rather, it is guided, by ‘ontological parsimony’ and ‘economy’ (WO 270; PL 69) on the one hand, and by standards of ontological admissibility like clarity and scientific respectability on the other. The question is: ‘how economical an ontology can we achieve and still have a language adequate to all purposes of science?’ (WP 201).

45

46

Quine and Davidson on Language, Thought and Reality

In pursuit of an optimal ontology we have to take ‘ontic decisions’, namely whether to admit entities of a certain type into our universe, or to reject them by avoiding quantification over such entia non grata (WO §50). These decisions are guided by pragmatic considerations based on a trade-off between the systematic efficacy (explanatory power) attained by admitting a certain type of object and the ontological economy achieved by excluding it (WO 238, 264). (E) Goodmann and Quine (1947: 143–4) initially adopted a thoroughgoing nominalism, renouncing abstract objects ‘altogether’ on the basis of an irreducible ‘philosophical intuition’. Later Quine adopted a conditional nominalism (WO 243n, 268–9; FLPV 116–17). Ontological commitments are reduced by avoiding quantification over abstract objects as far as possible. Quine’s programme of ontological parsimony proceeds by ‘explication’. It does not aim at synonymy, but eliminates expressions in favour of alternatives which serve the same cognitive purposes while avoiding unwanted ontological commitments (WO §§53–4). We get rid of attributes, for example, by paraphrasing (4) Red is a colour

as (4
) For all x, if x is red then x is coloured

which avoids commitment to the existence of attributes. In the same vein, numbers and ordered pairs are reduced to classes. In general, Quine’s repudiation of abstract objects is contingent on the availability of an acceptable paraphrase of the scientific statements in which they feature. But there is an exception to this pragmatic stance, the ‘Flight from Intension’ (WO ch. VI). His repudiation of intensional notions is absolute as far as the language of science is concerned. Quine is a conditional nominalist, yet an unconditional extensionalist (OR 21; TT 182–4; RHS 397). This double standard draws not just on his preference for extensional languages like the first-order predicate calculus, on the grounds of their greater simplicity and familiarity. It is first and foremost based on a second famous epigram: ‘No entity without identity’ (OR 23; see FLPV 4, ch. IV; PL 1–2, 67–8). Positing entities of any sort requires criteria of identity, principles for individuating entities of that kind. Unlike abstract extensional objects (numbers, classes), Quine contends, intensions lack such criteria. The predicates ‘creature with a heart’ and ‘creature with a kidney’ have the same extension – they apply to all and only the same

Ontology

individuals – but without expressing the same attribute. According to Quine, the attempt to specify what distinguishes these attributes is doomed, because it relies on the idea that the two predicates differ in meaning, which in turn proves to be vacuous. 2

Intensions and criteria of identity

Logical explication always involves a terminological elimination, but one which goes hand-in-hand with an illumination of the subject matter of the explicated terms (WO 265; MVD 91). Quine’s elimination of intensions is more radical. In his canonical notation we cannot express the things we can express in our intensional idiom of modal notions or propositional attitudes. According to Quine, however, this merely goes to show that the idiom was irrelevant for scientific purposes in the first place (WO 221; OR 23). Quine’s own preferred ontology contains two categories, physical objects and classes, a category of abstract extensional objects he retains since it is indispensable for explicating numbers (WO §§53–5; WP 242– 5; FLPV 117; SS ch. 1, 40–1). Thus his conditional nominalism results in Platonism, albeit of a selective and grudging kind. For the hope of dispensing with abstract terms altogether is kept alive by the prospect of technical innovations. In this vein, Field (1980) tried to avoid classes by reformulating physics without reference to numbers. Ontological parsimony can also lead to the opposite result. Thus Quine contemplated a ‘brave new ontology’ that achieves economy by dispensing with physical objects in favour of classes (FM 164; TT ch. 1). Quine’s conception of ontology is prima facie reasonable. It avoids both the Platonist Scylla of accepting abstract entities uncritically, and the nominalist Charybdis of rejecting them dogmatically. A scientific and pragmatic spirit seems to prevail. Nevertheless, each of the five pillars of Quine’s approach can be questioned. (B), the idea that even physical objects are posits, not to mention the plot of depositing them in favour of classes, is incompatible with the fact that we can perceive such objects directly, without inferring their existence from more basic data, and with his own insight that they are paradigmatic examples of what is real (WO 3; FM 159; TT 9; see Hanfling 2000: 208–11). Furthermore, (C) threatens to be inconsistent with (E), because Quine’s criterion of ontological commitment is either blatantly inadequate or presupposes intensional notions, contrary to his purely extensional

47

48

Quine and Davidson on Language, Thought and Reality

ontology. One line of attack maintains that the commitment relation is intensional in that it suffers from referential opacity. Commitment to Twain, the story goes, does not entail commitment to Clemens, even though both ‘Twain’ and ‘Clemens’ refer to the same person (Scheffler and Chomsky 1958). But this objection takes for granted that one cannot be committed to something without knowing that one is. On a more plausible view, what a person is committed to cannot be ascertained simply from establishing what statements she would assent to under conditions of ignorance. If Sarah maintains that Twain is the greatest American humourist, then she is thereby committed to the existence of Clemens. If Sarah does not know that Twain is Clemens she will not recognize this commitment, but that does not detract from the fact that she is committed. Two more compelling reasons to suspect intensionality have been pointed out by Richard Cartwright (1954). The first is that the idea of ontological commitment involves modal terms: a theory is committed to those objects which must exist for it to be true. The same goes for Quine’s criterion: theory T is committed to objects of kind K iff it is necessary that if T is true, objects of kind K are among the values of its variables. Quine himself frequently uses phrases like ‘must be’ or ‘has to be’ in this context (FLPV 13–14, 103; WP 205). But he thinks that a purely extensional version can be given: ‘to say that a given existential quantification presupposes objects of a given kind is to say simply that the open sentence which follows the quantifier is true of some objects of that kind and none not of that kind’ (FLPV 130–1). For example, the statement ‘There are black holes’ presupposes the existence of black holes because ‘x is a black hole’ is true of all and only black holes. This formulation is extensional, all right; but for that very reason it is also inadequate. It implies that false existential quantifications are not committed to any entities whatever, since their open sentences (i.e. predicates) are not true of any objects whatever. The theory of phlogiston, for example, would not be committed to the existence of a substance emitted during combustion, simply because the open sentence ‘x is phlogiston’ is not true of anything. Quine could avoid this consequence by modifying the above criterion: to say that a given existential quantification presupposes objects of kind K is to say that, if the open sentence is true of any objects, it is true of some objects of kind K, and none which are not of that kind.

Ontology

Unfortunately, this creates a new shortcoming. If the antecedent is false, that is, if there are no objects of which the open sentence is true, then the conditional will be true, irrespective of what kind K is at issue. Accordingly, if an existential quantification is false, it is thereby committed to everything that does not exist. By this token, since classical mechanics is committed to the existence of the ether, it is also committed to the existence of phlogiston and of the Homeric gods, which is an exceedingly unkind imputation. Quine cannot avoid the objection through replacing the material implication by entailment. This would bring in an intensional notion, since to say that p entails q is to say that ‘p→q’ is analytic or necessarily true (FLPV 130; WP 137–42). One might try to salvage Quine’s position by holding that a theory T is committed to objects of kind K iff the most parsimonious translation of T quantifies over objects of kind K. But this move once more lumbers Quine with intensional notions. For one thing, it would seem that the most parsimonious translation is the one which gets by with as few ontological commitments as possible. For another, translation is a paradigmatic intensional notion, since a translation has to preserve meaning. This also indicates a second reason for suspecting intensionality. Quine’s criterion applies directly only to theories which are phrased in terms of objectual quantification. With respect to other theories, it runs: a theory is committed to all those entities which must be reckoned among the values of the variables of a quantified translation, if the theory is to be true (see FLPV 105, 131). It transpires, therefore, that Quine’s flight from intensions is incompatible with his criterion of ontological commitment. What about the standard of ontological admissibility that motivates the flight? It demands that the existence of a type of entity (events, for example) can only be accepted if we can provide ‘criteria’ or ‘conditions’ of identity for entities of that type. ‘We have an acceptable notion of class, or physical object, or attribute, or any other sort of object, only insofar as we have an acceptable principle of individuation for that sort of object. There is no entity without identity’ (TT 102; see OR 19). Whether this demand is reasonable depends on what one makes of criteria of identity. A criterion is a way of recognizing something, or a feature by which we can recognize something. A criterion of identity is something by which we can recognize the correctness or incorrectness of a statement of identity. Questions of identity make sense only once we have specified what kind of thing is at issue. One cannot simply ask ‘Is this the same as that?’, but only ‘Is this the same S as that’, where ‘S ’ is a sortal term like

49

50

Quine and Davidson on Language, Thought and Reality

‘table’ or ‘planet’. Accordingly, a criterion of identity is a feature which determines whether or not an object falling under ‘S’ that we encounter in one context or refer to in one way is the same as an object falling under ‘S’ that we encounter in another context or refer to in another way. Quine’s slogan can be understood as insisting that objects of a type S can be posited only if it is possible to specify formal criteria of identity which apply to all objects of that kind in a rigid and clear-cut fashion. In that case, however, the slogan is implausible. Our generally accepted statements and theories refer to and quantify over a wide variety of things for which no such formal criteria of identity are in sight. Indeed, the only uncontroversial examples concern abstract sortals like ‘direction’ and ‘set’ (set x is identical with set y iff x and y have the same members). It can even be questioned whether such criteria are available for physical objects, which Quine concedes to be paradigms of what exist (Strawson 1997: 39–43). Quine could weaken his standard of ontological admissibility by insisting not on a general formal principle, but only on enumerability. If we are to treat Ss as distinct objects, questions of identity must be determinate and clear-cut enough to allow for counting. Thus Quine insists that propositions, the purported meanings of sentences, ‘have to be the same or distinct absolutely’ (WO 203). In the same spirit, Davidson writes that questions of meaning must be sharp, ‘if meanings are entities’ (ITI 101). If Ss are genuine entities, it must be possible to establish how many of them there are. This weaker principle excludes many of the intensional objects Quine detests. There is no way of answering the question of how many attributes Reims cathedral has, or of how many propositions there are in the Critique of Pure Reason. Even specific subcategories of attributes by and large do not allow for enumeration. There are no ways of establishing how many colours, intellectual qualities, character traits, literary styles, affective attitudes, smells, and ways of walking or talking there are, to name but a few. Yet this is also the Achilles heel of the weaker standard. As Strawson has pointed out, these things are ‘as distinctive, as easily recognizable, as anything in human experience’; and only a philistine form of scientism would deny that they are the subjects of ‘pointful predications’ (1997: 23, 35). Indeed, enumerability is a question not of the degree to which certain types of entities are real, but of the degree to which the corresponding sortals are regimented. There are ways of counting the attributes Bohr’s 1913 theory ascribes to electrons or the axioms of Newtonian mechanics, simply

Ontology

because these theories draw precise distinctions in these respects. Even loads and cargoes, things of a particularly artificial and derived type, can be precisely counted, following the stipulations of shipping contracts or customs regulations. Conversely, although mountains are supremely real, there is no cast-iron method of counting them. This leaves an even weaker reading of Quine’s slogan. ‘There is nothing you can sensibly talk about without knowing, at least in principle, how it might be identified’ (Strawson 1997: 22). In this sense, however, we do possess criteria of identity for determinables like colour or character trait simply by virtue of knowing how to distinguish the corresponding determinates. Although we cannot count colours or character traits, we can distinguish different colours and character traits. That is to say, we can tell that an object is red rather than green, or that a person is arrogant rather than timid. Identifiability is all we require for the purpose of talking meaningfully about things of a certain kind. But though catholic in principle, this standard of ontological admissibility can exclude purported types of things, namely if their alleged identifiability turns out to be spurious. Quine’s thesis that translation is indeterminate implies that this is the case for meanings. We think that we can identify meanings in the sense that we can specify what meaning an expression has. But, Quine alleges, this impression is deceptive, since incompatible specifications have an equal foundation in fact. On this reading, ‘No entity without identity’ is not so much a standard of ontological admissibility as a standard of referential clarity. We must be clear as to what we are talking about. Even this standard faces resistance, however. In recent years it has been common to insist that talk about entities of a certain kind requires only ‘metaphysical’ or ‘constitutive’ principles of individuation, rather than ‘epistemic’ or ‘evidential’ criteria of identity (e.g. Williamson 1990: 148; Lowe 1997: 622). As long as one can specify what the identity of objects of a certain kind consists in, what it is for x to be the same S as y, it does not matter whether we can establish whether they are the same. Following this lead, Quine’s opponents could argue that even the insistence on identifiability amounts to a confusion of epistemological and ontological issues. This complaint is unjustified. The requirement of identifiability is semantic or conceptual. Perhaps ontological questions are prior to epistemological ones. Yet semantic questions are prior to both, since matters of meaning antecede matters of knowledge and of fact. There can be true or

51

52

Quine and Davidson on Language, Thought and Reality

false answers to the question of whether entities of type S exist only if that question is meaningful to begin with. But it can be meaningful only if the term ‘S’ is itself meaningful. And how could it be, if the relevant facts leave underdetermined what S we are talking about? Even champions of metaphysical principles of individuation admit that not just any necessary and sufficient conditions of identity do the trick. It will not do, for example, to state that expression x has the same meaning as y iff x and y are synonymous. Principles of individuation must be informative and non-circular. In the absence of such principles, talk about one thing being identical with or different from another will be vacuous. Our discussion so far yields the following result. Quine’s flight from intensions will be justified if and only if translation and meaning are indeed indeterminate. In that case, however, he must leave his criterion of ontological commitment behind as well. 3

Ontology and singular terms

Quine’s criterion of ontological commitment applies only to theories formulated in his canonical notation. Among ‘champions of ordinary language’, this invites the complaint that the criterion has no bearing on assessing the existential assumptions of our unregimented beliefs (whether everyday or scientific). In Quine, one finds two distinct responses to this objection. The first is that questions of ontological commitment do not arise for ordinary language because of its unruly nature (FLPV 106; TT 9). Accordingly, the aim of regimentation is not to bring out the commitments of our actual conceptual scheme, but to replace the latter by a better one in which there are clear existential assumptions. Unfortunately, this response is incompatible with the very idea of ontological commitment, which is to capture the existential assumptions of our unregimented beliefs and theories. In fact, other passages grant this point. Regimentation serves the purpose of ‘understanding the referential work of language and clarifying our conceptual scheme’, not just of reforming that scheme (WO 158; see FLPV 105–6, 117–18; PT 77). Why? Because an appraisal of our actual ontological commitments is a precondition for revising them. It is only proper, therefore, that Quine defends his criterion of ontological commitment on the grounds that the quantifiers of canonical notation capture the sense of our ordinary ‘there is’ (see section 4).

Ontology

Quine’s second response is that clarity about our actual statements is best achieved by translating them into canonical notation (FLPV 107; PL 3, 9–10, 16; RHS 228). Ordinary language conceals our ontological commitments, hence our beliefs must be paraphrased in canonical notation. But this amounts to a crucial concession. Canonical notation must capture the relevant aspects of our non-philosophical discourse. Therefore Quine’s criterion of ontological commitment must be judged as an account of the existential assumptions of our non-regimented discourse. As such it is problematic in several respects. The first one is the elimination of singular terms. Quine concedes that singular terms are ‘indispensable in practice’, for example in mathematics. Nevertheless, he insists, no cogent objections have been raised against the eliminability of singular terms in principle (SS 60; RDH 326–7). But this is far from obvious. For one thing, if singular terms are indeed indispensable in practice, they will at best be paraphraseable, not eliminable. For to say that x is indispensable to y is to say that y cannot get by without x. The point is not simply pedantic, especially given the pragmatist/instrumentalist aspirations behind canonical notation. Canonical notation is supposed to reveal and minimize our ontological commitments. Its more general purpose, however, is to facilitate the explanation and prediction of experience. If singular terms are essential to this overall cognitive enterprise, as Quine rightly grants, they simply cannot be eliminated from canonical notation. For another, Perry (1979) has argued powerfully that it is impossible to even paraphrase away indexicals without loss of content and explanatory power. A third difficulty for Quine is that quantifiers and variables are no less paraphraseable than singular terms. In the combinatory logic of Schonfinkel variables are supplanted by so-called predicate functors. ¨ Ironically, Quine himself helped to popularize combinatory logic. He recognizes that his criterion of ontological commitment does not apply to such systems directly, but insists that it applies unproblematically to their translation into quantified formulae (SLP ch. 23; SS 33–5, 101–5; WP 304– 5). However, this leaves intact the main problem. If the paraphraseability of singular terms is a reason for denying that they carry ontological commitments, the paraphraseability of variables will be a reason for the same denial concerning variables (Haack 1978: 47–8). A final worry is that Quine’s paraphrases may be intelligible only given a prior understanding of the singular terms they replace. This worry arises at several levels. Quine himself intimates that ‘x pegasizes’ can be

53

54

Quine and Davidson on Language, Thought and Reality

understood as meaning ‘x = Pegasus’. This comment is not meant to define the predicate, for if it were, the singular term would not have been eliminated. But without this unofficial explanation, the predicate remains unexplained. Russell did not face this difficulty, since he explicated names through complex descriptions. Quine recognizes that in many cases suitable descriptions are hard to come by. But the contrived predicates through which he tries to avoid this problem can be made intelligible only by reference to the names they replace. The elimination of singular terms may also eliminate the possibility of explaining other parts of Quine’s own canonical notation (Strawson 1959: 194–8, 215–16; 1971: ch. 3). Quine himself suggests that the distinct roles of singular and general terms must be explained via their functions within singular predications like ‘This rose is red’ (WO §20). In that case, however, it is difficult to see how the notion of predication could make sense without that of reference. To predicate is to classify a thing along certain parameters, and this presupposes the idea of reference, of singling out the object to be characterized or classified. Consequently, Quine’s canonical notation trades on a tacit understanding of the singular terms which it is the explicit purpose of the notation to avoid. An analogous objection can be raised concerning variables and quantification. Far from the notion of a singular term relying on quantification, the latter tacitly relies on propositions which combine singular terms and predicates. This is not to say that the quantifiers can be explained simply by reference to conjunctions or disjunctions. That explanation fails when it comes to quantification over infinite or indefinite domains. But the question remains whether objectual quantification is intelligible without an understanding of singular predicative propositions such as ‘Fa’. We cannot understand what it is for all or some objects in the domain to be F unless we understand what it is for a particular object to be F. Such an understanding requires an understanding of singular reference and hence of singular terms. After all, Quine himself stresses that variables are the formal equivalents of pronouns. Yet the anaphoric reference of pronouns clearly presupposes non-anaphoric reference by other singular terms. Quine holds the reverse position. In domains which are both finite and specified, ‘ontological considerations lose all force; for we can then reduce all quantifications to conjunctions and alternations and so retain no recognizably referential apparatus’ (WP 216). If we only quantify over a domain of objects a1 to an , ‘(x)Fx’ will of course be tantamount to ‘Fa1 & Fa2 & . . . & Fan ’ and ‘∃xFx’ to ‘Fa1 ∨ Fa2 ∨ . . . ∨ Fan ’. Nevertheless, Quine’s claim is

Ontology

perverse. The singular terms which occur in the conjuncts and disjuncts are not only ‘recognizably referential’, they are the very paradigms of referring expressions. Quine resists this obvious point on the grounds that in a language without quantifiers ‘the very distinction between names and other signs lapses . . . since the mark of a name is its admissibility in positions of variables. Ontology thus is emphatically meaningless for a finite theory of objects, considered in and of itself’ (OR 62, see 106; ML 26–8; RHS 397). But this claim would fail, even if Quine were right to maintain that only singular terms are accessible to quantification (Kunne 1983: chs. 1.1, ¨ 3.3). Quine defines a singular term as one which can be replaced by variables of quantification. But consider a language which contains a finite number of names yet no quantifiers. In such a language, this definition does not apply, yet we can still distinguish singular and general terms. If it contains expressions like ‘is the same as’ or ‘=’, singular terms can be defined as those which can occur to either side of the identity-sign (Dummett 1981: 59–60; Tugendhat 1982: ch. 3). And if it does not, we can fall back on Strawson’s proposal (1971: ch. 5). There is an asymmetry between general and singular terms, in that the former but not the latter belong to systems of mutually exclusive expressions. In non-quantified statements like ‘Tom Tower is 30m high’, we can replace the general term but not the singular term in such a way that it is logically impossible that the result should have the same truth-value as the original. Many replacements for ‘Tom Tower’ will in fact alter the truth-value, but no replacement from the substitution class of singular terms will do so necessarily. Again, many replacements of ‘is 30m high’ will not lead to a change of truth-value, but there are always replacements that will, in our case from the system of height-attributions (‘is 100m high’, etc.). Far from being the only way of distinguishing singular from general terms, Quine’s explanation is not even the best. For this reason, singular terms are not tied to quantification. Indeed, Quine’s conclusion that ontological questions do not arise for singular predicative statements is a reductio ad absurdum of his position. If the idea of ontological commitment is to make any sense at all, then saying that Jesus came from Nazareth commits one to the existence of Jesus and of Nazareth, whether or not one indulges in quantification over infinite domains. Of course, we can only express these commitments in terms of quantifiers (vernacular or formal). But it does not follow that only existential statements, let alone statements quantifying over infinite domains, carry the commitment. Consequently,

55

56

Quine and Davidson on Language, Thought and Reality

by confining ontological commitments to quantification over infinite domains, Quine violates the general idea of ontological commitment which he purports to spell out. These objections rule out attaching exclusive ontological significance to variables at the expense of singular terms. At the same time, Quine has good reasons for resisting the opposite extreme (WP 204–5; OR 93–5, 106; PT 27–8; SS 32–3). Even singular terms in subject position need not commit one to the existence of an entity, as in ‘Pegasus has wings’. Conversely, we are capable of committing ourselves to objects that cannot be named individually, for example through the true statement that there are unspecifiable real numbers. Finally, ontological commitments that attach to singular terms can be expressed in terms of quantification, because the statements concerned entail or presuppose first-level existential statements. In normal cases, ‘Fa’ will imply both ‘∃xFx’ and ‘∃x (x = a)’ (OR 93–7; WO §37). Consequently, Quine may yet be right to focus on quantification, not because singular terms are eliminable or without ontological import, but because singular predicative statements standardly imply existential statements. 4

Ontology and quantification

Quine maintains that his criterion of ontological commitment is a truism (TT 175; PT 25–7). Theories take to exist what they say that there is, or what must exist if they are to be true. This point is at odds with his restriction of ontological commitment to quantification over infinite domains, but it is fundamentally correct. Eighteenth-century chemistry, for example, was committed to the existence of phlogiston, a hypothetical substance emitted during combustion; for if there is no such substance, that theory cannot be true. Nevertheless, Quine’s criterion captures what we take to exist only if his objectual ‘∃x’ actually encapsulates the existential commitments of our ‘there is’ and its cognates. Now, the criterion is specifically designed to ensure that only the variables that replace singular terms carry ontological responsibilities. Abstract singular terms like ‘redness’ but not concrete general terms like ‘red’ commit us to attributes. This is directed against two kinds of ‘offenders’ (WO 240–1; see FLPV 9–13). On the one side, there are what I call ontological inflationists, who hold that even concrete predicates like ‘is red’ commit us to intensions. On the other side, there are ontological deflationists, who hold that even employing abstract singular

Ontology

terms like ‘redness’ or quantifying over numbers does not commit us to dubious abstract objects. Prima facie, at least, the inflationist offenders are right by Quine’s own lights. For general terms in predicative position are also accessible to quantifiers and hence seem to commit their users to the existence of attributes (Dummett 1981: 59–61; Strawson 1971: 65–6). (5) Betty is witty and Sarah is witty

Fa & Fb

entails not just (6) Someone is witty

∃xFx

but equally (7) There is something Betty and Sarah both are (viz. witty)

∃ (a & b)

In repudiating this objection, Quine relies on the objectual interpretation of quantification. He alleges that locutions like (7) are ‘meaningless’ and do ‘violence to grammar’ (ML §34). This allegation is based on the conviction that all genuine existence statements amount to ‘There is a thing/object x such that’ and can hence be captured by objectual quantification. Such is simply the intended sense of the quantifiers ‘(x)’ and ‘(∃x)’: ‘every object x is such that’, ‘there is an object x such that’. The quantifiers are encapsulations of these specially selected, unequivocally referential idioms of ordinary language. (WO §49; see FLPV 102–6; OR 97, 106; PL 89; TT 174–5)

On this assumption, (7) is indeed infelicitous. It would have to be rendered as (7
) There is an object  such that a is  and b is .

Yet (7) does not claim that there is an object – the property of being witty – Betty and Sarah both are, but that there is something they both are, namely witty. But why should quantification have to be understood objectually? Quine’s answer is: ‘Variables are pronouns, and make sense only in positions which are available to names’ (WP 198; see ML §34). However, in natural languages pronouns can occur in other positions. To be more precise, these languages feature not just pro-nouns, as in ‘Betty is witty but she is

57

58

Quine and Davidson on Language, Thought and Reality

gentle’, but also pro-adjectives (‘Not just the Kremlin is red, the Tower is so too’) and pro-verbs (‘Sarah goes on vacation, and so does Peter’). Quine ignores the fact that ‘There is something which’ has a wider scope than ‘there is an object which’, the idiom on which he focuses. ‘Something’ is syntactically transcategorial: it can quantify into the positions of both singular terms, as in (6), and of predicates, as in (7). Pace Quine, therefore, there are perfectly acceptable ‘idioms of ordinary language’ which capture 1983: ch. 3.3). quantification into predicate position (ML §16; cf. Kunne ¨ The question arises of how the predicative quantification of (7) should be understood. The standard move is to conceive of it in terms of substitutional quantification (Haack 1978: 50–6). But that move is susceptible to Quine’s objections. Substitutional quantification cannot deal with objects no one has ever labelled or attributes no one has ever ascribed (OR 63–7, 106; Q 35–6). Fortunately, this difficulty can be avoided by a third option, an understanding of ‘∃’ which is neither objectual nor substitutional but what Prior called ‘non-nominal’ (1971: ch. 3; see Kunne ¨ 2003: ch. 6; Hugly and Sayward 1996: 174–208). It amounts to a nonsubstitutional understanding of quantification into predicate position. The quantifiers bind not name variables but predicate variables. That is to say, the substitution class of the variables is not the class of singular terms but the class of predicates. By contrast to substitutional quantification, the variables have not just substituends but also values, a range of objects with which they are associated, namely properties. By contrast to objectual quantification, the substituends of predicate variables do not name these values (attributes), they ascribe them. As Quine himself stresses (WP 204; PL 28, 66–7), a predicate like ‘is a fish’ applies to every fish without in addition naming a property. But properties can be introduced into discourse not just by referring to them, as in (4), but also by ascribing them to something – as in (4
). It emerges, therefore, that Quine cannot convict the inflationist offenders simply by ruling statements like (7) out of court. One can make existential claims without using the objectual quantifier, and his criterion of ontological commitment is too narrow. According to Quine, all values are objects (WP 182). But in that case, to be is not simply to be the value of a variable. There is something Betty and Sarah both are. Yet what they both are is not an object – the property of being witty; what they both are is witty. While ‘wit’ is the name of a property, ‘witty’ is not. This gives rise to an even more serious predicament. Quine’s criterion of ontological commitment turns accessibility to quantification into

Ontology

the hallmark of existential assumptions. But even if one stuck to the untenable restriction to the position of singular terms, replaceability by ‘something’ would transpire to be a trivial operation. If you insulted me, you did something; and if it is certain that I will die, then something is certain (Rundle 1979: 199–200). Most importantly, replaceability by ‘something’ does not entail that the replaced phrases even purport to refer to an object or something that exists. If Pegasus does not exist, then something does not exist, and if the round square is a Meinongian nightmare, then something is a Meinongian nightmare. We can talk about and quantify over ‘all kinds of everything’,1 including things that do not or cannot exist. Consequently one cannot impute serious ontological responsibilities to phrases merely because they can be replaced by ‘something’. Quine could retort that such responsibility must attach to ‘there is something . . .’ and its cognates. It may indeed sound awkward to say that there are things that do not exist. But there is a use of vernacular quantifiers that does not necessarily imply existence. There are, for example, fictional characters with a stammer, and there is nothing awkward in denying that fictional characters exist. And even if one were to insist that fictional characters exist, one could not maintain that they are real. It is more plausible to hold that quantification requires enumerability. Yet even this requirement is too strong for quantifiers like ‘some’. Some attributes are instantiated and some propositions are true, even though we have seen that there is no canonical way of counting attributes or propositions. Reference does not presuppose reality, existence or even enumerability, but rather identifiability. It must be possible to specify what we are talking about, whether it be through a name, a description or by ostension (direct or deferred). We can refer to and quantify over those things we can identify, which includes fictional characters among other things. By this token, Quine’s answer to the ontological question is wrong. Not everything exists. We refer to and quantify over things that do not. And Quine is even further off target when he invokes ‘the ordinary usage of “real”’ in support of his contention ‘Everything, of course, is real’ (WP 225). These conclusions in turn lend credence to the deflationist offenders, whom Quine charges with underestimating their ontological commitments. Though reference to and quantification over things is much more pervasive than Quine allows, it may also be less harmful. To be sure, there 1. As in the song ‘All Kinds of Everything’ by D. Lindsay and J. Smith: ‘Dances, romances, things of the night, sunshine and holidays, postcards to write, budding trees, autumn leaves, a snowflake or two, all kinds of everything remind me of you!’

59

60

Quine and Davidson on Language, Thought and Reality

is a difference between fictional and abstract entities, in that reference to and quantification over the latter does involve existential assumptions of some kind. But the question remains whether it commits us to a dubious Platonist metaphysics. 5

Ontological deflationism

Platonism holds that abstract objects like numbers, properties and propositions inhabit a super-natural world beyond space, time and causation, a world to which we – or at any rate suitably trained logicians and mathematicians – have access by a kind of ‘intellectual intuition’. Nominalists protest that this hinterworld is a myth, and that the signs which apparently refer to its denizens are flatus vocis that can be avoided in respectable discourse. Ontological deflationists try to undermine this debate. For them, the way to avoid Platonism does not lie in purging our theories of reference to and quantification over abstract objects. Like Quine, they view the prospects of this enterprise as dim. Unlike Quine, they regard the attempt to avoid intensional abstract terms as equally futile. In fact, even the prima facie favourable case of properties provides ample scope for such scepticism (see Putnam 1975a: ch. 19; Prior 1967). Unlike the Platonists, however, the ontological deflationists do not accept that such failure would lumber us with mysterious entities. Once we clarify what it is for abstract singular terms to refer, they claim, we shall realize either that they do not refer to genuine objects at all or that the existence of abstract and intensional objects is a perfectly intelligible and low-key affair. I shall discuss three different versions of deflationism: existential, objectual and linguistic. Existential deflationism is based on Ryle’s claim that the term ‘exists’ is ambiguous. In the sturdy spatio-temporal sense in which material objects exist, abstract objects do not, but they were never meant to by our pre-philosophical statements about numbers, properties or propositions (1980: 23–4). The dispute between Platonism and nominalism would then be based on a failure to keep apart different notions of existence. Quine concedes that ‘in a derivative way’, there are ‘two senses of existence’, a ‘common sense’ and a ‘philosophical sense’. In the former, a nominalist can accept ‘There are prime numbers’ as a fac¸on de parler. But the philosophical sense in which the nominalist must deny that there are numbers ‘is rather to be respected as literal and basic than deplored as a philosophical

Ontology

disorder’. In that literal ontological sense, someone adverse to Platonism must either ‘paraphrase or retract’ the statement that there are prime numbers (OR 98–100; see Q 35–6). According to the Oxford English Dictionary, ‘to exist’ is ambiguous, namely between ‘to have a place in the domain of reality’ and the more demanding ‘to live’. ‘Lenin exists’, for example, is true on the first interpretation, false on the second. But this is not the ambiguity between ‘existence’ and ‘existence in space and time’ claimed by Ryle. For Quine, that claim falls foul of the proper criterion for ambiguous terms. A term is ambiguous iff in one and the same context it can make for both a true and a false statement. On this basis, he can insist that the difference between stones and numbers is not one between two kinds of existence, but between two kinds of objects (WO 241–2, §27; FLPV 3, 131; Rundle 1979: §3; 1990: 74–7). However, our preceding discussion suggests that statements about fictional characters, e.g. ‘Quasimodo exists’, might satisfy this criterion: they are true or false depending on whether or not ‘exists’ is understood as presuming reality. Furthermore, Quine’s criterion of ambiguity can be contested. If change of truth-value must be possible in all contexts, then it excludes paradigms of lexical ambiguity: in ‘I had a drink of port’ the understanding ‘naval harbour’ makes for nonsense rather than a statement with a different truth-value. If change of truth-value must be possible only in some context, then univocal terms come out as ambiguous: a single utterance of ‘This wine is awful’ can express something true and express something false depending on one’s standards, but ‘awful’ is not ambiguous. Finally, the argument can be turned against Quine. If ambiguity is captured by his criterion, this only goes to show that it is a rare phenomenon, which does not exhaust the possibility of equivocations. The criterion entails, for example, that the differences between ‘Socrates is snub-nosed’ and ‘Socrates is the husband of Xanthippe’ cannot derive from an ambiguity of ‘is’. For in the context of ‘Socrates . . . snub-nosed’, glossing ‘is’ as ‘is identical with’ does not turn a true statement into a false one, but into ungrammatical gibberish. Nevertheless, Quine would be the first to acknowledge that there is a fundamental difference between the ‘is’ of predication and the ‘is’ of identity. Similarly, even if the word ‘exists’ is univocal, statements claiming the existence of concrete and abstract objects respectively may yet have a different semantic role. This would explain

61

62

Quine and Davidson on Language, Thought and Reality

why a sentence like ‘There are dogs, prime numbers and properties’ has an air of zeugma; it is reminiscent of ‘The bench, the liquor and the exam were hard’. Existential deflationism is closely linked to objectual deflationism, a position that derives from Wittgenstein. It maintains that so-called abstract objects are not really bona fide objects at all. They lack many of the characteristic features of ordinary objects, for example the capacity to be visualized or pointed at (Wittgenstein 1978: 262–3; see 1960: 19, 64; Dummett 1981: 70–80; Dilman 1984: 71). This claim invokes the commonsense notion of an object as a distinct, relatively enduring and sufficiently large material object. However, there is also a wider logical conception associated with Frege. An object is simply anything one can refer to: a is an object iff ‘a’ is a singular term. Quine shares that conception, except that for him the vehicles of reference are variables rather than singular terms. An object is anything that is the value of a variable of quantification. Wittgenstein complained that according to the logical conception even the coincidence of an eclipse of the moon and a court case is an object (1975: §§93, 115). Kunne (1983: 41) has replied that there is a perfectly intelligible ¨ and non-philosophical sense in which one can regard this coincidence as an object. One can refer to it and claim of it, for example, that it was an object of heated debates. But this only goes to show that even the vernacular operates with two distinct notions of an object – object of reference or discourse on the one hand, material object on the other. It does not show that there is no difference between the two. Although holes and gaps are objects of reference, ‘The hole in the object grew larger’ and ‘There is a gap between the two objects’ definitely do not imply ‘The object in the object grew larger’ or ‘There is an object between the two objects’. For all his rhetoric against Ryle, even Quine writes: ‘If drastic difference of objects makes for a difference of sense [concerning ‘exists’], then I must concede that the difference between abstract and concrete should indeed do so.’ He resists that conclusion only on the grounds of his general scepticism about criteria of identity for ‘senses’ (Q 191). Accordingly, his objection is not so much that terms like ‘exist’ and ‘object’ are univocal, but that the notion of ambiguity, like that of meaning, is vacuous to begin with. Objectual deflationism now comes to this. One must grant that numbers, attributes, propositions, etc., are objects of reference. But this is only to grant that we can identify and talk about them. Just as the concrete singular term ‘Betty’ occupies a referring position with respect to ‘is witty’ in (5), we can move to second-level predication in which a corresponding

Ontology

abstract singular term can occupy a referential position relative to a higher level predicate: (8) Wit is desirable in a friend.

This shows that properties are topics of speech, namely of higher-level predication. However, the analogy between abstract and concrete objects is ‘strictly limited and purely logical’ (Strawson 1971: 74). Although ‘wit’ indicates what attribute we are talking about, it does not single out something which we could point at if only we ever managed to visit the Platonic realm. As long as one really confines oneself to the logical notion, there is nothing mysterious about numbers, attributes and propositions being objects. (8) applies a principle of classification (‘desirable in a friend’) to a property, which for these purposes is designated by a nominalization of the firstlevel predicate. To make sense of that possibility, all we need is the idea that human beings have the ability to engage in second-order discourse, that is, to refer to and to classify properties. We do not need to postulate a realm of entities that are just like material objects, except that they inhabit a realm beyond space, time and causation. The Platonist view that such a realm is required arises only when the logical notion of an object is illicitly combined with distorted connotations of the everyday notion of a material object. On a minimal logical conception of ‘object’ that avoids these pitfalls, numbers, attributes and propositions are indeed objects. But that concession is itself minimal. It does not license Quine’s fear that reference to or quantification over numbers, attributes and propositions involves the heavy metaphysical prize of postulating a mysterious kind of reality.

6

Linguistic frameworks and logical paraphrase

Quine faces a further challenge, namely by linguistic deflationism. According to this position, associated with Carnap and Wittgenstein, ontological questions are not just a low-key affair, once freed from metaphysical metaphors, but downright vacuous. Admittedly, there are genuine questions of existence. But these are specific; they concern the existence of particular kinds of things and can be solved by the special sciences, including logic and mathematics. By contrast, ontological statements like ‘There are material objects’ or ‘Universals do not exist’ are at best misleading

63

64

Quine and Davidson on Language, Thought and Reality

expressions of allegiance to certain conceptual frameworks or ‘forms of language’ (see Dilman 1984: chs. 1–3; 1996; Arrington 1996). In this vein, Carnap (1956: Supl. A) distinguished between two kinds of existence questions. ‘Internal questions’ are raised from within a ‘linguistic framework’ (an interpreted formal language) and involve the existence of particular (groups of) entities. ‘External questions’ concern the framework itself and involve the existence of a whole system of entities. Scientific questions like ‘Are there neutrinos?’ or ‘Are there prime numbers greater than 1010 ?’are of the former kind, while philosophical questions like ‘Are there mental events?’ or ‘Do numbers exist?’ belong to the latter. Internal questions are ‘theoretical’ in nature, that is to say they allow of true or false answers which are reached by proceeding in accordance with the rules constituting the framework (rules concerning methods of observation and computation). By contrast, external questions are either meaningless or ‘practical’ in nature. They do not concern the ultimate constituents of reality, as the metaphysical tradition had it, but merely the adoption of a particular linguistic framework. The question of whether numbers exist is either internal to our mathematical framework, in which case the answer is trivially ‘yes’, or it boils down to the question of whether we should adopt a ‘number language’, in which case it is to be answered on the basis of pragmatic considerations. But an affirmative answer to that pragmatic question involves no commitment to the existence of numbers, and hence no Platonist metaphysics. Quine’s view that scientific and philosophical existence questions do not differ in kind was developed in explicit opposition to Carnap (WP ch. 19; OR 91–3). The main thrust of his critique is that the external/ internal distinction is so intimately tied to the analytic/synthetic distinction that it collapses with the latter. Recent commentators have rightly repudiated this charge of guilt by association (Bird 1995; Yablo 1998a: 235–7). The two distinctions do not run parallel; for instance, analytic existence claims in mathematics are not external. Admittedly, Quine’s attack on the analytic/synthetic distinction threatens the possibility of distinguishing between the external linguistic framework and the internal empirical statements made within it (WP 211). But we shall see in the next chapter that this attack is less formidable than its reputation. Unfortunately, Carnap for his part has given us no reason for accepting that philosophical questions of existence differ sharply from scientific ones in concerning the adoption of a way of speaking (Stroud 1984: 180–97; Hale 1987, ch. 1). But perhaps this lacuna can be filled by exploiting a

Ontology

widely noted feature of ontological questions and statements. Unlike ordinary existence claims and questions, they often strike even philosophers as peculiar or ridiculous. Can there really be a serious question as to whether numbers, properties, languages or computer programmes exist? A remark by Wittgenstein (1967a: §58) helps to elaborate this unease. If an ontological statement like (9) The colour red exists

means anything at all, it is either an empirical claim, namely (9
) There are things which are red

or a statement about language, namely (9∗ ) The word ‘red’ has a meaning in our language.

Accordingly, ontological claims are either trivially true (it is obvious, for example, that there are red things), or they do not concern reality but our conceptual framework, just as Carnap had it. These two interpretations of (9) are indeed natural. Still, Quine would insist that there is a third legitimate interpretation, namely something like (9#) Canonical notation features the abstract singular term ‘the colour red’ (or the ensuing quantifications), and not just general terms like ‘red’

For Quine, the question is whether we must include such terms in a language that captures ‘all traits of reality worthy of that name’. And he would further maintain that to answer this question is the proper task of ontology. But Carnap and Wittgenstein can retort that (9#) is just as linguistic as (9∗ ); it amounts to nothing other than an answer to an external question, namely the question of whether we should adopt a linguistic framework involving abstract colour terms. Interestingly, Quine concedes that in discussing philosophical problems of existence we tend to shift from talk about objects to talk about words. But he denies that this ‘semantic ascent’ marks a significant methodological difference (WO §56; see FLPV 15–18, 44–6, 77–9). Semantic ascent, he assures us, occurs in the sciences as well. Furthermore, in so far as an ontological question like that concerning the existence of colours is a pragmatic question about the adoption of a ‘linguistic framework’, it is no more so than the question of whether our conceptual framework should reckon with the existence of dodos and quarks. In both cases the alteration of the framework is subject to the same empirical constraints. Accordingly,

65

66

Quine and Davidson on Language, Thought and Reality

the fact that ontologists often talk about words does not show that ultimately their questions concern language rather than reality. Here we have reached the crux of the matter. According to Quine’s naturalism, ontological reforms of canonical notation are on a par with the adjustment of scientific theories to new empirical evidence. ‘Each such reform is an adjustment of the scientific scheme, comparable to the introduction or repudiation of some category of elementary physical particle’ (WO 123). Ontological statements ‘admit of evidence, in the sense that we can have reasons, and essentially scientific reasons’ for including numbers or classes in the range of values of our variables, and against including propositions, attributes, or unactualized objects (OR 97). If reference to a type of entities can be avoided through logical paraphrase, this shows that entities of this type do not exist, just as the Michelson–Morley experiment established that the ether does not exist. Conversely, if reference to a type of entities is both part of our best scientific theories and immune to logical paraphrase, this shows that entities of this type do exist, just as the experiments of Cowan and Reines established that neutrinos exist. By this token, the paraphrases of (4) into (4
), and of (10) Betty possesses wit

into (10
) Betty is witty

are on a par with scientific discoveries concerning the nature of reality. However, the idea that logical paraphrase has such momentous consequences can be challenged. Paraphrase is a symmetrical relation (Alston 1958; Wright 1982: ch. 1.5). If adequate, the paraphrases will say the same as the originals. In that case, however, (4
) implies that red is a colour no less than (4), and (10
) implies that there is a property Betty possesses no less than (10). Nevertheless, it seems that the existential commitments of (11) The average Briton drinks 100 litres of beer per annum

are indeed brought out by the contextual paraphrase (11
) If you divide the litres of beer consumed by Britons each year by the number of Britons, you get 100

which makes no reference to the average Briton. In this spirit, Roger Teichmann (1992: 21–4) has argued that there can be an asymmetry

Ontology

between S and S
, namely if the sentences which parallel S in their logical properties of entailment and compatibility are more often of the form S
. For example (12) Humility is a virtue

is properly rendered as (12
) If someone is humble, she is (pro tanto) virtuous

because its logical powers differ from those of (13) Fido is a dog.

Unlike (13), (12) entails statements of the form ‘If a is F, then a is G’ (‘If Jane is humble, she is virtuous’) and does not entail statements like ‘a is something that b is not’ (‘Fido is something that Betty is not’). But if we disregard the difference between first- and second-order predication, (12) does entail statements of this form, e.g. ‘Humility is something that greed is not’, namely a virtue. In one crucial respect, moreover, (12) is like (13). Both entail a statement of the form ‘There is at least one F’. There is a difference between (12) on the one hand and (11) on the other. While (11) entails neither that there is an average Briton nor that there is someone who drinks 100 litres of beer per annum, (12) entails both that humility exists and that there is at least one virtue. Pace Quine, moreover, the switch from (10) to (10
) is no more than an ‘insignificant quirk of grammar’ (WO 241). It involves a change not in existential beliefs, but only in linguistic form. Quine protests that there is a logical difference between singular and general terms. But that there is a difference between the terms ‘wit’ and ‘is witty’ does not entail that there is a difference in existential assumptions between the sentences (10) and (10
). For the difference between the terms is cancelled out by the remainder of the respective sentences. The moves from (10) to (10
) and from (12) to (12
) do not diminish our commitment to attributes. In fact, (12
) obfuscates rather than illuminates (12). By contrast, in cases like (11), Quinean paraphrase does clarify our ontological commitments. But even there it does nothing to reduce them. The reason why (11
) brings out the commitments of (11) is precisely that the former was never a statement about a putative object – the average Briton – to begin with. Someone who entertains the (in any event curious) desire of marrying the average Briton will simply have misunderstood the original statement (11).

67

68

Quine and Davidson on Language, Thought and Reality

While it is clear how an explicans can be superior to an explicandum in terms of clarity, it is unclear how it can avoid ontological commitments. Whether someone asserts or is committed to the existence of certain things depends on what she says, not on how she says it. One cannot repeal these admissions simply by modifying one’s way of expression. Ergo, the possibility of paraphrasing sentences in a way that avoids explicit reference to entities of a certain type does not reduce our ontological commitments. To use Alston’s example, someone who is afraid of policemen would be reassured if she were convinced that there are no policemen, but not by the fact that one can paraphrase ‘There is a policeman around the corner’ by ‘Policemanhood is instantiated around the corner’. The Quinean extensionalist nicely fits Strawson’s description ‘Committed in thought to what we shun in speech, we should then seem like people seeking euphemisms in order to avoid explicit mention of distasteful realities’ (1997: 58). It is indeed part of the business of science to avoid existential commitments, but not through logical paraphrase. By postulating both phlogiston and oxygen, Priestley’s theory creates empirical and theoretical problems that are avoided by Lavoisier’s theory, which does away with the former. However, this is not a matter of paraphrasing Priestley’s theory, but of developing a new empirical theory that contains no reference to phlogiston, in whatever manner, and yet accounts for the empirical data. Similarly, nuclear physicists may want to avoid commitment to quarks, for example because they cannot be observed in isolation. But to that end it does not suffice to reformulate every sentence implying the existence of quarks, for example, into one implying the instantiation of quarkhood. Rather, it is a matter of postulating a new set of basic particles which account for the empirical data without having the drawbacks of quarks. In short, it is a matter of scientific investigation, not of armchair paraphrase `a la Quine. Quine’s hostility to intensions is based not on empirical evidence, but on a priori scruples concerning the conceptual framework within which that evidence is weighed. He admits that the philosopher’s task differs from the scientist’s ‘in detail’, in that the former ‘makes explicit’ the assumptions that guide empirical science (WO 275–6; see OR 98–9). Yet he insists that his ontological reforms are nonetheless indirectly connected with experience through their place within a web of beliefs. But even if accurate, this holistic picture provides no argument for assimilating the extensionalist purge of language to the scientific quest of limning the ultimate traits of reality. That two things are connected does not show that there are no qualitative differences between them.

Ontology

I conclude that Quine’s conception of ontology is flawed. There is nothing philosophically daring about referring to or quantifying over abstract objects, or claiming that they exist, provided that we can explain what it is that we are talking about. What has to correspond to our abstract terms is not a system of dubious entities, but an intelligible use. Quine may yet show that in the case of intensional terms the appearance of intelligibility is deceptive (see chs. 7–8). But until then, we can simply answer the question ‘Do numbers, attributes and propositions exist?’ with a resounding ‘Yes!’, since this follows from true and well-established statements. The existential question has been settled, and the philosophical task turns to clarifying what it is for numbers, attributes and propositions to exist, and in what sense they count as real. ‘Ontological’ problems are solved not by positing or depositing entities. Instead, they are solved by clarifying what our various existential claims amount to, and by explaining and contrasting both general ontological notions like ‘object’, ‘thing’, ‘existent’, ‘real’, and notions from specific forms of discourse, like that of a natural number, of an uninstantiated property or of an unexpressed proposition. Quine’s logical paraphrase will constitute an important part of such an enterprise. Unlike scientific research, however, it tells us nothing about reality, but at best something about our existential claims and the conceptual framework within which they are made. At this point, Quine might redraft his project of ontological reform. Perhaps our ontological commitments are reduced not by paraphrasing our best actual beliefs, but simply by replacing these beliefs through different ones. Indeed, my suggestion of reducing ontology to the clarification of our existential beliefs seems to ignore the possibility that our non-philosophical beliefs – including those of science – are based on false existential assumptions. But this move is not available to Quine. His naturalism leaves him without a platform from which to criticize the claims of scientific disciplines. And in so far as Quine is not just a naturalist but also a pragmatist, he has no platform for interfering even with non-scientific forms of discourse, provided that they are cognitively successful. More importantly still, the revisionist move is inherently dubious. To settle the question of whether nonphilosophical existence statements are true, all one can do is to establish whether the conditions obtain that these statements claim to obtain. And while philosophers can help in clarifying what these conditions are, they hold no special brief for establishing whether they obtain. At best they can show that some existential statements could not possibly be true, because they involve inconsistent concepts like that of a round square. To suppose

69

70

Quine and Davidson on Language, Thought and Reality

otherwise is to go back to the pre-Kantian and un-Quinean idea of philosophy as a super-science capable of overruling the special sciences. Rather than saving Quine, the protest highlights an incoherence in his position. On the one hand, his naturalism construes philosophical problems of existence as factual or internal. On the other, his extensionalism repudiates the answers which, on this construal, are appropriate, because of metaphysical scruples for which he has left no room. Right now, however, it is time to turn to the source of his view that ontology is continuous with science, namely his attack on the analytic/synthetic distinction.

3

Analyticity, apriority and necessity

Statements like ‘g = 9.81 m/sec2 ’ or ‘Radioactivity causes cancer’ may be physically necessary. Their denial may be incompatible with the laws of nature, but it is not contradictory. By contrast, it seems that statements like ‘∼(p & ∼p)’, ‘2 + 2 = 4’ and ‘All material objects are located in space’ are necessary in a stronger sense. They must be true, no matter what the world and the laws of nature are like, since their denial is contradictory. By a similar token, our knowledge of such statements seems to be a priori. It is independent of experience, not as regards its origin (such knowledge is not innate), but as regards its validity: we can justify these statements without any reference to experience. Radical empiricists reject these ideas. They claim that all truths are contingent and that all knowledge must be based on experience – even that of apparently a priori disciplines like logic, mathematics and metaphysics. According to Mill, the truths of mathematics are in effect wellsupported inductive generalizations. As Frege pointed out (1953: §§7–10), this is dubious, since it makes the truths of pure mathematics depend on contingent physical facts. For this reason, the logical positivists settled for a moderate empiricism. Logic and mathematics, they conceded, are necessary and a priori; but they do not amount to knowledge about the world. For all a priori truths are analytic, that is, true solely in virtue of the meanings of their constituent words. Logical truths are tautologies which are true in virtue of the meaning of the logical constants alone, and analytical truths proper can be reduced to tautologies by substituting synonyms for synonyms. Thus ‘All bachelors are unmarried’ is transformed into ‘All unmarried men are unmarried’, a tautology of the form ‘(x) ((Fx & Gx) → Gx)’. Necessary propositions, far from mirroring the super-empirical essence of reality or the structure of pure reason, are true

[71]

72

Quine and Davidson on Language, Thought and Reality

by virtue of the conventions governing our use of words (Ayer 1971: 21–4, ch. 4). This position is known as conventionalism. It comprises an analytic theory of a priori knowledge, according to which only analytic truths can be known independently of experience, and a linguistic doctrine of logical necessity, according to which the source of such necessity lies in language. Views of this kind were not confined to logical positivism, but shared by some ordinary language philosophers. Quine attacks them from the perspective of radical empiricism. He denies that there is a qualitative difference between contingent propositions and the allegedly necessary propositions of logic, mathematics and philosophy. But unlike previous radical empiricists, Quine does not simply reduce necessary propositions to empirical generalizations. Instead, he backs his assimilation by an ingenious line of argument which questions the very distinctions that have traditionally been used to set philosophy and science apart: analytic/synthetic, necessary/contingent, a priori/a posteriori, conceptual/factual. Quine’s crusade has revolutionized this entire debate, and some of his points against the logical positivists are decisive. I shall argue, however, that he has not demonstrated the spuriousness of these distinctions, and that his empiricist assimilation of the necessary and a priori to the contingent and a posteriori is untenable. Section 1 resists his claim that the notion of analyticity is obscure because it can only be explained through other intensional notions. Next I partly endorse his attack on the idea of truth in virtue of meaning (section 2), while insisting that it does not preclude a linguistic account of analytic propositions as constitutive of meaning/concepts (section 3). In section 4, I argue that this account is compatible with the fallibilism supported by Quine’s epistemic holism. In fact, a distinction between conceptual and factual propositions is essential if one is to make sense of Quine’s holistic picture of a web of belief and of meaningful discourse (section 5). Section 6 is more sympathetic. Although Quine’s attacks on essentialist modal logic are unconvincing as they stand, he is fundamentally right to resist the idea that there are metaphysical necessities independent of our way of conceptualizing the world. 1

The circularity charge

In ‘Two Dogmas of Empiricism’, Quine tries to discredit two positions of logical positivism: the analytic/synthetic distinction on the one hand,

Analyticity, apriority and necessity

reductionism on the other. The conclusion of these combined attacks is a ‘blurring of the supposed boundary between speculative metaphysics and natural science’ and ‘a shift towards pragmatism’ (FLPV 20), namely the idea that pragmatic considerations apply across the board, and not just to the external questions singled out by Carnap. Quine does not accept the conventionalist account of logical truths as true by virtue of meaning/convention, and he denies that they differ essentially from empirical truths. Yet he does not reject the notion as unintelligible. A logical truth is one in which only the logical particles ‘occur essentially’, that is, one which ‘remains true under all reinterpretations of its components other than the logical particles’ (WP 80–1, 110, 130; FLPV 22–3). At the same time, he challenges the proponents of the analytic/synthetic distinction to provide an equally clear explanation of analyticity. Quine shows that this notion belongs to a cluster of intensional concepts including ‘synonymy’, ‘self-contradiction’, ‘necessity’, ‘definition’ and ‘semantical rule’. These concepts can only be explained by using other members of this family. Analytical truths are propositions which can be reduced to tautologies by substituting synonymous expressions (definitions) for certain constituents. Conversely, two expressions are (cognitively) synonymous if they are interchangeable not just salva veritate but salva analyticitate, which is to say that they are interchangeable even in intensional languages that contain operators like ‘Necessarily . . .’ or ‘. . . is analytic’ (FLPV 20–32). But none of these concepts can be explained in a purely extensional idiom, by means of terms like ‘truth’ or ‘reference’. Quine thinks that this amounts to a vicious circularity. He concludes that intensional concepts cannot be adequately explained, and that the corresponding dichotomies (analytic/synthetic, necessary/contingent) are ill-founded (FLPV 37, 46, 56–7). In one respect, Quine here grants too much. Substitutability salva analyticitate is not sufficient for synonymy or the identity of propositions and attributes, since in that case all necessary propositions and all necessarily co-extensional predicates would have to be synonymous. More importantly, his circularity charge against analyticity is flawed, as Grice and Strawson have shown in their ‘In Defense of a Dogma’. They first demonstrate that Quine has not provided sufficient reasons for discarding the distinction, even if he is right in complaining that it has not been adequately explained (1956: 142–3). Philosophers agree in their application of the analytic/synthetic distinction to an open class of cases, and they also agree on what count as borderline cases. This projective quality of the

73

74

Quine and Davidson on Language, Thought and Reality

distinction shows that there is a difference between analytic and synthetic propositions, even if it has yet to be clarified.1 Quine maintains that the existence of problematic cases is due to the unclarity of the dichotomy (FLPV 32). But he cannot convict the notion of analyticity on these grounds without also convicting an extensional notion he accepts, namely ‘is true’. While it is unclear whether, for example, ‘Everything green is extended’ is analytic, it is equally unclear whether it is true (Grice and Strawson 1956: 153–4; Carnap 1990: 428–9). Should a point of green light, for example, count as extended or not? In fact, the recognition of borderline cases actually presupposes a grasp of the analytic/synthetic distinction. How else are we supposed to recognize what cases lie on the border? The second defence of the alleged dogma is that an explanation which fulfils Quine’s demands is ‘hard to come by’ only because his standards for a satisfactory clarification are inappropriately high (Grice and Strawson 1956: 147–8). For he demands, first, a definition of ‘analytic’ stating necessary and sufficient conditions, and, secondly, a reduction of intensional notions to extensional ones. But the first condition cannot be met even in the case of extensional notions such as truth. Therefore it cannot be considered a precondition for a concept’s making sense without casting doubt on the intelligibility of extensional notions Quine condones. This defence is certainly an example of British understatement, and must be sharpened. Even if all extensional notions could be defined analytically, Quine’s standards of clarification would not just be inappropriate but downright incoherent. This becomes clear once we turn to his second demand, namely that ‘analytic’ be defined in purely extensional terms. It must be emphasized that Quine requires an explanation of the meaning of ‘analytic’. His demand for an explanation that makes ‘analytic’ intelligible is not merely a request that we define the term’s extension. For a specific formal language L1 Carnap defined a metalinguistic predicate ‘analytic in L1 ’ by listing all the analytic propositions 1. Harman (1999: chs. 5–6) criticizes this conclusion. By parity of reason, he maintains, one might conclude that there is a distinction between witches and non-witches, since these terms have been applied to an open class of cases. But this response confuses the existence of a distinction with the existence of things falling on either side of it. There is a legitimate distinction here, roughly between women with and women without certain super-natural powers. It is just that nothing falls in the former class. People who believe otherwise are not suffering from conceptual unclarity (indeed, they mean the same by ‘witch’ as Harman), they are wrong in denying that the class of witches is empty. Harman further suggests that a charge of circularity in definition does bite with respect to technical notions like analyticity. However, as Quine himself acknowledges, the analytic/synthetic distinction corresponds to commonsensical distinctions and intuitions.

Analyticity, apriority and necessity

or ‘meaning postulates’ of L1 (1956: Supl. B). Quine rejected that strategy, on the grounds that it only explains ‘analytic in L1 ’ but not ‘analytic’ (FLPV 33–4, 134). This objection must surely rest on the requirement that an explanation of ‘analytic’ give its intension, and not merely its extension. For the problem with Carnap’s suggestion cannot lie with the restriction to a specific language L1 , since otherwise one could define a general notion ‘analytic in Ln ’ by using the definitions for particular Ls. It follows that Quine’s challenge itself depends essentially upon the use of an intensional concept (‘meaning’ or ‘intension’), whilst he forbids any response which grants itself the same licence. It is not inconsistent for an extensionalist to demand that the meaning of ‘analytic’ be explained; but it is inconsistent to reject putative definitions solely on the grounds that they employ intensional concepts. If intensional notions were unintelligible per se, then Quine’s challenge would be unintelligible; yet in that case, it would neither allow of nor deserve a response. There is another inconsistency between Quine’s insistence that the notion of analyticity be reduced to extensional ones and his requests for an explanation of ‘analytic’. For the latter demand cannot be fulfilled without using those concepts to which, by Quine’s own lights, the explanandum is synonymous, that is, the notions he prohibits. Consequently Quine’s circularity-charge comes down to the rather odd complaint that ‘analytic’ can be explained only via notions with which it is synonymous, and not via notions with which it is not synonymous. It is clear, therefore, that the circle of intensional notions need not be avoided; it is not vicious in the least and it does not set ‘analytic’ apart from any other concept. Accordingly, the circularity-charge does not reveal a shortcoming of intensional notions, but only incoherent ideas about explanation on Quine’s part. One might try to save him by maintaining that his idea of an explanation is not to provide synonyms but to ‘explicate’, that is, to translate into his canonical notation. Ironically, this would mean that Quine has been complaining that the notion of analyticity has been explained, when it should have been explicated. In any event, it would not salvage the circularity-charge. The idea that legitimate concepts must be translatable into a purely extensional language presupposes that intensional notions have been discredited, which is what the circularity-charge set out to do. Nonetheless, it remains possible to claim that explaining analyticity by way of other intensional notions is a case of obscurum per obscurius. But this requires independent arguments to the effect that ‘analytic’ and its intensional relatives are obscure in the first place. Indeed, Quine’s demand for

75

76

Quine and Davidson on Language, Thought and Reality

a clear explanation is itself a smokescreen. Behind it lies an urge not to clarify intensional notions, but to remove them from canonical notation – the ‘flight from intension’. As a result, Quine has to repudiate not just the philosophical notion of analyticity, but also everyday notions like meaning the same as or having meaning (Grice and Strawson 1956: 145–6). In response to Grice and Strawson, Quine has done three things. One is to soft-pedal. Thus he states that he never meant to repudiate the intuitions that sentences like ‘No bachelor is unmarried’ are special in that their denial betokens linguistic misunderstanding. At the same time, he denies that these intuitions sustain a sweeping epistemological dichotomy between ‘analytic truths as by-products of language and synthetic truths as reports on the world’ (WO 66–7). Next, Quine accuses Grice and Strawson of the ‘fallacy of subtraction’, namely of arguing that if a sentence s1 is meaningful and means the same as another sentence s2 , then there must be a meaning, an abstract entity with which both are associated (WO 206–7). But one need not reify ‘meanings’ in order to speak, as Grice and Strawson do, of the meaning of an expression or of two expressions having the same meaning. Knowledge of meaning is not acquaintance with an object. Whereas ‘A knows Willy Brandt’ and ‘Willy Brandt is identical with Karl Frahm’ entail ‘A knows Karl Frahm’, ‘A knows the meaning of “superfluous”’ and ‘The meaning of “superfluous” is identical with the meaning of “redundant”’ do not entail ‘A knows the meaning of “redundant”’. Nor do they entail that there is an entity x such that ‘superfluous’ means x and A knows x. ‘The meaning of s1 ’ does not designate a genuine object, it is equivalent to ‘what s1 means’. The ‘what’ here is not a relative pronoun, as in ‘what s1 is written on’, but an interrogative pronoun (Latin quid rather than quod). It indirectly introduces a question, namely: ‘What does s1 mean?’ That s1 has a meaning means that there is an answer to this question; that A knows the meaning of s1 means that he can answer it; and that s1 has the same meaning as s2 means that the same answer can be given for both (see Austin 1970: 96–7; Rundle 1979: §47). In fact, Quine’s third response belies his claim to respect our semantic intuitions, because it denies that answers to these innocuous questions can be given. He maintains that ‘intensions’ must be rejected because they lack clear criteria of identity: there is no objective way of telling whether two expressions mean the same (OR 23; FLPV 4, ch. 4; PL 1–2, 67). Yet Quine grants that criteria of identity for intensions could be provided, if we could appeal to notions like analyticity, necessity or synonymy: two

Analyticity, apriority and necessity

predicates mean the same attribute, for example, if they are synonymous. Quine rejects that solution precisely because he repudiates these notions as unclear (FLPV 152; OR 19–20; PL ch. 1, 67–8; TT 105). This means that in his attempt to show that intensional notions like analyticity are obscure he cannot take for granted that they lack criteria of identity without circularity, a truly vicious one this time. A non-circular argument is in the offing only if Quine succeeds in establishing that meaning is indeterminate, without assuming that intensional notions are obscure. Quine is right, therefore, when he insists that doubts about intensional and intentional notions must be supported by the thesis of the indeterminacy of translation rather than the other way around (RBG 198–9; see Hookway 1988: 139–41; Boghossian 1997: 331–2; and ch. 7, section 5 below). 2

Holism and truth by virtue of meaning

Although some die-hards still set store by it, Quine himself quietly dropped the circularity charge as early as 1953 (WP 138–9). His second line of argument is more promising. It starts out from the so-called ‘Duhem– Quine’ hypothesis. Our statements do not admit of confirmation or disconfirmation individually, but face the tribunal of experience only as a whole. When we derive a prediction p from a scientific theory T we rely not just on T, but also on auxiliary assumptions about the antecedent conditions A, as well as on mathematical propositions and logical laws, M and L: (T & A & M & L) → p. If we observe that ∼p, we have therefore refuted not T, but only the antecedent as a whole. From this Quine concludes that our beliefs form an all-inclusive and tightly knit ‘fabric’ or ‘web’. ‘The unit of empirical significance is the whole of science’ (FLPV 42). In the first instance, this is an epistemic holism or holism of confirmation. In the second instance, however, it leads to a semantic holism or holism of meaning. Quine subscribes to a verificationist conception of meaning, according to which the meaning of a sentence is in principle determined by the evidence that would count for or against it. But if empirical evidence attaches only to systems of sentences rather than to individual sentences, so will ‘empirical meaning’ or ‘content’ (FLPV 37; OR 78–82; Gibson 1981: 80–1): epistemic holism + verificationist theory of meaning = semantic holism.2 2. For the relation between Quine’s holism and the holisms of Duhem and Carnap see, respectively, Vuillemin 1986 and Isaacson 1992.

77

78

Quine and Davidson on Language, Thought and Reality

Quine uses this holism to attack the logical positivists’ conception of both analytic and logical truths. The analytic/synthetic distinction is based on the idea that each individual truth involves a ‘linguistic’ and a ‘factual component’, and that the latter is zero in the case of both logical and analytical truths (FLPV 41–2; WP 108). (1) Brutus killed Caesar

seems to owe its truth both to what its constituent expressions mean and to a contingent fact. By contrast (2) Brutus killed Caesar or Brutus did not kill Caesar

seems to owe its truth purely to the way we use ‘or’ and ‘not’. Officially, Quine has no qualms about distinguishing a linguistic and a factual component of truth, as long as this is done for a web of sentences rather than individual sentences. At the same time, he insists that one cannot define analytic truths as those without factual content, because it is impossible to distinguish between a linguistic and factual component at the level of individual truths. Holism also removes the need for according to logic and mathematics a special, linguistic content. They have empirical content, namely in virtue of allowing the derivation of predictions from scientific theories (WP 132; FLPV 42; PL ch. 7; RHS 207). This line of argument depends heavily on Quine’s epistemic holism. Originally, this holism portrayed science as an undifferentiated whole in which everything is connected to everything else. But this monolithic picture bears little resemblance to the realities of scientific research (Lauener 1982: 78–9; Sokal and Bricmont 1998: 62–3). It is difficult to see how either elementary arithmetic or observation sentences about brick houses on Elm Street could depend on theories concerning messengerRNA. Quine himself later moved to what he calls ‘moderate holism’. He recognized that in practice it is never the totality of science which is tested, but a set of beliefs that can be isolated. He also conceded that observation sentences normally confront experience separately (FLPV viii; EEW 313–14; WB 13; TT 71). With this concession, however, the holistic argument against analyticity loses its bite. If individual observation sentences have a special status because of their direct connection with experience, why shouldn’t individual sentences like (2) have a special status because they lack any such connection? Furthermore, the concession does not go far enough. According to moderate holism, an individual non-observational sentence s is associated

Analyticity, apriority and necessity

with corroborating or refuting evidence, and hence has empirical content, only on the assumption that certain other sentences are true. But it does not follow that s as such lacks empirical content. When Sherlock Holmes suspects Moriarty of having committed the murder, there can be evidence for this particular claim (e.g. the presence of fingerprints), even though its connection with the suspicion is contingent on certain forensic assumptions. An individual statement can have empirical content, which is why we can ask how it squares with the evidence. Quine’s other criticisms cut deeper. They do not simply invoke holism against otherwise plausible claims about individual statements, but question the very idea that some sentences are true solely in virtue of the meaning of their constituents, independently of the world. This suggests that there is something about a sentence like (2) – its logical form or structure, the meaning of its constituents or a semantical rule – which forces us to hold on to it come what may. But the idea that ‘meanings’ – abstract or mental entities – coerce us to accept sentences like (2) is the ‘myth of a museum’ (OR 27, see 19). Even according to moderate holism, any sentence can be abandoned if necessary. Only the periphery of our ‘web of beliefs’, consisting of observation sentences, confronts experience directly. But even the centre, consisting of the allegedly a priori sciences of logic and mathematics, is indirectly linked to experience. If a scientific prediction is refuted, we could react by abandoning the mathematical and logical propositions used in deriving it from the theory in question (FLPV 42–3; WO 275–6; WP 245; PL 5–6, 99–102). Accordingly, we can abandon analytic sentences, and the idea that their logical form or meaning might prevent us from doing so is mystifying. Obviously, conventionalists cannot subscribe to such a Platonist picture. But without it, the idea of truth by virtue of meaning boils down to the claim that necessary propositions are true by virtue of definition: they follow from the definitions of their constituents (Waismann 1968: 124–6). This takes us back to the proposal that analytic propositions are either logical truths or reducible to logical truths by substituting synonyms for synonyms. But it faces two difficulties. The first has dogged logical positivism from its inception (Carnap 1956: 226–8). Non-empirical propositions like (3) (x) (y) (x is warmer than y → y is not warmer than x)

or (4) Nothing can be red and green all over

79

80

Quine and Davidson on Language, Thought and Reality

resist such reduction, since the ingredient terms are not definable in the appropriate way. Consequently, they would have to be classified as synthetic a priori. The second difficulty was first pointed out by Quine. Standard definitions cannot create truths, since they are simply conventions of abbreviation which license the replacement of a definiendum by a definiens. Consequently, definitions only transform truths, they cannot create them. For example, given the definition (5) Bachelors are unmarried men

one can transform the analytic truth (6) All bachelors are unmarried

into the logical truth (7) All unmarried men are unmarried.

‘What is loosely called a logical consequence of a definition [e.g. (6)] is therefore more exactly describable as a logical truth [e.g. (7)] definitionally abbreviated’ (WP 79). Conventionalism has not accounted for this original logical truth, and therefore it has failed to show that all necessary truths are true by virtue of definition. Quine grants, however, that there is a kind of convention capable of creating truths. Unlike ‘discursive definitions’, which simply record preexisting relations of co-extension or synonymy, ‘legislative postulates’ or ‘definitions’ are capable of making sentences true. Adopting sentences on the basis of ‘deliberate choices’, which are justified only ‘in terms of elegance and convenience’, renders those sentences true by convention (WP 117–20; see FLPV 24–7). In this fashion, we can even render logical truths true by convention. We define the meaning of the logical constants a novo by specifying contexts in which they are to be true and contexts in which they are to be false. The former ‘become true by fiat, by linguistic convention’ (WP 89–91; see LoC 49–60). Yet Quine denies that legislative postulation redeems conventionalism. He objects that this procedure can be extended beyond logic and mathematics to empirical science (WP 100–102, 121–2). But, in his own words, all this shows is that in addition to analytic propositions being true by convention, ‘it is likewise a matter of linguistic convention which propositions we are to make analytic and which not’ (LoC 64). Conventionalists can happily accept that we are at liberty to adopt analytic statements even

Analyticity, apriority and necessity

within empirical science, namely for the sake of defining scientific terms, as in the case of Carnap’s meaning postulates. At the same time, Quine’s own discussion intimates reasons against his initial concession that legislative definitions, at any rate, are capable of creating truths. ‘Verbal usage is in general a major determinant of truth’ (WP 108, see 113–16; FLPV 42). For any sentence s, s is true iff for some p, s says that p and p. All that conventions do is to determine what a sentence says; whether what it says is true is another question, to which linguistic conventions are irrelevant. Therefore, a sentence cannot owe its truth-value exclusively to meaning or conventions (Lewis 1946: ch. 5; Pap 1958: 103–8; Boghossian 1997: 334–6). Indeed, what could it mean for a convention to create a truth? Of course, we can choose to assume that a certain statement is true, in the course of constructing hypotheses or for the sake of argument. But this does not render that proposition true. In the sense in which the fact that the cat is on the mat might be said to render true the statement that the cat is on the mat, conventions cannot be said to render anything true. 3

A linguistic account of analyticity

These points militate against the view that analytic propositions are truths following from meanings, conventions or definitions. But they do not show that there is no fundamental difference between analytic and synthetic propositions. To be sure, no workable distinction will accommodate all the disparate claims made by the various proponents of an analytic/synthetic distinction (see Bealer 1998). Nevertheless, at least one such distinction is legitimate, and it does tie analytic propositions to meaning or language. In fact, it is a distinction Quine himself has come to accept. The early Quine condemned analyticity as a ‘pseudo-concept which philosophy would be better off without’ (WP 171). He rejected any qualitative distinction between necessary and contingent propositions. There is only a difference of degree between the logical and mathematical beliefs at the centre of our web and empirical beliefs closer to the periphery: the former are more ‘obvious’ or ‘firmly accepted’, that is, we are more reluctant to abandon them. Furthermore, logic and mathematics share this obviousness with the empirical observation-sentences right at the periphery. The only legitimate notion of analyticity is a quantitative one: ‘stimulusanalytical’ truths are those which are accepted under any circumstances

81

82

Quine and Davidson on Language, Thought and Reality

(and by all speakers). But this will not distinguish analytical propositions from very well-confirmed empirical propositions like ‘There have been black dogs’ or ‘Lions roar’ (WP 102, 112–13; WO §14; OR 86; PL 96–7; RHS 206). In 1974, however, Quine countenanced a stronger definition: analytical sentences are those which are learned together with language, that is to say, their non-acceptance signifies that a person has failed to learn the meaning of these expressions: ‘a sentence is analytic if everybody learns that it is true by learning its words’ (RR 79; see RHS 93–5). Versions of this idea can be found in various proponents of an analytic/synthetic distinction (e.g. Grice and Strawson 1956: 150–1; Boghossian 1997: 334). According to Quine, it may also ‘approximate the layman’s intuitive conception’ of the analytic as being ‘just a matter of words’ (PLSP 52–3). He is right. Outside of philosophy, no one would say that a Bushism like ‘I am mindful not only of preserving executive powers for myself, but for my predecessors as well’ betokens a risible ignorance of a Platonic form, or of the intrinsic modal nature of reality, or of the holistic web of our beliefs. Instead, people would say that it displays ignorance of what the word ‘predecessor’ means. At the same time, Quine insists that this commonsensical distinction does not have the epistemological and metaphilosophical importance traditionally accorded to it, and that, in fact, it lacks ‘explanatory value’ (WP 113; see WO 66–7; PT 54–5; Hookway 1988: 37–46; Gibson 1988: 93–6). One of his complaints is that the distinction concerns only the acceptance of the sentences in question and hence does not capture an ‘enduring trait’ of the truths thus ‘created’ (WP 119–21; RHS 95, 138, 206). But this defect can be remedied. For Quine’s definition includes a wanton genetic element that is easily jettisoned. Whether a sentence s counts as analytic should not depend on how it came to be accepted, for example on whether it was acquired as part of language acquisition. Rather, it should depend on its subsequent status. What counts is whether the sincere refusal to accept s is compatible with having understood s. Thus one can define as follows: A sentence s expresses an analytic proposition:= s is self-evident, i.e. one cannot understand s without accepting it, independently of empirical information.

Quine insists that in pressing for a clear explanation of analyticity he is demanding no more than ‘a rough characterization in terms of dispositions to verbal behaviour’ (WO 207). And he propounds what Dummett

Analyticity, apriority and necessity

(1978: 377–8) calls Quine’s ‘inextricability thesis’: there is no way of disentangling convention and experience as determinants of our disposition to linguistic behaviour. But whatever the merits of Quine’s behaviourist requirement, my definition fulfils it. This is brought out by a reformulation: A sentence s expresses an analytic proposition:= if a speaker x sincerely denies or rejects s, this shows either that x fails to understand s, or that x is deliberately employing s in a novel sense.

Whether in a linguistic community the refusal to accept s is treated as a criterion for factual ignorance or for a divergence in linguistic understanding is evident from the reaction of its members. Consider the following two statements: (8) My seven-year-old daughter understands Russell’s theory of types. (9) My seven-year old daughter is an adult.

To (8) we would typically react with disbelief and by demanding evidence. By contrast, we fail to understand (9) and will demand an explanation of what, if anything, the speaker means by an adult. In response to (8) we might say ‘I don’t believe what you say’ – provided that we take the statement to be in earnest. In response to (9) we might say ‘I don’t understand what you mean’ – provided that we do not take it as shorthand for ‘My seven-year-old daughter behaves like an adult’. The contrast between analytic and synthetic falsehood is displayed in the difference in use between terms like ‘doubt’ and ‘incredible’ on the one hand, and terms like ‘misunderstanding’ and ‘unintelligible’ on the other. Of course, this distinction between analytic and synthetic propositions allows of borderline cases. Nevertheless, it is clear and manifest in linguistic behaviour. It also helps the logical positivists out of one quandary, because my definition of ‘analytic’ covers cases like (3) and (4). But it does not necessarily rehabilitate their overall position. For one thing, it might exclude complex logical and mathematical propositions, thereby allowing for synthetic propositions a priori. It is generally accepted that one can understand such propositions without knowing whether they are true. One can try to extend the definition to those propositions that can be derived from self-evident propositions by inferences that themselves must be acknowledged by anyone who understands them. Even with this proviso, however, many mathematical theorems may turn out to be synthetic, since neither they nor the steps required to derive

83

84

Quine and Davidson on Language, Thought and Reality

them from self-evident axioms are evident in this way (see PL 98; WB 43–6; Kunne 1983: ch. 5.5). ¨ At the same time, my line of reasoning supports a linguistic conception of at least those necessary propositions that are analytic in my sense. If ignorance/knowledge of analytic propositions is ignorance/knowledge of meaning, then these propositions have a special connection with meaning. But what is that connection, if they are not true by virtue of meaning? The most promising suggestion is that necessary propositions do not follow from the meanings of their component signs, rather, they partly determine or constitute these meanings. According to Wittgenstein, the propositions of logic implicitly define the logical constants. Whether a specific transformation of symbols is licensed or not is one aspect of the established use and hence of the meaning of the terms involved. For example, that we treat ‘∼∼p → p’ as a tautology and double-negation elimination as a rule of inference contributes to the meaning of ‘∼’. If the rule were changed so that ‘∼∼p’ entails ‘∼p’, the meaning of ‘∼’ would change correspondingly. The idea resurfaces in contemporary conceptual role semantics: ‘if some expressions mean what they do by virtue of figuring in certain inferences and sentences, then some inferences and sentences are constitutive of an expression’s meaning what it does, and others aren’t’ (Boghossian 1997: 353). This move also accommodates (3) and (4). Rather than following from logical truths and definitions, they are themselves implicit definitions or explanations of our vocabulary for thermal qualities and colour. Analytic propositions are conceptual. Unlike synthetic propositions, which apply concepts to make statements of fact, they are constitutive of our concepts. They determine the meaning of expressions, in that they lay down what combinations of words are meaningful, and what transformations of empirical sentences are licensed independently of experience. Its linguistic appearance notwithstanding, the typical role of (6) is not to make a true statement of fact about bachelors, but to explain the meaning of ‘bachelor’. This is why we do not verify (6) by investigating the marital status of people we regard as bachelors, and why its denial betokens not factual ignorance but linguistic misunderstanding.3 3. According to Boghossian (1997: 348–56), it is conventional that we express an analytic proposition by assigning the value ‘true’ to a sentence like (2); but the truth so expressed rests on facts about a ‘logical object’ that obtain even if we no longer treat (2) as true. This Platonist option avoids the idea of truth by convention. But I remain sceptical that it can avoid the myth of a museum and preserve the idea that analytic propositions are constitutive

Analyticity, apriority and necessity

On occasion, Quine follows Russell in holding that necessary propositions are factual propositions that describe the most pervasive traits of reality (PL 15; RHS 206). In that case, however, it should be possible to do for necessary propositions what we can do for obvious empirical truths, namely to specify what it would be for them to be false.4 Quine assumes that this can be done, since he denies that there is a substantial difference between propositions which are ‘obviously’ or ‘trivially’ false and the allegedly nonsensical negations of necessary propositions (FLPV 5; WO 228–30; PL 21, 32; Q 191–2). In support of Quine, Harman (1999: 119–23) maintains that even the laws of non-contradiction or identity ‘can be conceived to fail to hold’, since we may refuse to assign a truth-value to sentences like ‘∼(p & ∼p)’ or ‘a = a’. But there is a difference between conceiving that a certain form of words is no longer used and specifying what it would be for a proposition to be both true and false, or for a thing not to be identical with itself, or for white to be darker than black. In our language, there is no coherent description of these cases. This peculiar feature is intelligible if analytic propositions are implicit definitions or explanations of meaning. (6) cannot be overthrown by the putative description ‘This bachelor is married’, because it rules out the latter as nonsensical. (6) is necessary because, in our language, it makes no sense to apply both ‘married’ and ‘bachelor’ to one and the same person. In fact, elsewhere Quine concedes that it is unexplanatory to hold that necessary propositions describe the most pervasive traits of reality, and that, in so far as necessity is intelligible, it resides in language rather than in reality (WP 113, 106–7, 176; PL 95–6 and section 5 below). Consequently, of meaning, since it regards them as statements about and validated by meanings. I therefore favour a Wittgensteinian account. It acknowledges the difference between a sentence and what it expresses, while insisting that sentences like (6) typically express a linguistic rule, a standard for the meaningful use of words. This normativist account avoids both the myth of a museum and the idea of a special truth by convention. We confer a special status on certain sentences by treating them as constitutive of meaning; yet in that case we have not created a factual truth, but adopted a linguistic norm. Normativism is compatible with the fact that we call sentences like (6) true, since the same holds, e.g., of ‘The chess-king moves one square at a time’, which undoubtedly expresses a rule. The crucial point is that being true amounts to something different in such cases, namely to expressing a rule that is actually in force. For this reason, the position is also immune to the charge that rules/conventions can be relevant only to what a sentence says, not to whether what it says is true. Whether a certain rule is in force can be determined by human practice (Glock 2003). Normativism insists that what must be the case, unlike what is the case, is determined not by reality but by our rules. Even if Platonism prevails over normativism, however, it would be cold comfort to Quine, since it will show that the ‘myth’ of a museum is no such thing and that analytic propositions are simply truths about abstract objects. 4. This demand is not met by saying, e.g., that for ‘Black is darker than white’ to be false would be for black to be lighter than white. Required is an informative specification of a possible situation that would falsify the proposition.

85

86

Quine and Davidson on Language, Thought and Reality

he has no knockdown argument against the idea that analytic propositions are conceptual, that is, constitutive of the meaning of our terms.

4

Holism and the a priori

Quine does not just question the intelligibility of the concept of analytical truth, he also straightforwardly denies that there are a priori truths, statements which are unassailable by experience and hence confirmed come what may. According to epistemic holism, when we consider for revision a statement of the form ‘(x)Fx’ in the face of new experiences, we make choices. We may choose to reject one of the statements involved in deducing the prediction, such as ‘(x)Fx → Fa’, or even to discard a refuting observation like ‘∼Fa’ ‘by pleading hallucination’. That there is a choice in dealing with ‘recalcitrant experience’ means two things. First, ‘any statement can be held true come what may, if we make drastic enough adjustments elsewhere’; second, no statement is immune from revision, since it can be abandoned for the sake of upholding others (FLPV 42–3). Pace Mill, mathematics and logic are not directly based on inductive generalization. But they have empirical content because they are indirectly connected to observation sentences in the web of beliefs. Consequently their connection to experience differs from that of scientific theories not qualitatively, but only in degree. This ‘doctrine of gradualism’ (PT 100) implies that there are no a priori propositions in the traditional sense. Being absolutely unfalsifiable is not a special property that some propositions possess. Many allegedly a priori propositions have been relinquished, and others would be, given new empirical evidence. The impossibility of abandoning mathematical or logical propositions is merely psychological or pragmatic. It derives from the fact that revising such centrally located beliefs violates the ‘maxim of minimum mutilation’, according to which the overall system is to be disturbed as little as possible. In fact, even such radical changes have on occasion been proposed, as is shown by constructivist mathematics and intuitionist or quantum logic (PL 7, 86; FLPV 44; Q 142; PT 14–15; PLSP 54; see Gibson 1981: 174; Hookway 1988: 39–41; Harman 1999: 127–8). This fallibilism also seems to rule out the idea that certain propositions could have a special conceptual role, one that antecedes experience and factual truth. However, many of the stock examples of revisable analytic truths look dubious on closer scrutiny. Putnam has alleged that (10) All cats are animals

Analyticity, apriority and necessity

is not analytic, because we would abandon it if it turned out that the things we call ‘cats’ are radio-controlled automata (1975a: ch. 15). But if this means that it could turn out that cats can be controlled through radio signals, it simply shows that some animals could turn out to be automata. On the other hand, if it means that what we call ‘cats’ could turn out to be inorganic robots, it is far from obvious that the envisaged discovery is even epistemically possible, i.e. compatible with what we know. Whether in the face of it we would say that there never were any cats, or that our cats must have been replaced by automata, or that cats need not be animals, is up in the air. Furthermore, if the example is licit, it will actually support rather than challenge the connection between analytic propositions and concepts. Those people who persist in calling (10) analytic under these circumstances are clearly operating with a different concept of a cat from those who do not, namely one which applies only to things with certain properties (including that of being an animal), rather than to those things to which we have previously applied the label ‘cat’ (in the style of a rigid designator). Matters get even hairier when it comes to logic and mathematics. As we saw, there is a difference between no longer using a certain sentence and abandoning a necessary proposition. If one takes this lesson to heart, it becomes difficult to envisage a linguistic practice that abandons the law of non-contradiction. Of course, we can envisage people who always utter sentences of the form ‘p & ∼p’. But on what grounds could we say that they use these sentences to express a contradiction? And if such grounds could be specified, would they not also cast doubt on the idea that these people engage in reasoning or even in meaningful discourse? At any rate, this is suggested by Quine’s own claim that a radical translator should not ascribe illogical beliefs (PL chs. 6 – 7 and chapter 7, section 4 below), and it is explicitly urged by Davidson. But in that case, someone who abandons the law of non-contradiction would not just ‘be left making mutually contrary predictions indiscriminately, thus scoring a pure ratio of success over failures’ (CBG 309); rather, he would not be making predictions in the first place. Although their claims are exaggerated, Quineans can point to cases in which we have abandoned not just a form of words, but certain conceptual connections. However, this recognition is compatible with a dynamic distinction between conceptual and factual propositions, one that allows for changes of meaning and concepts. We can alter or modify our concepts by using the relevant expressions in a certain way, notably by altering the explanations that constitute the meaning of these expressions. In fact, such changes are commonplace in science and everyday discourse. Empirical

87

88

Quine and Davidson on Language, Thought and Reality

propositions are ‘hardened’ into conceptual ones (Wittgenstein 1978: 325), while the latter lose their privileged status and are abandoned. For example, ‘An acid is a substance which, in solution, turns litmus-paper red’ lost its conceptual status – acids now being defined as proton-donors – and is now used to make an empirical statement which holds true of most, but not all acids. Conversely, ‘Gold has 79 protons’ originally expressed an empirical discovery, but is now partly constitutive of what we mean by ‘gold’. The idea that revisability rules out a distinction between conceptual and empirical role amounts to a fallacy. The fact that a Prime Minister can be relegated to an ordinary Member of Parliament does not entail that there is no difference in political status between the Prime Minister and a Member of Parliament. By the same token, the fact that we can deprive certain propositions of their conceptual status does not show that they never had such a status in advance of the conceptual change. Changes of the conceptual framework can themselves be motivated by theoretical considerations ranging from new experiences over simplicity and fruitfulness to sheer beauty. But this does not mean that empirical and conceptual propositions are on a par. For such changes are themselves distinct from theoretical changes like the falsification of a theory (Carnap 1956: 208; Grice and Strawson 1956: 156–7). Strictly speaking, there is no such thing as the falsification of a conceptual proposition. For the necessary status of the latter is constitutive of the meaning of some of its constituent expressions. For example, (11) Nobody under the age of 10 can be an adult

partly determines what we call an adult. If we were to allow a statement like (9) My seven-year-old daughter is an adult

because she has amazing intellectual capacities, let’s say, we would not have falsified (11). For allowing (9) amounts to a new way of using ‘adult’ which modifies the concept of an adult. Consequently (11) and (9) would not contradict each other, since ‘adult’ means something different in each case. This means that, although conceptual propositions can be abandoned, they cannot be falsified in the sense in which empirical propositions can. A difference in status is preserved, since the empirically motivated abandonment of a conceptual proposition differs from admitting the falsity of an empirical proposition. It is conceptual in the sense of involving a change in

Analyticity, apriority and necessity

the meaning of the expressions in question. What is abandoned or revised is not a factual truth about the world, but the definition/explanation of a word. Some sceptics have objected that the distinction between conceptual and theoretical change is just as problematic as the original analytic/ synthetic dichotomy, since there are no clear criteria for distinguishing between conceptual and theoretical change (Waismann 1968: 156–96; Hookway 1988: 37–8). In reply it should be conceded that the distinction is no sharper than that between conceptual and factual propositions. Such sharpness can only be expected if conceptual propositions are simply stipulated, as in Carnap’s artificial languages. When we turn to existing languages (whether ordinary or technical), there will inevitably be borderline cases and fluctuations. But this does not show that the distinction is unclear. Indeed, a clear criterion is in the offing. A revision involves the meaning of an expression e if it involves a change in the explanation of e, in what it makes sense to say with e, and in the logical relations between statements that include e. For example, in Euclidean geometry it makes no sense to say that a light-ray travelled along a straight line and returned to its place of departure. In Riemannian geometry, on the other hand, it makes no sense to deny this. Putnam has directed a more searching set of objections against the claim that every case in which it appeared as if an analytic/conceptual proposition is falsified amounts to a change in the meaning of the terms involved. He maintains first, that such changes cannot be characterized as changes of meaning and, secondly, that if they could, scientific progress would be reduced to triviality. Putnam denies that there is a distinction between conceptual and theoretical change, between cases in which a term is redefined and those in which we discover new facts about the thing denoted by the original concept. He backs this claim through the following example taken from Quine (WP 74–5). In Newtonian physics momentum was defined as ‘mass times velocity’. It soon turned out, moreover, that momentum is conserved in elastic collision. But with the acceptance of Einstein’s Special Theory of Relativity a problem emerged. If momentum was to remain a conserved quantity, it could not be exactly equal to rest mass times velocity. Consequently it was not only possible but also rational for Einstein to revise the statement that momentum is equal to mass times velocity, in spite of the fact that this statement was originally a definition. The view that he thereby altered the meaning of ‘momentum’ is mistaken. For it implies that we are now talking about a different

89

90

Quine and Davidson on Language, Thought and Reality

physical magnitude. ‘But no, we are still talking about the same good old momentum – the magnitude that is conserved in elastic collisions’ (1988: 11; my emphasis; see 1975b: 42–50). However, as this quotation shows, Putnam’s argument trades on the possibility of oscillating between two different definitions of momentum. What we are still talking about is momentum in one of the two senses that the term previously had, namely ‘whatever quantity is preserved in elastic collision’, while giving up the other of ‘mass times velocity’. The plausibility of his story turns on the fact that before Einstein both could equally be regarded as constitutive of the meaning of ‘momentum’. Since the two seemed to coincide invariably, there was no need to decide which one of them should have definitive status and which one should be regarded as empirical. This changed when it was discovered that mass times velocity is not strictly preserved in elastic collision. What Einstein did in reaction to this discovery was to accord definitive status exclusively to ‘preserved in elastic collision’, which amounts to altering the rules for the use of the term ‘momentum’. Scientific revolutions of this kind illustrate that a. scientific concepts and conceptual propositions change; b. scientific concepts are frequently held in place by more than one connection, more than one explanation; c. in certain cases there is no answer to the question ‘Which one is the definition and which one is an empirical statement?’

(c) may seem to support Quine and Putnam, but actually illustrates an important lesson of our dynamic account of conceptual truth. In cases where a phenomenon x (e.g. a disease) is universally associated with several others (e.g. fever, presence of a virus), distinguishing those phenomena which accompany x as a matter of definition from those which accompany x as a matter of empirical fact may be a stipulation. The status of sentences does not just change diachronically. Even when the use of a term is relatively stable, tokens of the same type-sentence can be used either conceptually or factually by one and the same person in different contexts. It may even be indeterminate whether a token-sentence expresses conceptual or empirical relations. For there may not have been a need to decide, as in the case of momentum before Einstein. The possibility of leaving open the precise status of certain statements and connections is perhaps a precondition for scientific progress. Nevertheless, there is a distinction between conceptual and factual connections. For once the question of logical status arises, it is possible to distinguish

Analyticity, apriority and necessity

between those connections which are then adopted as constitutive of meaning (conservation) and those which are abandoned (mass times velocity). And with respect to specific scientific experiments or lines of reasoning, it is often possible to decide whether or not particular sentences are used empirically or as a definition (although there need not always be an answer). Quine considers the case of two physicists. Both postulate a particle under the label ‘neutrino’, but one of them assigns to it a rest mass while the other does not. Quine maintains that it is absurd to ask whether they disagree over the definition of ‘neutrino’ or over whether neutrinos have rest mass (WP 16; SS 70). But at least in some circumstances, the question has a clear answer. If the first physicist postulates a particle with rest mass to explain strong interaction whereas the second postulates a particle without rest mass to explain β-decay, then they disagree over a definition. But if their dispute is whether the particle carrying the surplus energy of β-decay has rest mass, then it is a divergence over facts. That there may be a fluctuation between conceptual and empirical status, and even an indeterminacy, does not obliterate the difference between the two roles. Putnam’s second argument runs as follows. The distinction between conceptual and theoretical change implies that scientific changes which involve abandoning analytic propositions never provide better answers to a given question (‘What is momentum?’) but rather attach old labels (‘momentum’) to new things (the quantity preserved in elastic collision). This not only trivializes scientific revolutions, it presents them as based on a fallacy of equivocation. However, this suspicion is unfounded. As we have seen, conceptual change can be initiated by factual discoveries, in our case that mass times velocity is not preserved in elastic collision. Conceptual propositions determine the network of connections between our concepts. Along with methodological rules, they are part of those fundamental aspects of scientific theories which Kuhn (1970) has called ‘paradigms’. Such paradigms determine the meaning of key scientific expressions. But they do more than simply label things. They inform complex scientific practices in that they provide a way of making predictions and of dealing with recalcitrant experiences. The result of conceptual change in science is not mere renaming, therefore, but a new way of theorizing about the world. This means that changes to our conceptual framework may be far from trivial as concerns both their grounds and their results (see Brown 2001). Although proponents of different paradigms will not simply disagree over facts, they

91

92

Quine and Davidson on Language, Thought and Reality

can have a genuine cognitive dispute, not just about how to explain specific observations, but about the question of what conceptual framework accounts best for the empirical phenomena. For example, the controversy between classical mechanics and general relativity can be seen as a dispute about what geometry is best suited for describing spatial relations on a cosmological scale. 5

The deep need for the convention

It emerges that the distinction between conceptual and factual propositions is not just clear but also compatible with fallibilistic holism, and that it has epistemological significance in discriminating two ways in which our beliefs can change. It remains to tackle Quine’s second reason for denying that such a distinction has explanatory value, which emerges from his critique of the linguistic doctrine of the logical positivists. According to Quine, any explanation of logic by appeal to conventions (rules/definitions), leads to a vicious regress. If we fix the meaning of the logical constants by stipulating that certain inferences are to be valid, we must cover an infinite number of cases. To this end, we must resort to general conventions. Thus one might suggest that the meaning of ‘and’ is determined by the stipulation (12) Let all results of putting a statement for ‘’ and ‘’ in ‘ and  implies ’ be valid.

Quine points out that this procedure involves a ‘self-presupposition’ of basic concepts, since such general conventions already use the ‘logical idiom’ (‘all’, ‘and’, etc.). Consequently this idiom cannot acquire its meaning through such conventions. We could only communicate the conventions that are supposed to underpin necessary truths once these conventions are already in place (WP 103–4). Quine recognizes that there is a possible reaction to this difficulty. Why not say that we first observe conventions in our behaviour, without announcing them in words, and formulate them only subsequently? He even admits that this ‘accords well with what we actually do’ in formulating logical truths. But he remonstrates: ‘In dropping the attributes of deliberateness and explicitness from the notion of linguistic convention we risk depriving the latter of any explanatory force and reducing it to an idle label’ (WP 105–6). For we cannot distinguish behaviour which involves such implicit conventions from behaviour which involves no conventions at

Analyticity, apriority and necessity

all. From Quine’s behaviourist perspective, there is merely a quantitative difference between more or less firmly accepted statements. The fundamental bone of contention is therefore this: is it possible to distinguish between well-entrenched beliefs on the one hand, linguistic conventions or rules in this implicit sense on the other? Once more, we confront Quine’s inextricability thesis that factual belief and convention cannot be distinguished in linguistic behaviour. But once more we can resort to the distinction between disbelief and incomprehension. Even a linguistic community in which norms like (11) or (12) have not been explicitly enunciated can react with incredulity to utterances like (8), and with incomprehension to (9) or failure to acknowledge a valid inference. In fact, one can argue that some distinction, however implicit, between experience and convention is not just feasible but indispensable to meaningful language. Thus Wittgenstein wrote of ‘the deep need for the convention’, and maintained that the recognition of conceptual connections is a precondition of describing factual connections (1978: 65; 1975: 66). A possible rationale for this claim is that the assimilation of the conceptual to the factual sits uneasily with Quine’s own holistic picture of a web of beliefs (Quinton 1973: 216–17; Dummett 1978: 376–8; Wright 1980: 329–30). Although Quine recognizes that logical truths provide the connections within the web of beliefs (WO 12–13), he also holds that the germaneness of one belief to another is only a loose association, which reflects the relative likelihood of revising beliefs. Logical and mathematical statements at the centre of the web are ‘simply further statements of the system, certain further elements of the field’ (FLPV 42–3). But unless certain relations had a special logical status, there would be no web of beliefs adapting to experience. For observations at the periphery would not be logically linked to theories closer to the centre, and hence could not confirm or refute them. It would be unclear whether the general statement ‘(x)Fx ’ is incompatible with an observational statement like ‘∼Fa’, or one like ‘Fa’, or like ‘∼Ga’. Consequently, there could be no procedure for deciding what changes should be made in the face of recalcitrant experiences. What of a radical empiricist and behaviourist who is prepared to abandon even the holistic picture of a web of beliefs, someone willing to dispute, for example, Davidson’s thesis that any belief must always stand in logical relations to other beliefs (see chapter 9, section 3)? Here we must ask: what would a form of linguistic behaviour look like which lacked any distinct logical and conceptual structure? Let us consider first the case of a single term like ‘bachelor’. If all of the conceptual connections set up

93

94

Quine and Davidson on Language, Thought and Reality

by linguistic rules were transformed into empirical ones this would mean that all of the following statements could be rejected. (13) Bachelors are unmarried men (14) Bachelors are human beings (15) Bachelors are made of flesh and blood.

Under these circumstances anything at all could be called ‘bachelor’, since there would be no reason to deny that anything can fall under the concept. Consequently the use of this term would have become totally arbitrary. But this would simply mean that the expression had lost all meaning. It would have become senseless. Correspondingly, if we surrendered the norms governing the use of all our words, these words would lose all meaning. Of course our habit of uttering words might continue: a communal phonetic babbling without rules is a conceivable form of behaviour. However, it would not be a case of meaningful discourse. If anything can be said, nothing can be meaningfully said (Glock 1996b: 220–2). Both criticisms have been repudiated on the following grounds: to weave empirical beliefs into a web and to ensure a restricted use for our expressions, logical propositions and propositions like (13)–(15) need not be treated as necessary but only as true (Gluer ¨ 1999: 233–4). But this riposte creates a regress similar to that created by failure to distinguish between axioms and rules of inference in a logical system (Carroll 1895). If all propositions have the same status, namely of being contingently true, then treating any two propositions as standing in a logical connection will be contingent on further contingent statements. One cannot simply infer ‘Fa’ from ‘(x)Fx’, because this relation will depend on another statement, namely ‘(x)Fx → Fa’. But if this is once more a mere truth rather than a rule that entitles one to pass to the conclusion, then one is not entitled to infer ‘Fa’, but only the proposition ‘((x)Fx & ((x)Fx → Fa)) → Fa’, and so on. What is needed are relations between propositions which are not just external but internal, that is, constitutive of the two relata and hence not in need of underwriting by a third proposition (Glock 1996c: 189–91). ‘(x)Fx → Fa’ is a tautology, and this means that it is constitutive of the two propositions ‘(x)Fx’ and ‘Fa’ that the latter can be inferred from the former, without appeal to an additional proposition. Similarly, as a definition that constitutes meaning, (13) establishes an internal relation between the concept of a bachelor and that of being unmarried. It licenses the step from ‘Kant was a bachelor’ to ‘Kant was unmarried’ without appeal

Analyticity, apriority and necessity

to the (logical) truth ‘((x)(Fx →Gx) & Fa) →Ga’. But it could not provide such a direct licence if it were merely a further element of the web of belief. By a similar token, if statements like (13)–(15) were simply more firmly entrenched than others, the connection between applying ‘bachelor’ and withholding ‘married’ would be one of mere regularity, though rarely severed. In response to utterances like ‘I just met a married bachelor’ or ‘π is a bachelor’ I might be surprised, but I could not reject it as unintelligible or demand an explanation. But a predicate like ‘bachelor’ is meaningful only in so far as its application is tied to that of some predicates and incompatible with that of others. Without such special connections, we could employ words any way we please, and the difference between meaningful and nonsensical uses would vanish. Under such circumstances, linguistic utterances could lead one to form expectations concerning the future behaviour of the utterer, and one could use words with the intention to cause a certain result. But linguistic utterances would merely be empirical indicators of other phenomena, just as clouds indicate rain. They would have indicative value (natural meaning), but no genuinely linguistic meaning. My argument is intended to show not that we have to retain a certain number of beliefs – a point Quine accepts – but that some connections must be internal rather than external, and some propositions conceptual rather than factual. They must have the special status of licensing the transformation of factual propositions and of excluding certain combinations of words as nonsensical. Quine’s rejection of (implicit) conventions cannot make sense of language or meaning. This is not to deny that linguistic behaviour can be causally explained as a natural phenomenon. But in order to understand it as meaningful we have to regard it as subject to normative distinctions which he abhors. 6

Modal logic and essentialism

Quine follows the logical positivists in identifying the necessary, the analytic and the a priori. This is at odds not just with Kant, but also with contemporary essentialism. According to Kripke (1980: 34–9), the a priori is an epistemological category, necessity a metaphysical one, and analyticity a logical one. A truth is a priori iff it can be known independently of experience; it is necessary iff it is true in all possible worlds; it is analytic iff it is true by virtue of meaning. Kripke maintains that these categories differ

95

96

Quine and Davidson on Language, Thought and Reality

not just in their intension, but also in their extension; in particular, there are necessary a posteriori truths. Quine, by contrast, repudiates virtually all forms of necessity: ‘I see no higher or more austere necessity than natural necessity; and in natural necessity, or our attributions of it, I see only Hume’s regularities’ (WP 76, see 68–74; TT 120–1, 141; RHS 396–8; Q 139–42; CBG 244). The only modal notion he accepts is epistemic necessity relativized to context. When we say ‘Necessarily, p’, we indicate that ‘p’ follows from something that is taken for granted on this occasion. Quine’s rejection of metaphysical necessity is based on his well-known animadversions to quantified modal logic. His scepticism concerns not the formal feasibility of this enterprise, but its philosophical consequences. With characteristic foresight, Quine anticipated the essentialist implications of quantified modal logic by more than a decade. Ironically, instead of undermining modal logic, his warnings have led to a revival of essentialist metaphysics. What is more, this new essentialism can appeal to Quine’s own naturalism. Quine thinks that philosophy must eschew necessity and essences because it is continuous with science. But if some necessary truths, truths about the essence of things, are a posteriori, then philosophy is continuous with science precisely because it deals in such essences. Against conventionalism, Quine and essentialism agree that philosophy cannot be a matter of conceptual analysis. I shall argue that they are wrong on this. Quine’s attacks against modal logic are uncompelling as they stand. But he is right on the issue which unites him with conventionalism. There is no such thing as ‘metaphysical necessity’ (TT 174). The kind of modality sought by metaphysics ‘resides in the way we say things, and not in the things we talk about’ (WP 176). Quine distinguishes three ways of formalizing modal notions (WP ch. 15): I. as a predicate that applies to quoted sentences, as in ‘
“p”’ (‘ “p” is necessarily true’); II. as a sentence-forming operator that applies to closed sentences, as in ‘
p’ (‘it is necessarily true that p’); III. as an operator that extends to open sentences as well, as in ‘
Fa’ (‘a is necessarily F’) which entails ‘∃x
Fx’.

Quine regards (I) as harmless. If ‘analytic’ could be treated as an extensional metalinguistic predicate, ‘
“p”’ could be rendered in canonical notation as ‘“p” is analytic’. But even if it cannot, we can understand this locution simply as saying that ‘p’ is a truth of logic or mathematics.

Analyticity, apriority and necessity

(II) poses additional problems, in that it invites iterated modalities like ‘
p’ which Quine regards with suspicion, but in principle it can be reconstructed in similar metalinguistic terms. We need a separate modal logic only in areas where such metalinguistic paraphrase is unavailable. This holds for (III), quantified modal logic. Here it is claimed that some object has the property of being necessarily F. No complete sentence follows the modal operator, and hence the alleged necessity cannot be explained in terms of the analyticity of a sentence. According to Quine, however, quantifying into modal contexts is no more legitimate than quantifying into quotation. This verdict follows from two premises. First, quantification is to be understood objectually; second, modal contexts are opaque, the occurrence of singular terms within such contexts is not ‘purely referential’ (WP 160–4; FLPV 140–8; WO §41). Since the variables of objectual quantification can only take the place of purely referential terms, it follows that one cannot quantify into modal contexts. One way of resisting this argument is to repudiate the first premise and to construe quantified modal logic in substitutional terms. As we have seen, however, Quine has noteworthy qualms about substitutional quantification. What about the second premise? Quine supports it with the following argument. If modal contexts were transparent, the following argument would have to be sound: (P1 ) The number of planets = 9 (P2 ) 9 is necessarily greater than 7 (C1 ) The number of planets is necessarily greater than 7

But while the premises are true, the conclusion is false, since there might have been fewer than 7 planets. Accordingly, substitutability fails and ‘necessarily’ is opaque. This verdict is precipitate. We must distinguish between a de dicto interpretation of modal operators, according to which they concern whole propositions, and a de re interpretation, according to which they concern an individual thing. (C1 ) can be read de dicto, namely as (C1
) Necessarily: the number of planets is greater than 7

or de re, namely as (C1 ∗ ) Concerning the number of planets: it is necessarily greater than 7.

(C1
) is false, but it does not follow from the premises, since the definite description ‘the number of planets’ occurs within the opaque context

97

98

Quine and Davidson on Language, Thought and Reality

produced by ‘necessarily’. By contrast, (C1 ∗ ) does follow, but it is true, since it has the force of ‘The number of planets, viz. 9, is necessarily greater than 7’ (see Smullyan 1971). However, Quine also argues that the idea of de re necessity makes no sense to begin with. (C1 ∗ ) ascribes to the number 9 a special property, namely that of being essentially greater than 7. But the ‘doctrine that some of the attributes of a thing (quite independently of the language in which the thing is referred to, if at all) may be essential to the thing, and others accidental . . . leads us back into the metaphysical jungle of Aristotelian essentialism’ (WP 175–6). Quine tries to ‘evoke the appropriate sense of bewilderment’ by considering an individual, let’s call him John, ‘who counts among his eccentricities both mathematics and cycling’ (WO 199). According to essentialism, John would have to be necessarily rational and contingently two-legged as a mathematician, yet necessarily two-legged and contingently rational as a cyclist. But this leads to inconsistency: (16) (x)(x is a mathematician →
(x is rational)) & (x)(x is a mathematician → ∼
(x is two-legged)) (17) (x)(x is a cyclist →
(x is two-legged)) & (x)(x is a cyclist → ∼
(x is rational)) (18) John is a mathematician and John is a cyclist

From (16) and (18) it follows that John is necessarily rational, while from (17) and (18) it follows that he is not necessarily rational. Consequently, there is no principled way of assigning some properties as essential to a thing and others as accidental (FLPV 148, 155–6). What properties a thing possesses essentially depends on how it is described (e.g. as a mathematician or a cyclist). So-called essential properties are merely properties entailed by some currently salient description. However, while the essentialist is committed to some properties being essential to a thing, e.g. being greater than 7 to the number 9, she is not committed to the claim that John is essentially two-legged, since there are possible worlds in which he lost one leg as a child. Therefore she can avoid de re statements like (16) and (17), and opt for uncontroversial de dicto statements which do not lead to inconsistency in conjunction with (18) (see Marcus 1990: 237–8; Orenstein 2002: 158–9): (16
)
(x) (x is a mathematician → x is rational) & ∼
(x) (x is a mathematician → x is two-legged) (17
)
(x) (x is a cyclist → x is two-legged) & ∼
(x) (x is a cyclist → x is rational)

Analyticity, apriority and necessity

At this juncture, Quine is wont to complain that the idea of there being any essential properties is inimical to modern, non-Aristotelian science (TT 121, 128). For Quine, natural kinds are distinctions in our ‘subjective quality space’, that is, determined by the nature of human observers (OR ch. 5). By contrast, essentialism presupposes a taxonomic realism according to which the observed world itself divides unequivocally into different types. This taxonomic realism is incompatible with Quine’s instrumentalist conception of science. What is more, it may be at odds with scientific practice. Thus it has been argued that the essentialist understanding of biological species as determined by microstructure runs counter to modern biology (Dupr ´e 1993). This response leaves open the possibility of an essentialist metaphysics independent of science. Quine’s second complaint does not, since it questions the intelligibility of the essentialist framework. De re modality and essentialism are intimately connected to the idea of possible worlds. An essential property of x is one which x has in every possible world in which it exists. Quine has pointed out that it would follow that all identities are necessary (WP 176). But this is a consequence essentialists like Kripke are happy to defend (1980: 3–5, 97–110). However, the idea of a possible world is problematic. It is fundamentally unclear what possible worlds are, over and above a picturesque way of illustrating counterfactuals (Haack 1978: 190–3). Quine’s main objection to possible worlds, however, is the lack of criteria for transworld identity. There is no way of determining whether the possible fat man in that doorway is identical with the possible bald man in that doorway. And it makes sense to say that there is a possible world in which John lost a leg as a child only if we can identify someone in that world as John, just as we can identify John from moment to moment. But nothing on the modal side matches our well-developed criteria of identity over time. Unlike temporal stages, possible worlds do not have a unique ordering imposed on them (FLPV 4; TT 127, 173; see Hookway 1988: 118–19). Some essentialists concede the impossibility of providing criteria of transworld identity, but insist that it does not matter. After all, they suggest, it makes sense to say that John once was a precocious child even though we may be incapable of picking out that child (Plantinga 1974: ch. 6). But while we may not need operational procedures of identification, we do need to know what identity across possible worlds consists in. In the temporal case, we can say that the precocious baby is the one that later discovered a particular theorem. No equivalent explanation is in sight for the modal case. Again, it has been maintained that

99

100

Quine and Davidson on Language, Thought and Reality

de re modal talk does not depend on transworld criteria of identity, but the other way around: to call an object in a different world identical to x is just to say that its properties are properties x could have possessed (Yablo 1998b). However, a de re possibility is something that could hold of a particular individual; therefore one cannot without circularity explain the individuality and transworld identity of that thing by reference to de re modalities. According to Kripke, proper names and natural kind terms are rigid designators, that is to say, as long as they do not change their meaning, they refer to the same thing in all possible worlds in which their referents exist (1980: 48–9). Furthermore, biological organisms derive essentially from whatever origins they in fact derive from. The property that individuates a person is to have derived from a particular sperm and egg (1980: 113–14; Hookway 1988: 121–2; Forbes 1985: ch. 6). Essentiality of origin would solve the problem of transworld identity. However, the doctrine is problematic, and not just from the perspective of Quine’s elimination of singular terms. Even if one denies that proper names are equivalent to descriptions of their bearers, it is far from clear that they are completely independent of descriptions. Consider the possibility that Wittgenstein’s gametes developed into a person indistinguishable from Hitler, while Hitler’s gametes developed into a person indistinguishable from Wittgenstein. Who is to say that there is no choice but to conclude that the former is Wittgenstein (Grayling 1997: 74–5)? It is up to us whether a term is used rigidly, or whether its application is tied to the satisfaction of certain descriptions. This takes us to the crucial strategic question. According to Kripke, identity statements like (19) Hesperus is Phosphorus

and theoretical identifications like (20) Water is H2 O

are a posteriori yet necessary, because they involve rigid designators that pick out the same things in all possible worlds: the planet Venus in the former case, a substance with a particular microstructure in the latter. This classification is widely accepted. However, it has been contested by Quine and Davidson (SIO 29, 198; see also Stroll 1989: 271–2; Sidelle 1989: 105–6; Dupr ´e 1993: 25–6). Furthermore, even if one concedes that (19) and (20) are necessary a posteriori this will not commit one to accepting metaphysical necessities. As

Analyticity, apriority and necessity

Alan Sidelle has argued, one can see such statements as arising from two components: one which is necessary but conventional, and one which is empirical but contingent (1989: chs. 2–3; similarly Peacocke 1997: 244). The grounds of the necessity of (19) lie in a linguistic norm, namely that expressions like ‘Hesperus’ are to be used rigidly. Once we have determined empirically that Hesperus is Phosphorus, this convention ensures that the identity holds across all possible worlds. Similarly, the necessary a posteriori status of (20) can be seen as the result of combining a contingent discovery, namely that H2 O is the chemical composition of the stuff denoted by ‘water’, with a convention: whatever chemical composition paradigmatic instances to which we apply ‘water’ actually have, in any possible situation, nothing will count as water unless it has that composition. (19) and (20) may be de re in the sense that they are about (rigidly designated) subjects. But it does not follow that the ultimate source of their necessity is an independent modal reality. For there are no facts about who is who or what is what in possible worlds independently of our conventions of individuation. If this line of reasoning can be sustained, quantified modal logic and de re modalities do not rule out that the source of all non-physical necessity lies in our conceptual scheme. By the same token, they do not rule out that ‘metaphysical’ inquiries into the essence of natural kinds are investigations of complex conceptual connections and their interplay with contingent facts. I have also argued that these ideas do not fall prey to Quine’s attacks. The semantical result of my discussion, therefore, is that one must draw a distinction between necessary and contingent propositions, between the conceptual framework and its empirical employment. The metaphilosophical result is that one can maintain, contrary to both Quine and essentialism, that philosophy is a priori because its problems concern concepts rather than empirical facts. Even if I am right against both alternatives, this will not establish that a priori conceptual analysis is a legitimate philosophical enterprise. But it will eliminate the most powerful reasons for supposing otherwise.

101

4

Truth

Quine and Davidson base their accounts of truth on Tarski (section 1). But although they treat Tarski’s semantic theory as the indispensable starting point, they also stress its limitations. Section 2 deals with Davidson’s reasons for denying that Tarski has said all there is to say about truth, and his claim that truth is indefinable. I shall resist the latter claim, and argue that Tarski’s theory leaves even larger gaps than Davidson recognizes. In section 3, I maintain that all three go wrong in treating truth as a property of sentences and thereby relativizing it to languages. But while I condemn this sententialism, I defend Davidson’s pragmatist claim that the concept of truth indirectly presupposes the linguistic activities of human beings. Next I turn to three prominent accounts of truth that Davidson regards not just as incomplete but as inadequate, namely antirealism and correspondence theories on the one hand, deflationism on the other. I provide a qualified defence of Quine’s and Davidson’s attacks on correspondence theories and epistemic accounts (section 4). Section 5 concludes that the proper moral is to adopt a version of deflationism, though not the one implicit in Quine’s disquotational theory. 1

The legacy of Tarski

According to Aristotle, ‘to say of what is that it is, and of what is not that it is not, is true’. Tarski regarded his theory as a defence of ‘the classical Aristotelian conception of truth’, which he transposed into modern terminology as the view that ‘the truth of a sentence consists in its agreement with (or correspondence to) reality’. But he considered both Aristotle’s dictum and this paraphrase to be potentially misleading, and hence sought a ‘more precise expression’ of the classical conception. Tarski

[102]

Truth

describes his theory as semantic, and semantics as that part of logic which is concerned with the relations between linguistic expressions and the objects or states of affairs ‘referred to’ or ‘expressed by’ them (1949: 54–6; 1969: 63–4). Unlike Davidson, however, Tarski does not discuss truth out of an interest in its connection with meaning. Instead, he seeks a systematic account of what determines the extension of ‘true’ in a formal language. He regarded such an account as a definition of ‘true’ for that language, and insisted that it must meet two conditions. First, it must be ‘materially adequate’: ‘The desired definition does not aim to specify the meaning of a familiar word used to denote a novel notion; on the contrary, it aims to catch hold of the actual meaning of an old notion’, namely the classical Aristotelian one. Second, the definition must be ‘formally correct’, which means in particular that it must avoid semantic paradoxes like that of the liar (1949: 52–3; 1983: 152–3). In developing a criterion for material adequacy, Tarski starts out from the idea that the conditions for a particular sentence’s being true can be stated as follows: (1) ‘Snow is white’ is true iff snow is white

Following Davidson, sentences like (1) are known as ‘T-sentences’ (ITI 64). According to Tarski, (1) provides a ‘satisfactory explanation’ or ‘definition’ of the term ‘true’ as applied to the sentence ‘snow is white’ (1969: 64). This definition is not circular, because what occurs on the left-hand side is not the sentence itself, but the name of the sentence. Such a name can be formed either by putting the sentence in quotation marks, or through a ‘letter-by-letter description’ of the sentence (1949: 55; 1969: 64), what Davidson calls a ‘structural description’ (ITI 18n) and Quine a ‘spelling’ (WO 143; ML 23–6; PT 69–70). By providing a definition of ‘true’ in its application to ‘snow is white’, (1) also provides a ‘partial definition’ of the term ‘true sentence in English’. Analogous partial definitions can be given for any other English sentence. If the number of English sentences were finite, one could give a complete definition of that predicate through the logical conjunction of all these partial definitions. However, that number is potentially infinite, and hence we would require an infinite number of partial definitions (1949: 55; 1969: 65). Instead of a complete definition, therefore, what we have reached is a form to which each partial definition has to conform, namely (T) s is true iff p

103

104

Quine and Davidson on Language, Thought and Reality

where ‘p’ can be replaced by any English sentence in the indicative, and ‘s’ by a name of that same sentence. (T) is known as Tarski’s T-schema. It is not a complete definition, because it is a schema rather than a generalization over sentences. But it furnishes the desired criterion for the material adequacy of such a definition, known as ‘Convention T’ (1983: 187–8; 1949: 55): a definition of ‘true’ is materially adequate if it entails all equivalences of form (T). The requirement of formal correctness comes into play because not all instances of (T) are acceptable. Semantic notions like truth can give rise to paradoxes. A sentence like (2) The sentence on line 11 of page 104 of this book is false

will yield a contradiction when plugged into (T). According to Tarski, the fault lies with the fact that natural languages like English are ‘semantically closed’. They do not just contain expressions to talk about non-linguistic objects, but also names for expressions of the language and semantic terms like ‘true’. As a result, they permit the construction of self-referential sentences like (2). Tarski averts self-referential sentences and thereby the threat of paradox by confining the definition of truth to languages which are not semantically closed (1983: 157–64; 1949: 58–61). In such cases we can distinguish strictly between the ‘object language’ for which the semantic theory defines ‘true’, and the ‘metalanguage’ of the theory itself, which is used to talk about the object language. Truth turns into a metalinguistic property of the sentences of open languages. For any such object language L, Tarski defines a predicate ‘true’, but that predicate is itself part of the metalanguage M, not of L. Moreover, the definition is relative to a particular language L, which means that he defines not ‘true’ simpliciter but ‘true-in-L’. To satisfy Convention T, Tarski’s definition must entail all equivalences of the form (T), with ‘s’ being a name in M of a sentence of L, and ‘p’ being a translation of that sentence in M. This means that M must be ‘essentially richer’ than L. It must contain all the sentences of L, or translations of them, as well as devices for constructing a name for every sentence of L and semantic notions like ‘true-in-L’. It also means that M and L must be ‘formalized languages’ with an ‘exactly specified structure’. M contains a syntax capable of specifying all the sentences of L, since these are the items to which ‘true-in-L’ applies. Because L must be semantically open and formalized, Tarski regarded it as ‘very questionable’ that a definition of truth could be given for a natural language. For such languages

Truth

contain semantic notions and are not formally specifiable (1949: 57–62; 1983: 164–5). Davidson has been more optimistic on this score, to which we shall return in the sequel. If M includes L as a proper part, the sentences of L will be their own translations into M. In that case, the results are disquotational T-sentences like (1). (1) is an instance of a disquotational version of the (T) schema: (T
) ‘p’ is true-in-L iff p.

It is tempting to suppose that a general definition of ‘true-in-L’ can be achieved through universal quantification over (T
): (D1 ) (p) (‘p’ is true-in-L iff p).

Tarski, Quine and Davidson reject this quick fix because, for diverse reasons, they regard quantification into quotation marks as illicit (1983: 158–60; FLPV 140; WO 144; PL 12–13; ITI ch. 6; FDT 273). Tarski’s alternative is to achieve generality through axiomatization rather than quantification. He constructs a set of axioms which suffice to entail a sentence of form (T) for any sentence of L. This axiomatization specifies both the syntax and the semantics of L. The syntax determines what counts as a sentence in L. It does so recursively, by first providing a list of simple sentences and then stating what operations can be performed on them to construct complex sentences. The semantics of L is equally recursive. It specifies how the truth-values of complex sentences depend on those of their constituents. Unfortunately, this procedure does not suffice. Only sentences can be true or false. But even in a simple formal language like Quine’s canonical notation, sentences are constructed out of predicates or ‘open sentences’ like ‘Fx’, which are neither true nor false. To solve this problem, Tarski turns to a semantic notion – satisfaction – which applies to all sentences, open as well as closed, and then defines truth in terms of satisfaction. Satisfaction is a relation that holds between predicates and sequences of objects. Sequences are ordered n-tuples. They are like sets, except that the order of the elements matters. For example, ‘x is south of y’ is satisfied by but not by <Manchester, Reading>. Predicates have any number of free variables. This would mean that there is an infinite number of satisfaction relations, one between one-place predicates and single objects, one between two-place predicates and ordered pairs, and so on. To preserve a single relation, Tarski defines satisfaction as a relation between predicates and infinite sequences. A predicate ‘F(x1 . . . xn )’ is satisfied by the sequence

105

106

Quine and Davidson on Language, Thought and Reality

iff it is satisfied by the first n members of the sequence; subsequent members are immaterial. Given the notion of satisfaction and the syntax of L, truth can be defined recursively in three steps (PL ch. 3). First, ‘satisfied in L’ is defined for simple predicates by specifying for each of them the conditions under which it is satisfied by a sequence. Next, ‘satisfied in L’ is defined recursively for complex sentences (open or closed) by displaying how their satisfaction depends on that of simple sentences. Sentences are limiting cases of predicates, namely those without free variables. Only the i-th member of a sequence is relevant to whether the sequence satisfies a one-place predicate ‘Fxi ’. By the same token, no member is relevant to whether the sequence satisfies a zero-place predicate or sentence. Therefore, a sentence is satisfied by any sequence iff it is satisfied by all. This furnishes the final step, an explicit definition of truth: (D2 ) A closed sentence of L is true iff it is satisfied by all sequences.

Tarski’s definition is formally correct – notably in blocking the semantic paradoxes – and materially adequate: it defines truth in L in the sense that its axioms generate a T-sentence for each of its sentences. Both Quine and Davidson base their account of truth on Tarski (RLS 353; SCT 281). Quine extracts three main points. First, by defusing the semantic paradoxes, Tarski put one part of semantics, the theory of reference concerned with extensional notions like denotation and truth, on the secure path of a science, leaving behind the other part, the theory of meaning, in its intensional quagmire (FLPV 137–8; SS 62–5). Secondly, one of the reasons why truth is a clearer concept than meaning is that it is disquotational. The crucial lesson of Tarski’s theory is that ‘the truth-predicate is a device of disquotation’ (PL 12; see Q 213; PT 79–83). Thus in (1) the quotation-marks around the sentence ‘snow is white’ take us from talk about the world – in this case about snow – to talk about language; ‘is true’ cancels this semantic ascent and makes us once more snow-bound. Finally, Tarski is right to reject the ‘disappearance theory of truth’, the idea that truth is logically redundant or eliminable (PL 11; Q 214). Disquotation disposes of the truth-predicate only in those cases in which the object language is a proper part of the metalanguage. More is involved in (3) ‘Schnee ist weiß’ is true iff snow is white

namely the translation from the object- into the metalanguage.

Truth

Even that does not suffice for cases in which the sentences are not explicitly before us, such as ‘Whatever the Pope says is true’ or ‘All consequences of true sentences are true’. Both Quine and Davidson reject the attempt to paraphrase such sentences through quantification over sentences or propositions. Instead, they turn to Tarski, who provides a way of predicating truth even of sentences we can designate but do not know how to formulate. On the basis of a recursive definition of satisfaction for the language in question, Tarski can explicitly define a truth-predicate, and thereby eliminate it even in cases in which no sentence is to hand (ITI 38–41; SCT 284–5; FDT 273; PCT 8). Davidson employs the semantic theory in a way that is novel in two respects. First, he defies Tarski’s scepticism by applying it to natural languages. Secondly, he stands the semantic theory on its head. Tarski provided a recursive theory of truth for L which took for granted the notion of translation, namely of ‘s’ from L into M, and thereby the notion of meaning. By contrast, Davidson tries to provide a theory of meaning for L, which specifies the meaning of ‘s’ while taking for granted the notion of truth (ITI xiv, 134, 150, 172–3; SCT 286n). The basis for this reorientation is the idea that the meaning of a sentence is determined by the conditions under which it is true. Davidson seeks not a semantic theory of truth but an alethic theory of meaning. Moreover, in this theory T-sentences are not consequences of the stipulated rules of an artificial language, but empirical claims about an existing natural language. Yet in spite of his ultimate concern with meaning Davidson has increasingly discussed truth in its own right (IA 65). In fact, one can distinguish five different phases in his reflections. A Definition of Truth: In ‘Truth and Meaning’ (1967), Davidson thought that Tarski could provide both a complete definition of truth and an empirical theory specifying the truth-conditions and thereby the meaning of the sentences of natural languages. Later he abandoned the first ambition and came to employ the semantic theory exclusively as a way of specifying the semantic properties of natural languages (ITI xiv–xv; SCT 286+n). A Correspondence Theory: In ‘True to the Facts’ (1969), Davidson portrayed the semantic theory as a version of the correspondence theory, albeit one that evolves not around a correspondence between sentences and facts, but around satisfaction as a relation between words and objects (ITI 37). Later, he gradually abandoned that idea, and with it the label ‘correspondence theory’.

107

108

Quine and Davidson on Language, Thought and Reality

A Coherence Theory: ‘A Coherence Theory of Truth and Knowledge’ (1981) officially espouses a coherence theory, on the grounds that a belief can only be justified by another belief from a web of beliefs. But since Davidson holds that a coherent web of beliefs must be largely true, this coherence theory of justification is supposed to yield ‘correspondence without confrontation’ (of sentences with facts), and thereby realism concerning truth and the external world (SIO 137–8). A Pragmatist Approach to Truth: ‘A Coherence Theory of Truth and Knowledge’ provoked a debate with Rorty (1986). In its course, Davidson agreed that the title of the essay is misleading and that his ‘view of truth amounts to a rejection of both coherence and correspondence theories and should properly be classed as belonging to the pragmatist tradition’ (SIO 154; see SCT 302–3; IA 58). Note, however, that it belongs to that tradition not on the ground advocated by Rorty, namely that there is nothing philosophically interesting to say about truth, but on the incompatible ground that there is something interesting to say, namely how truth is connected with the thoughts and activities of human beings (SCT 279–82; FDT 265, 276–8). Truth as Indefinable: Davidson’s latest turn is to deny that Tarski has provided a full explanation of truth and to declare that truth is indefinable: ‘Truth is one of the clearest and most basic concepts we have, so it is fruitless to dream of eliminating it in favour of something simpler or more fundamental’ (SCT 314, see 297n; FDT 263–4, 275–6).

In spite of these largely terminological shifts (see ITI xiv–xv; SCT 286n, 302–4; RD 36–7), several ideas remain constant throughout. First, from the very beginning Davidson has rejected the traditional correspondence theory – ‘the strategy of facts’ – according to which truth is a relation of correspondence between thoughts or sentences on the one hand and facts, items which make them true on the other (ITI 41–3, 50). Secondly, this has not led him to adopt ‘antirealist’ or ‘epistemic’ accounts which reduce truth to what we are justified in believing or capable of learning. Thirdly, Davidson has been adamant in rejecting deflationary theories according to which there is nothing more to truth than what is captured by schemas like (T). Truth is not redundant; it is a substantive property, and there are important connections between truth and other concepts. Finally, Davidson has always found truth to be unproblematic, that is, ‘beautifully transparent’ and ‘primitive’ compared to concepts such as belief (SIO 139). This

Truth

is why he takes the notion for granted, and uses it to explicate meaning: ‘. . . meaning not only is a more obscure concept than that of truth; it clearly involves it: if you know what an utterance means, you know its truth conditions’ (FDT 278). All these convictions are at work in Davidson’s most recent pronouncements. On the one hand, he rejects deflationism. Tarski did not say ‘all there is to say about truth’, because he did not explain what the truth-predicates of different languages have in common (FDT 268; see SCT 288). Since Davidson regards any departure from Tarski’s paradigm as ‘crippling’ to semantics, for him the question is how the semantic theory should be supplemented (SCT 294–5). On the other hand, he repudiates two received ways of adding to Tarski, namely epistemic and correspondence theories. Davidson’s own supplementation evolves around two claims. First, there is an intuitive notion of truth which applies across languages, but it is both indefinable and unproblematic. Secondly, Tarski’s account should be enriched by linking the concept of truth to that of meaning and thereby to propositional attitudes and human activities. The connection is forged by a theory of meaning for natural languages which turns Tarski’s T-sentences into empirical claims about the truth-conditions (alias meanings) of the sentences of a particular natural language. It employs a general notion of truth, because these T-sentences apply the predicate to sentences of a different language (SCT 294–5, 309–26; RD 36; FDT 275–8). Davidson’s Aufhebung of the concept of truth in a theory of meaning has led commentators to pass over his remarks on truth as a mere prelude to his theory of meaning, or even to opine that he has no theory of truth, only a theory of meaning (Kirkham 1992: 38). Initially, Davidson played down the question of whether the semantic theory elucidates ‘the’ or even ‘a’ philosophically important conception of truth, or our ordinary use of ‘true’ and its cognates (ITI 24). And in recent writings, he often characterizes a theory of truth as a Davidsonian theory of meaning, ‘an empirical theory about the truth-conditions of every sentence in some corpus of sentences’ (SCT 309). To Davidson’s followers, such passages present an invitation to bypass the topic of truth. To less sympathetic readers, they amount to a refusal to face the general philosophical problem altogether. Having agreed to the general point that the meanings of sentences are determined . . . by rules which determine truth-conditions, we then raise the general question . . . what truth-conditions are conditions of;

109

110

Quine and Davidson on Language, Thought and Reality

and we are told that the concept of truth for a given language is defined by the rules which determine the truth-conditions for sentences of that language. (Strawson 1971: 180; see Dummett 1978: ch. 1)

This circularity frustrates Davidson’s original plan of explaining both meaning and truth in one fell swoop by reference to a Tarskian theory. As we have seen, Davidson subsequently abandoned this ambition and settled for a theory of meaning. But one can use a Tarskian theory to specify the meanings of sentences by stating their truth-conditions only if that theory does justice to the notion of truth. Otherwise, how could we assess whether Davidson is justified either in regarding truth as the key to meaning, or in claiming that the proper way to shed light on truth is to look at its role in a theory of meaning? The aforementioned passages notwithstanding, Davidson does not simply speak of truth when in fact he is discussing meaning. He initially defined a theory of truth along Tarskian lines, namely as a system which is materially adequate to the notion of truth by virtue of satisfying Convention T. That such a theory provides a theory of meaning is put forward as a substantial discovery (ITI 23–4), and one which presupposes that ‘Convention T embodies our best intuition as to how the concept of truth is used’ (ITI 195). Even in Davidson’s recent work, which highlights the limitations of the semantic theory, we read: ‘Tarski has told us much of what we want to know about the concept of truth’ (SCT 295). Furthermore, he justifies his redeployment of Tarski’s theory not just by maintaining that the latter can do service as a theory of meaning, but also by reflections about truth: a theory of truth can only amount to a theory of meaning because the concept of truth is substantial yet indefinable. Even if a theory of truth and a theory of meaning can both take the form of a Tarskian theory of truth-conditions, however, there is a difference between them: the former should shed light on the concept of truth, the latter on the concept of meaning. Ultimately, Davidson does not heed this difference. He shrugs off certain shortcomings the semantic theory has as an account of truth, while at the same time complaining about its failure to deliver on tasks exclusive to a theory of meaning. At any rate, this is what I hope to establish in the next section. 2

Has Tarski defined truth?

The semantic theory is a theory of truth in a very different sense from traditional theories like correspondence, coherence and pragmatist theories. The latter are explanations of the concept of truth, and they can

Truth

be stated in vernacular prose of no more than a paragraph. By contrast, Tarski provides a truth-theory, an axiomatic formal theory geared to the derivation of semantic statements about each sentence of an artificial language. Ironically, in spite of the elaborate nature of the exercise, it is a moot question whether and in what sense Tarski has explained truth at all. Davidson himself has come to accept that there are essential aspects of truth not captured by Tarski, and this has driven him to the conclusion that truth is so basic a concept as to be indefinable. He has adduced four arguments to this effect – the threat of paradox, the restriction to particular languages, the problem of empirical content and the Socratic predicament. Davidson is interested in natural languages that contain singular terms, unlike Quine’s canonical notation. A truth-theory for such a language defines denotation and satisfaction for simple predicates through axioms like the following: ‘snow’ denotes snow ‘grass’ denotes grass etc. for all singular terms

an object x satisfies ‘is white’ iff x is white an object x satisfies ‘is green’ iff x is green etc. for all simple predicates.

It then provides a recursive definition of satisfaction for complex cases in terms of the simple ones. This definition fixes the use of ‘satisfaction’ by establishing what sequences satisfy each (open or closed) sentence. But it does not eliminate ‘x satisfies y’ for variable x and y. Hence, as Quine points out, it is a definition only of ‘a lower grade’. Furthermore, the attempt to transform it into an explicit definition leads to Grelling’s paradox. An explicit definition of satisfaction is possible in a metalanguage M, but an explicit definition of satisfaction for M requires a yet stronger metalanguage M
, and so on. We are heading up a hierarchy of levels of satisfaction and of truth (PL 42–6; Q 147–9, 215–16; PT 84–90; SS 64–6). For this reason, Davidson renounces Tarski’s final step of converting the recursive characterization of satisfaction into an explicit definition of ‘true’, and requires only a ‘theory of truth’ (ITI 66, 72; SCT 296). But this does not avert the threat of paradox. The natural languages for which Davidson seeks to furnish a truth-theory are semantically closed: they contain semantic notions and means of talking about their own expressions. Davidson admits that he has no compelling answer to this problem. He thinks, however, that we can carry on ‘without having disinfected this particular source of conceptual anxiety’ (ITI 28–9). He hopes that the theory can be confined to segments of natural languages without semantic terms like ‘true’. But this militates against his own requirement that a

111

112

Quine and Davidson on Language, Thought and Reality

theory of truth/meaning must cover the whole language (see chapter 5, section 2). Davidson’s second response is that ‘there may in the nature of the case always be something we grasp in understanding the language of another (the concept of truth) that we cannot communicate to him’ (ITI 29; see PCT 16). Quine’s hierarchy of truth-predicates beckons here. On pain of contradiction, we can never elucidate the concept of truth of the language of our last theory of truth. That is why Tarski ‘proved’ truth to be indefinable by showing that no single concept of truth can be defined in a consistent language. There can be no definition of the form ‘For all languages L and all sentences s in L, s is true in L iff . . . s . . . L’ (SIO 155–6; SCT 285–6, 294–6; RD 36; FDT 265, 269–70). This is indeed an inevitable outcome of Tarski’s resolution of the paradoxes. For all that, it remains an astonishing confession of semantic impotence. It lends credence to alternative solutions according to which self-referential sentences like (2) are not well-formed sentences capable of being true or false (White 1970: 97–8; Haack 1978: ch. 8). If these alternatives can be sustained, Tarski’s distinction between object- and metalanguage will not be necessary to prevent the paradoxes. Nor is it sufficient, since a strict distinction between truth-predicates of different levels is not always feasible (Kripke 1975). Consequently, one may sympathize with Quine’s conclusion that ‘all is not well’ (Q 216). He applies this verdict to the concept of truth, but it may be more apposite for the Tarskian approach to truth he shares with Davidson. Basing the recursive definition of reference (denotation and satisfaction) on a finite and exhaustive list of the basic cases also leads to another impasse. Such an enumeration cannot indicate how these terms apply to new cases, and hence how they apply in general. As Black (1948) and Field (1972) have pointed out, Tarski’s procedure provides only an extensional equivalence between definiendum and definiens. As a result, it is incapable of dealing with the addition of new terms. If we have a definition of satisfaction for L and we add to L the predicate ‘x flies’, it will entail that ‘x flies’ is not satisfied by an object that flies, or by anything else. For this reason, Tarski’s theory does not amount to a genuine reduction of semantic notions to physical and syntactic ones, even though no semantic terms occur on the right-hand side of the aforementioned axioms. Davidson concurs, but unlike Field he does not deplore this fact, since he regards such reduction as neither feasible nor desirable (ITI 217–18; SCT 287, 297n, 314).

Truth

A related objection hails from Dummett. Tarski did not define a concept ‘s is true in L’ for variable L. Instead he showed how to define a oneplace predicate ‘s is trueL ’ for specific formal specified languages, but without showing what these definitions share, and hence without showing what the truth-predicates for different languages have in common. As a result, Tarski’s definitions tell us nothing about the ‘point’ of the concept of truth, which, according to Dummett, is that by convention we aim at truth whenever we speak. He illustrates his point through the following analogy. If we want to know what winning is, it is not enough to be told what counts as winning in each of a number of different games. For the rules of these games take for granted rather than explain the classification of the final positions into winning and losing (1978: xx–xxi; ch. 1). Against this argument, Ramberg (1989: 135–6) objects that what we really need to know is not what winning is, but rather ‘how to play the game’: the aim is to elucidate not the general concept of truth, but the semantic structure of particular languages. This response ignores the fact that Dummett is concerned with that general concept. It also ignores Dummett’s contention that one cannot even ‘play the game’ (speak a language) properly without having grasped its aim (the concept of truth). Davidson’s own reservation about the analogy is more plausible. Unlike playing games, speaking a language does not have a single conventional purpose, not even speaking the truth (ITI 267–8; RD 36). But this leaves intact Dummett’s basic point: if we want to know what truth is, it is not enough to know how the truth-conditions of sentences of individual languages depend on the semantic properties of their constituents. Davidson later conceded this point. Tarski fails to elucidate the general concept of truth, because he does not tell us what the various truth-predicates have in common (SCT 285–6; RD 36). It is tempting to think that Convention T does just that. It is not confined to a particular language, since it maintains that for any language the definition of its truth-predicate must entail all true T-sentences (ITI 65–6). Still, Davidson denies that Convention T furnishes a general explanation of truth. His reason is that Convention T is acceptable only because T-sentences are ‘clearly true’, something we can recognize only because we already have a pre-analytic understanding of the general concept of truth (ITI 218). But if this complaint were licit, it would count equally against any explanation of truth that invokes conceptual truths about truth. The explanation of any concept relies on a pre-analytic understanding of the explanandum against which the explanation is tested.

113

114

Quine and Davidson on Language, Thought and Reality

Davidson further complains that although Convention T tells us what kind of ‘pattern truth must make’ in any language, it does not tell us how to ‘identify the presence of such a pattern’ in people’s behaviour. Tarski takes the translation from L into M for granted, and does not address the question of ‘how we know that a theory of truth for a language is correct’ (SCT 295; see ITI 214; QVT 439). However, ‘theory of truth’ here means a theory which specifies the meaning of sentences of L by stating their truth-conditions. Davidson’s complaint is that Tarski does not tell us when the sentences of L quoted on the left-hand side of T-sentences mean the same as the sentences of M used on the right-hand side. But this is simply not incumbent upon a definition of truth. It is one thing to explain what the English term ‘true’ means, and we can arguably do so by reference to disquotational T-sentences like (1). It is another thing to specify under what conditions we would call individual English sentences ‘true’. Unlike the latter, the former does not presuppose an understanding of how the truth-conditions of English sentences depend on their structure and constituents. A familiar reaction to Tarski’s theory is that disquotational T-sentences are so trivial as to be comical. In response, Tarski’s defenders have made two points. The first is that what counts is not the trivial nature of the T-sentences but the far from trivial process of deriving them from the axioms of the theory (ITI xv, 50–1; Platts 1979: 32). However, what is nontrivial here is only what this process reveals about the logical form of the object-language sentence ‘s’, not what it tells us about the phrase ‘true’ and its cognates. Consequently, this response makes Tarski’s theory non-trivial as a theory of meaning, but not as a theory of truth. The second response is that T-sentences are informative; they make empirical claims about the quoted sentence of L (ITI 73–4; Evans and McDowell 1976: vii–xi; Platts 1979: 13; Evnine 1991: 80; Wiggins 1997: 9). This is obvious in cases in which the object language is not part of the metalanguage, as in (3). (3) expresses a contingent truth about German. It could have been the case that grass’s being green rather than snow’s being white makes ‘Schnee ist weiß’ true. It might be held that (3) cannot be a contingent truth, because it is an essential feature of German that snow’s being white, and only snow’s being white, verifies what is expressed by ‘Schnee ist weiß’. But it remains an empirical question whether the speakers for which the theory is constructed speak German rather than some other language. Again, T-sentences like (3) are derived from the semantic axioms of German

Truth

without any reliance on empirical information. But whereas for Tarski the axioms are stipulations that create an artificial language, for Davidson they are empirical conjectures about what the terms of a pre-established natural language mean. Davidson is right, therefore, that nothing prevents us from taking T-sentences as empirical (RD 36; SCT 288–94). Doing so, however, is incompatible with Tarski’s project. Whereas Davidson tries to specify the meaning of the sentences of a language L, Tarski tried to define ‘true-in-L’. If we assume the synonymy of ‘Schnee ist weiß’ and ‘snow is white’, as is proper when our purpose is to define truth, (3) is a conceptual rather than an empirical truth. This is central to Tarski’s claim that T-sentences constitute partial definitions of ‘true’. Rightly so: rejecting such equivalences is incompatible with a proper understanding of ‘true’. That T-sentences are trivial is a strength rather than a weakness. A definition of truth is responsible only to the way ‘true’ and its equivalents in other languages are used, not to the semantic properties of English or any other language as a whole. An explanation of truth need not tell us what the sentences of a given language mean, but only what ‘true’ means. Davidson, by contrast, now maintains that truth should be treated as indefinable. It is a ‘primitive’ or ‘basic’ concept which cannot be defined ‘in terms of clearer or even more fundamental concepts’. Therefore we should abandon any attempt of analysing or defining truth. We should also abandon ‘attempts at substitutes for definitions’ in the guise of a ‘brief criterion, schema, partial but leading hint’; Davidson is here thinking of schemata like (T), which deflationists regard as capturing the core feature of truth. This verdict is partly based on general pessimism about the feasibility of definitions. Ever since Socrates, philosophers have failed spectacularly to define any but the most trivial concepts. Concepts like truth, knowledge, belief, cause, etc., are of philosophical importance precisely because they are ‘the most elementary concepts we have, concepts without which (I am inclined to say) we would have no concepts at all’. Instead of trying to define them in terms of more fundamental concepts, we should simply try to relate these basic concepts to each other (FDT 264, 276). This proposal resembles Strawson’s idea that philosophical analysis should be ‘connective’ rather than ‘reductive’ (1992: ch. 2). Unfortunately, Davidson mistakenly identifies the quest for definitions with the attempt to analyse concepts into simpler elements (Grayling 1997: 168–76). But his failure is mainly terminological. If his objections to realism, antirealism and deflationism are sound, no illuminating definition of truth

115

116

Quine and Davidson on Language, Thought and Reality

will be in sight. Furthermore, there is nothing obscurantist about connective analysis. As Davidson makes admirably clear, to reject the quest for reductive or analytic definitions means neither that ‘we can say nothing revealing’ about truth, nor that the concept is ‘mysterious, ambiguous or untrustworthy’ (FDT 265). Davidson recognizes that Tarski has not given the intension of truth even for particular languages (SCT 294–5). But in that case he should also renounce his claim that any predicate which applies to sentences of L iff their translations in the metalanguage are true simply is the predicate ‘true-in-L’ (ITI 65; SCT 289). That the predicate defined by Tarski applies to all and only the true sentences of L no more guarantees that it is the truthpredicate than the fact that ‘cordate’ applies to all and only creatures with kidneys shows that it is the predicate ‘renate’. Furthermore, like Quine (PL 42; Q 214; PT 81–2; SS 66–7) Davidson is adamant that ‘the totality of T-sentences fixes the extension of the truth-predicate uniquely’ (ITI 218; see 65, 73–4; SIO 138–9; SCT 294–5). But that claim is inaccurate. The totality of T-sentences does not tell us which sentences of L are in fact true, not even in the case of artificial languages; instead, it tells us under what conditions they are true. The inaccuracy conceals a crucial point about the semantic theory. It allows one to derive sentences about the conditions under which sentences are true, and thereby reveals how the truth-conditions of such sentences depend on their structure and constituents. But this leaves open a central question, namely what truth-conditions are conditions of. According to the early Davidson (ITI 70), we can establish what we mean by calling a sentence of a language L true by showing how its T-sentence follows from the axioms of the Tarskian theory of L. This sounds prima facie plausible. Just as we explain what a vixen is by specifying the conditions under which something is a vixen, we explain what truth is by specifying the conditions under which something is true. Moreover, if, by contrast to a Davidsonian theory of meaning, we take disquotation or translation for granted, T-sentences like (4) ‘p & q’ is true iff p and q

are just as conceptual as (5) x is a vixen iff x is a fox and x is female.

Nevertheless, (4) reveals the conditions under which ‘p & q’ is true only in the sense of showing how its truth-value depends on that of its

Truth

components. As a result, it may display the logical form of ‘p & q’ or its meaning. But it does not explain what we mean by calling this sentence true, let alone what we mean by calling a sentence true. This shortcoming has its roots in Convention T. Like Davidson, Tarski expects from a theory of truth both too much and too little. He expects too much, in that the definition should not just entail for each sentence s under what conditions it is true – which any explicit definition will do – but that this statement of truth-conditions should be derived from the semantic axioms of the language and that its derivation should reveal the logical structure of s. He expects too little, because he ignores the question of what it is for a sentence to be true, the question of what the fulfilment of truth-conditions amounts to. An analogy may help to illustrate this difference. It is one thing to know whether a specific legal contract is valid. It is another thing to know under what conditions contracts are valid in a particular legal code. And it is yet another thing to know what it is for a contract to be valid. The first kind of knowledge is empirical, and it corresponds to the knowledge of whether a particular sentence is true. The second kind of knowledge is a priori in the sense that it requires knowledge not of empirical facts but of a particular system of rules. It corresponds to the ability to derive T-sentences from the axioms of a Tarskian theory. But only the third kind of knowledge provides us with an analysis or explanation of what a valid contract is (what ‘valid contract’ means). Tarski provides this kind of explanation only in so far as he provides an explicit definition of truth as satisfaction by all sequences. But that definition is only a technical device. Its conceptual meat lies in the idea that the truth-conditions of sentences depend on the reference of their constituents – the denotation of singular terms and the conditions under which its predicates are satisfied (ITI 216–18). But Tarski defines denotation and satisfaction only in the sense of listing what singular terms denote what objects and what predicates are satisfied by what sequences. This once again leaves open what it is for a singular term to refer to an object or for a predicate to be satisfied by a sequence. Davidson candidly admits that the semantic theory fails to explain or analyse the notions of reference and satisfaction (ITI 216–18). But he regards this as a virtue, because he treats reference and satisfaction as purely theoretical notions, that play ‘no essential role in explaining the relation between language and reality’ (ITI 225, see 221–2). Only the T-sentences themselves count, the axioms from which they are deduced

117

118

Quine and Davidson on Language, Thought and Reality

do not. Davidson distinguishes between explanation within a theory and explanation of the theory. Within a Tarskian theory, the concept of reference is essential, because it explains how the truth-conditions of sentences are determined. But the connection between the semantic theory as a whole and actual linguistic behaviour relates only to the truth of sentences not to the reference of their constituents. Even if Davidson were right on this, however (see chapter 7, section 4), it would not absolve him from the obligation to explain reference and satisfaction. That only T-sentences feature in the empirical evidence for a theory does not mean that the basic notions employed in the theory can remain unexplained. To suppose otherwise is to confuse the evidential basis of a theory with its conceptual apparatus. Thermodynamics needs to explain the notion of entropy, even though the latter does not feature in the observations against which it is tested. What is more, the semantic theory faces a formidable task here, since the explanation of satisfaction seems to take us right back to truth. To say that grass satisfies ‘x is green’ is to say that ‘x is green’ holds true of grass, which in turn is simply a roundabout way of saying that it is true that grass is green. In so far as Tarski’s theory sheds light on the concept of truth, it is due not to its mesmerizing formal apparatus but to its starting point in logical equivalences like (1). What do such equivalences tell us about truth, and how should they be formulated? Tarskian theories tend to bury such questions under a barrage of technicalities. Now we need to dig them up again. We must turn from Tarskian truth-theories, which are tools for specifying truth-conditions for the sentences of a particular language, to theories of truth, which try to elucidate the concept of truth. 3

Unbearable truth-bearers

In the previous section we found ample reason to sharpen Davidson’s verdict that Tarski has not said ‘all there is to say about truth’. The current section scrutinizes two interrelated assumptions that are shared by Tarski, Quine and Davidson. The first is that truth and falsehood attach primarily to sentences. The second is that truth is relative to a language, as with Tarski’s predicates ‘true-in-L’, or relative to speakers and times. These assumptions raise the vexed question of what the bearers of truth and falsehood are. One traditional answer is mentalism. It attributes truth to ideas, beliefs or judgements, and conceives of these as mental phenomena which are the inalienable property of individuals. Another

Truth

answer is Platonism, which attributes truth to thoughts or propositions that exist timelessly, irrespective of whether human beings understand, believe or propound them. The final answer is nominalism, which holds that what is true or false are neither mental nor abstract objects, but material ones, namely sentences. The mentalist position can be disposed of through the objections of Platonists like Frege. Even if believing a truth is a phenomenon in the mind of an individual, the truth itself, what is believed, is not. For truths can be communicated, and one and the same truth can be believed or rejected by different individuals. Nominalism does not face this problem, since its preferred truth-bearers are intersubjectively accessible. Often a nominalist account is simply taken for granted. Thus Quine maintains that ‘there is surely no impugning the disquotational account, no disputing that “snow is white” is true if and only if snow is white’ (PT 93). At other times it is defended as being the ‘simplest’ or ‘most convenient’ account, or on the grounds that ‘no confusion results’ from adopting it (Dummett 1981: 444; Tarski 1949: 53+n; ITI 34, 43–5). But that an assumption is convenient does not entail that it is correct. At issue is not the convenience of stipulations, but the question of what ‘true’ can be applied to or is primarily applied to. Tarski was alive to this point. His definition of truth aimed ‘to catch hold of the actual meaning of the old notion’. The claim that sentences are truth-bearers is ambiguous. It can mean either that a truth is a token-sentence, a particular utterance or inscription, or that it is a type-sentence, an acoustic or typographic pattern which can be instantiated by indefinitely many tokens. This second position was the one in effect held by Tarski. Since types are universals rather than particulars, this position is not nominalist, though it still constitutes a form of sententialism. In any event, the idea that the bearers of truth and falsity are type-sentences runs into a difficulty (Strawson 1971: ch. 1). Tokens of one and the same type can be used to express either a truth or a falsehood, depending on who uses it, where and when. This is not just the case with sentences involving indexicals (‘You owe me ten dollars’), but also with those involving most kinds of proper names (‘Jack loves Jill’). In response to this difficulty, Quine and Davidson have modified Tarski’s position. Both have at various times toyed with either one of two ways of fixing the truth-values of sentences (SS 77–8 and ITI 33–4, 43–5, 58, 74, 213). The first is to relativize truth to speakers and times. In this vein, Davidson views ‘truth as a relation between a sentence, a person, and a

119

120

Quine and Davidson on Language, Thought and Reality

time’ (ITI 34; see 135–6, 151; PCT 12). Accordingly, the theorems of a truththeory `a la Davidson are T-sentences of a novel kind, such as ¨ (6) ‘Ich bin glucklich’ is true in German as spoken by x at t iff x is happy at t.

Similarly, Quine tries to do justice to this ‘relativity to times and places’ by holding that truth cannot be a feature of a sentence, but is rather ‘a passing trait of a sentence for a man’ (WO 191–2). ‘Occasion sentences’ with an indexical element (‘I am happy’, ‘It is raining’, ‘That’s a rabbit’) are ‘true on some occasions, false on others’ (PT 3). The second option is to claim that it is token-sentences – utterances or inscriptions – rather than type-sentences which are true or false. ‘What are best seen as primarily true or false are not [type-]sentences but events of utterance’ (PL 13; see PT 78–9). Davidson insists that theories of truth are primarily concerned with sentences as uttered on particular occasions and that utterances and inscriptions are the bearers of truth (ITI 118, 201; SCT 286, 309–10; FDT 275–6). In the end, relativizing truth and ascribing it to speech acts may come to the same thing. To hold that a sentence is true ‘as (potentially) spoken’ by a person at a time is tantamount to saying that it is the (actual or potential) utterance of the sentence by a person at a time which is true. This switch to token-sentences leads to a hard-headed nominalism. But it does not solve the problem of variable truth-values: if Smith and Jones cooperate in writing ‘I am 30 years old’ on a blackboard, that token could be both true and false. Furthermore, there are more truths than sentence-tokens. For example, not all truths of the form ‘x > 1’ could ever be uttered (Kunne 1983: 277–8). Finally, token-sentences are particu¨ lars, acoustic events in the case of speech, visible material objects in the case of writing. They are located in space and time, and they have causes and effects. By contrast, truths or falsehoods are non-physical, atemporal and non-spatial. A spoken token-sentence can last for two seconds, be loud or high-pitched, but a truth cannot. A written token-sentence can be 10cm long, consist of ink or chalk, and one can turn it upside down. But a truth cannot occupy any space or consist of any material stuff. One cannot destroy a truth by wiping a blackboard, or by setting fire to a piece of paper, and one cannot turn it upside down, except metaphorically (Baker and Hacker 1984: 183). There is a final option for sententialists. They may concede that the bearers of truth and falsity are types rather than tokens, that is, abstract

Truth

entities which are proof against vicissitudes like destruction through negligence or arson. To avoid the problem with Tarski’s original suggestion, they can opt for type-sentences with a fixed rather than variable truth-value. In addition to suggesting that the bearers of truth are tokensentences, Quine also insisted that this role can be occupied by ‘eternal sentences’. Eternal sentences are type-sentences – ‘repeatable linguistic forms’ or ‘sound patterns’ – from which all indexical elements (pronouns, tenses, etc.), as well as all ambiguity and vagueness, have been removed, for example ‘Bernard Ortcutt owes W. V. Quine ten US dollars on 15 July 1968’ (WO 191; PT 79; SS 77–8). Unfortunately, this suggestion works only if all truths expressed by occasional sentences can be captured by eternal sentences, something that has been questioned for sentences involving indexicals and proper names (Perry 1979; Kripke 1980: 80–5, 106). Furthermore, the criteria of identity for truths are not the same as those for sentences even of the eternal variety. One and the same truth can be expressed not just by different token-sentences (e.g. different utterances of ‘Snow is white’), but also by different type-sentences, such as ‘Vixens can catch rabies’ and ‘Female foxes can contract hydrophobia’, or ‘Snow is white’ and ‘Schnee ist weiß’. Moreover, some of the features highlighted by opponents of sententialism (White 1970: 7–18; Hacker 1996a: 299–300) distinguish truths from type-sentences just as much as from token-sentences. What is true, that is, what is said, asserted or claimed, can be in German, but unlike a typesentence, it cannot be German. A type-sentence can contain six words or two commas, it can be elegant or clumsy, ungrammatical, hard to pronounce or badly punctuated, but none of this can be said of a truth. What is true, for instance that the universe is unlimited yet finite, can be astonishing or remarkable. But sentences are remarkable in a different way, e.g. by virtue of containing eleven consecutive occurrences of the word ‘had’. There are several responses to these arguments. First, it might be objected that they hold of truths, but need not hold of the bearers of truth, which may therefore be sentences after all. This response fails. The bearers of truth are what ‘is true’ applies to, and hence they are what is true. But what is true is a truth. If it is true that p, then that p is a truth. Consequently, what holds of truths also holds of the bearers of truth. Next, sententialists might insist that some truths simply are German or hard to pronounce, and so on. But if this insistence is to be more than a petitio principii, they need to explain what it means for a truth to be German, other than that the sentence which expresses it is German. Moreover, they

121

122

Quine and Davidson on Language, Thought and Reality

would also have to insist that ‘Snow is white’ and ‘Schnee ist weiß’ simply do not express the same truth. They are committed to the view that while both of these sentences are truths, they are not the same truth, since one of them is English and the other German. But that conclusion blatantly violates the way we individuate truths. We actually apply the terms ‘true’ and ‘false’ not to the act of saying or writing something, nor to what is used to say or write it (a token of a typesentence), but to what is said or written, for example, that there are mountains or that uranium is radioactive, and also to what is made in saying it, an assertion or statement. Indeed, T-sentences like ‘“Snow is white” is true if and only if snow is white’ are simply unheard of in ordinary English. To be sure, we sometimes speak of words or sentences as true, but this is elliptical for what is said when we use them. Even those who postulate other truth-bearers often acknowledge that we actually use ‘true’ in this way (e.g. Platts 1979: 38–9). Our protagonists remain unimpressed. Davidson grants that ‘speech acts are not said to be true’. Nevertheless, he insists that ‘verbal felicity apart’ there is no reason not to call the utterance of a sentence true, and that we can explain what it is to make a true statement ‘in terms of the conventional relations between words and things that hold when the words are used by particular agents on particular occasions’ (SCT 310; ITI 44–5). Quine for his part pours scorn on the shortsighted but stubborn notion that a mere string of marks on paper cannot be true, false, doubted or believed. Of course it can, because of conventions relating it to speech habits and because of neural mechanisms linking speech habits causally to mental activity. (SS 94; see FLPV viii)

The notion is indeed stubborn. Yet it is shortsighted only in the sense of being based on a closer look at the relation between truth, sentences and conventions. Conventions relating sentences to occasions of utterance indeed play a crucial role for truth. But they establish what is said by the use of a sentence on a particular occasion. And it is what is said that is either true or false. It might be objected that the metaphysical nature of truth cannot be determined by reference to how we use ‘true’ and its cognates. But even if natural kinds have metaphysical essences to be discovered by science, truth is not an object or kind about which we could make such discoveries (see CT 105). It is hard to see how our concept of truth could be completely independent of how we explain and use ‘true’ and its cognates. Could it turn

Truth

out, for example, that, the ordinary use of non-philosophical philistines notwithstanding, truths are actually a species of Amazonian termites? On what grounds can we rule out that suggestion or a serious proposal like mentalism, if not by reference to what it makes sense to say about truths, for example that they cannot crawl but can be communicated? Nor can my points be dismissed as concerning only the allegedly superficial pragmatic aspects of our use of ‘true’. If any features of that use have semantic significance, they are that we combine ‘is true’ with that-clauses rather than quoted sentences, and the way in which we individuate truths. There is a version of the objection which is both more sophisticated and closer in spirit to Quine and Davidson. They might hold that ‘it does not make sense to say . . .’ creates an intensional context. If so, that it does not make sense to say things about truths which it does make sense to say about sentences does not show that truths are not identical with sentences. Our alethic vocabulary picks out a part of physical reality (tokensentences), but does so in a non-physical manner, just as, according to anomalous monism, our mental vocabulary picks out neurophysiological phenomena in an irreducibly mental manner. There is a kernel of truth in this response: truths are not part of a reality beyond the physical world. But it does not rehabilitate sentences as truthbearers. For one thing, it is far from obvious that ‘it makes (no) sense to say . . .’ creates an intensional context. ‘It makes no sense to say that the central truth of the theory of relativity is 2cm long’ and ‘That E = mc2 is the central truth of the theory of relativity’ does entail ‘It makes no sense to say that that E = mc2 is 2cm long’. In any event, the fact there are more truths than token-sentences remains an insurmountable obstacle to identifying the former with the latter. ‘True’ applies primarily to what is said rather than to type- or tokensentences. This ‘impugns’ not just Quine’s disquotational account but also Tarski’s semantic theory and Davidson’s alethic theory of meaning. It is wrong to relativize truth to speakers and times. Equally, the idea of truth as a passing trait of non-eternal type-sentences violates the atemporal nature of truth by turning it into a property that can be contracted or lost like measles. It ignores the distinction between ‘true’ and ‘true of ’. The statement ‘children worked in the mines in England in the eighteenth century’ has not ceased to be true, although it has ceased to be true of England that children work in its mines (White 1970: 23–4). As Quine himself recognizes, when a scientific theory is falsified ‘we do not say that it had been true but became false’ but that it turned out not to be true after all (SS 67).

123

124

Quine and Davidson on Language, Thought and Reality

Finally, it is wrong to relativize truth to languages. Tarski, Quine and Davidson insist that we must distinguish between ‘true in English’ and ‘true in German’ because one and the same acoustic or typographic type can have different meanings in different languages (1969: 64; FLPV viii; ITI 98). But this possibility does not entail that truth is relative to a language. Outside formal semantics, there is no such thing as ‘true in English’ or ‘true in German’. What can be relative to a language is what is said by using a particular inscription or sound. Truth itself is a property of what is said, not of the words used in saying it; ergo, it is not relative to the language in which it is said and ‘true’ is not a metalinguistic predicate. By the same token, Davidson is wrong to maintain that Convention T embodies our best intuition about truth, or that there is an ‘essential’ connection between truth and translation because truth can only be explained by appeal to translations from object- to metalanguage (ITI 194–5). Although his sententialism is untenable, Davidson is right to insist on some link between the notion of truth and the linguistic activities of human beings. He ascribes to Dewey the view that ‘nothing in the world, no object or event, would be true or false, if there were not thinking creatures’. He defends this claim on the grounds that without creatures using sentences, nothing would count as a sentence, and hence the concept of truth would ‘have no application’ (SCT 279, 300; see FDT 276–7).1 This idea has affinities with the later Wittgenstein, and it is anthropocentric in the pragmatist tradition. But it is not committed to absurd antirealist claims. Davidson does not hold that without people there would be no mountains, or that it is for human consensus to decide whether it is true that there are mountains. Rather, he maintains that the notion of truth in some way depends on thinking creatures. In fact, one can distinguish two different claims in this position. One is that the concept of truth exists only due to the activities of thinking creatures; the other that it can only be explained by reference to such activities. The first of these claims is directed against a Platonist account of concepts, the second against a Platonist account of truths. Quine and 1. It is difficult to square this idea with Davidson’s claim that the existence of a language does not depend on anybody speaking it, because it is an abstract object, unobservable and changeless (SP 255–6). If languages do not depend on speakers, how could sentences? In my view, Davidson is unwittingly operating with two different notions of a language, a technical one employed in formal semantics (e.g. Lewis 1983: ch. 11), and the ordinary one shared by pragmatism and common sense. In any event, to maintain that languages can exist without speakers is implausible. It implies that languages cannot die out, and that, as regards existence, there is no difference between Spanish and Mohican.

Truth

Davidson accuse Platonism of construing sentences as names and of positing self-subsisting entities – propositions – for them to name, entities which are not just abstract but intensional and hence bereft of criteria of identity (WO 246 and ITI ch. 3; DD 232). Fortunately, the tenability of this stance does not hinge on the success of Quine’s crusade against intensions, but on the difference between sentences, names and that-clauses. What is true – a truth-bearer – is also what can be believed or said – the so-called ‘content’ of a propositional attitude. And what is true/believed/said is denoted by that-clauses: it is true/believed/said that p. But although that-clauses are noun-phrases, they refer to objects only in the weak logical sense of an object of discourse (see chapter 2, section 5). What is said is not like what is eaten (a cake). The word ‘what’ here introduces a that-clause, not the name of a thing. Like the ‘what’ in ‘what x means’, it is an interrogative rather than relative pronoun. If A and B believe the same truth, this does not mean that there is an object beyond space and time to which they severally stand in the relation of believing. It just means that the same answers can be given to the question ‘What does A believe?’ and ‘What does B believe?’ What they both believe can be expressed by the same sentences and designated by the same that-clauses. The debate between Platonism and nominalism is misguided in so far as it turns on the existence and longevity of certain entities. At the same time, both contain kernels of truth. Platonism is right to point out the difference between a saying and what is said. Nominalism is right in insisting that what is said is not an abstract entity which exists separately from the utterance or inscription. It is distinct from the utterance only in the sense in which, for example, a pound is distinct from a pound note. The note does not name a pound, but to present the note is to present a pound. Equally, the sentence does not name or express an abstract entity, but to make the utterance in the appropriate circumstances is to say something – to utter a truth or falsehood. To insist that different token-sentences can say the same thing is not to relate them to a single abstract entity. Instead, it is to group or classify actual or potential token-sentences in a certain way, namely according to what, given linguistic conventions, they say or can be used to say. This supports Davidson’s thesis that we can identify truths only by relating them to speech acts (ITI 44–5; see SCT 309). There are truths no one has ever stated or even entertained. Mysticism notwithstanding, however, all truths are capable of being stated or being entertained. Furthermore, what distinguishes two unexpressed truths is evident from

125

126

Quine and Davidson on Language, Thought and Reality

the declarative sentences that express them. For example, the difference between the most important truth discovered by Newton and the most important truth discovered by Einstein is evident from the difference between what is said by uttering ‘F = ma’ and what is said by uttering ‘E = mc2 ’. Davidson’s anthropocentrism points in the direction of three antiPlatonist claims which do not presuppose either sententialism or antirealism: the concept of truth would not exist were people not capable of calling some things true and others false; truths are sayables and thinkables; and we identify truths by classifying actual or potential utterances according to what they say or would say. The concept of truth is linked to the concept of what people could say and hence to the concept of a language. The argument is the reverse of, but compatible with, Davidson’s idea that one cannot have the concepts of meaning and of a language without having the concept of truth. Though sentences are not the bearers of truth, language is the medium of truth, because it is in language that we say things. 4

Epistemic and correspondence theories

Realists regard truth as entirely independent of our beliefs and tend to promote a correspondence theory; antirealists ‘humanize truth’ by making it dependent on our beliefs, on our ability to learn the truth, or on what can be verified by finite rational creatures. Davidson concedes that both positions answer to powerful intuitions: antirealism to the urge of avoiding scepticism by cutting what is real or true down to what we can know; realism to the idea that reality is independent of what we believe. Nevertheless, he rejects both alternatives. ‘Realism, with its insistence on a radically nonepistemic correspondence, asks more of truth than we can understand; antirealism, with its limitation of truth to what can be ascertained, deprives truth of its role as an intersubjective standard’ (SCT 309, see 297–8). Among the antirealist positions Davidson attacks is Quine’s view ‘that truth is internal to a theory of the world and so to that extent is dependent on our epistemological stance’ (SCT 298, see 306; SIO 144–5). Davidson bases this allegation on Quine’s slogan that ‘truth is immanent’ (TT 21–2). In some passages that slogan boils down to the relativity of truth to a language (RLS 353; see PL 19–20), a claim Davidson accepts (QVT 437–8; PCT 16), wrongly, I argue. In other passages, however, Quine maintains

Truth

that truth is relative not just to a language, but to a ‘system of the world’, a global scientific theory that can be expressed in more than one language: ‘there is no extra-theoretic truth, no higher truth than the truth we are claiming or aspiring to as we continue to tinker with our system of the world from within’ (EEW 327; see RHS 316). Quine also maintains that two systems of the world ‘can be logically incompatible and empirically equivalent’ (RIT 179; see EEW; WO 23; PL 6–7; TT 28–30; PT 95–6). Whatever experience – past, present and future – counts for or against one system also counts for or against the other; yet the truth of the one entails the falsehood of the other and vice versa. This thesis of the underdetermination of theory by evidence is fuelled by Quine’s epistemic holism. He reasons that any theory whatever can be reconciled with any ‘conceivable evidence’ whatever by making appropriate adjustments within the web of beliefs. Underdetermination is contentious as a thesis about scientific method (Laudan 1996: chs. 2–3). It also creates a problem about truth. Let us assume that there can be two empirically equivalent global theories. Then how are we to characterize their truth-values? Quine has oscillated between two replies (RHS 156–7; PT 98–101). The first is ‘sectarian’: ‘one of two systems of the world must be deemed false’ (RHS 156; see TT 21–2). The other one is ‘ecumenical’ and reckons both theories true, on the grounds that to do otherwise would be contrary to empiricism (TT 29). Davidson comes down on the ecumenical side, Quine eventually on the sectarian side. Ironically, both have accused the other of importing epistemological considerations into the concept of truth. Davidson characterizes sectarianism as the view that a speaker ‘at a given time operates with one theory and, for him at that time, the theory he is using is true and the other theory false. If he shifts to the alternative theory, then it becomes true and the previously accepted theory false’ (SCT 306). Davidson protests that one and the same non-indexical sentence cannot be true at one time or for one person, but not at another time or for another person. This is correct. But on a charitable reading, Quine’s sectarianism maintains only that we must ‘deem’ or ‘recognize as’ true whatever theory we currently occupy, not that this theory is true. It is not a form of relativism about truth, but a reaffirmation of naturalism: we cannot get outside of our skins, and there is no ‘tribunal’ of truth other than our best scientific theory (RI 497–8; RLS 353; WO 24–5). Although this defuses the charge of relativism, it reduces Quine’s immanentism to a point about justification. It turns into a point about truth

127

128

Quine and Davidson on Language, Thought and Reality

only through his conviction that the world cannot ‘deviate’ from a theory that conforms to all possible observations (TT 22). But that conviction inevitably leads to an ecumenical position. It implies that two global theories T1 and T2 which conform to all conceivable evidence must both be faithful to the world and hence both be true. Davidson’s ecumenical position is equally flawed, however. If T1 and T2 do not just differ in their conceptual apparatus but are logically incompatible, they may be equally warranted; yet, by definition, they cannot both be true (see RHS 157; PT 97–100). This whole debate between Quine and Davidson is predicated on a misguided assumption, namely that truth is a matter of fitting the evidence. We cannot transcend our own best evidence in deciding what is true. But one role of ‘true’ is to indicate the potential gap between what we have evidence for and what is actually the case. When they are not side-tracked by empiricist allegiances, both Quine (RI 496–7) and Davidson recognize the untenability of such epistemic conceptions, including coherence and pragmatist theories, Dummett’s semantic antirealism and Putnam’s internal realism (SCT 307–8; CT 107–8). At the same time, both resist the lure of correspondence theories. Davidson originally accepted Tarski’s assessment that the semantic theory constitutes a version of the correspondence theory. Even then he rejected the ‘strategy of facts’ according to which a sentence is true iff there is a fact to which it corresponds. But he felt that the semantic theory ‘deserves to be called a correspondence theory because of the part played by the concept of satisfaction’. The property of being true is explained in terms of ‘a relation between language and something else’, just as the property of being a mother is explained by a relation between a woman and her child. In the theorems of the theory ‘there is no trace’ of correspondence, ‘no relational predicate that expresses a relation between sentences and what they are about’. T-sentences simply disquote or translate. Yet satisfaction is essential to the derivation of ‘all those neutral snow-bound trivialities’, and satisfaction is indeed a relation between bits of language and bits of the world (ITI 49, 51, see 37, 48, 67, 194). Subsequently Davidson moved away from the idea that this ‘strategy of satisfaction’ constitutes a correspondence theory: ‘there is nothing interesting or instructive to which true sentences might correspond’, and the idea that there are things which make sentences or beliefs true is unexplanatory (SCT 303; see ITI xviii, 70; SIO 154–5). Tarski puts words into a relation (satisfaction) with entities. But the entities to which sentences

Truth

correspond – infinite sequences – are ‘counterintuitive and contrived’; furthermore, ‘all true sentences would correspond to the same entities’ (SCT 302n; see RD 37–9; FDT 266–8). By (D2 ), every true sentence is satisfied by all sequences, therefore nothing about the truth of a particular sentence is explained by its being satisfied by a particular sequence. Davidson is right to insist that the strategy of satisfaction implies that all true sentences ‘correspond’ to the same entities. Is he also right to hold that the strategy of facts is lumbered with the same unpalatable consequence? In support of this claim, he employs what has come to be known as the ‘slingshot’ argument. (6) ‘Snow is white’ corresponds to the fact that snow is white

is a perfect candidate for expressing the kind of relation sought by traditional correspondence theorists. The slingshot purports to reduce this idea to absurdity. It rests on two assumptions: I. co-referential terms can be substituted salva veritate; II. logically equivalent sentences can be substituted salva veritate.

By (II), we can move from (6) to (7) ‘Snow is white’ corresponds to the fact that ˆx (x = x & snow is white) = ˆx (x = x).

ˆ’ is the class abstraction operator and means ‘the class of objects x such ‘x that’ (WO 164). ˆx(x = x) is the class of objects which are self-identical, that is, the universal class which includes everything. On the left-hand side of the identity statement we have the name of a class of objects which meet two conditions: they are such that they are identical with themselves and that snow is white. If it is true that snow is white, that class will be identical with the universal class, otherwise it will be the null-class. ˆx(x = x & snow is white) is identical with ˆx(x = x) iff snow is white. That is why ‘snow is ˆ(x = x & snow is white) = ˆx(x = x)’ are logically equivalent. white’ and ‘x Furthermore, by (I) we can proceed from (7) to ˆ (8) ‘Snow is white’ corresponds to the fact that ˆx(x = x & grass is green) =x (x = x). ˆ(x = x & snow is white)’ by the name Here we have replaced the name ‘x ˆ(x = x & grass is green)’. Both refer to the universal class, since ‘x both ‘snow is white’ and ‘grass is green’ are true. Finally, the sentence

129

130

Quine and Davidson on Language, Thought and Reality

ˆ(x = x & grass is green) = ˆx(x = x)’ is logically equivalent to ‘grass is green’. ‘x By (II), we therefore move to (9) ‘Snow is white’ corresponds to the fact that grass is green.

By the same token, any true sentence corresponds to any and all facts (ITI 42; CT 106). The explanation of truth as correspondence to facts is therefore vacuous. There are several protective measures against the slingshot. The most promising one is to impugn (II), and with it the step from (7) to (8) (Searle 1995: 221–6). In contexts of the form ‘the fact that φ’, we can substitute co-referential terms within φ, just as (I) has it. If snow is my favourite substance, we can move from (6) to (6
) ‘Snow is white’ corresponds to the fact that my favourite substance is white.

But it does not follow that we can replace φ itself by any logically equivalent sentence, as (II) requires. The fact specified in (6) is simply not identical with the one specified in (7), and hence the step from (6) to (7) is illicit. Even though the slingshot misfires, the strategy of facts is still in trouble. According to Quine facts suffer from the same difficulties concerning criteria of identity as propositions (WO 246–7; see TT 39, 81–3; RI 496). This complaint is doubly problematic. Even if the indeterminacy of translation were to establish that there are no criteria of identity for intensions, facts are arguably extensional in a restricted sense: although sentences with the same truth-value cannot be substituted salva veritate (the fact that snow is white = the fact that grass is green), sentence components with the same reference can (the fact that snow is white = the fact that my favourite substance is white). At the same time, Quine rightly complains about the idea that facts are ‘intervening elements’ between sentences and the world (PL 1–2). According to the correspondence theory, ‘a statement is true only if there is something in the world in virtue of which it is true’ (Dummett 1978: 14). However, facts are not ‘in the world’ (Strawson 1971: 195–9). The fact that Fifth Avenue is five miles long and the fact that it is thirty feet wide are different facts, but the only concrete object involved is Fifth Avenue (WO 247). Unlike objects or events, facts are not located in either space or time. The fact that the Battle of Hastings was fought in 1066 is not to be found on the battlefield, and unlike the battle itself, it is not an event with a beginning and an end. Facts are neither here nor there, neither now nor then.

Truth

This point is connected to another one. The strategy of facts includes in the entity to which a sentence corresponds not just the things the sentence is about, but also whatever the sentence says about them. ‘Dolores loves Dagmar’ and ‘Dolores is next to Dagmar’ are about the same two things, but if true they correspond to different facts, because they involve different relations. This avoids the problem of the strategy of satisfaction, namely that different truths ‘correspond’ to the same entity. The drawback: ‘it becomes difficult to describe the fact that verifies a sentence except by using that sentence itself ’ (ITI 49). Truths and facts are not logically independent items that stand in a relation of correspondence, they ‘were made for each other’ (Strawson 1971: 197). By definition, as it were, the fact that p is the fact that verifies the truth that p; and the truth that p is the truth that is verified by the fact that p. In this spirit, Quine writes that facts are ‘simply projections from true sentences for the sake of correspondence’ (Q 213). Similarly, Davidson holds: ‘Nothing, however, no thing, makes sentences or theories true’ (ITI 194). Admittedly, there is a sense in which the fact that p makes true the statement that p, but it is logical rather than causal. It is like ‘telling lies makes one a liar’ rather than ‘drinking alcohol makes one dizzy’ (Rundle 1979: §§42–3). Once we abandon the idea of facts as independent worldly items that make propositions true, the correspondence theory boils down to a platitude: (D3 ) It is true that p iff it is a fact that p.

But ‘it is a fact that p’ does not refer to a worldly item – that p – and characterizes it as a fact. The absence of any genuine reference is perspicuous in the adverbial construction ‘in fact’ (Rundle 1979: 324–5): (D4 ) It is true that p iff, in fact, p.

Instead of supporting a genuine correspondence theory, (D4 ) has a distinctly deflationary whiff about it. 5

Deflationism and minimalism

Davidson mentions various views that could be used to define deflationism (SCT 283–4; QVT 438; FDT 265, 272). One is that Tarski completely captured the notion of truth, another that truth is disquotational and nothing more. Both characterizations ignore the fact that there are deflationary theories which do not build on Tarski, and in particular do not

131

132

Quine and Davidson on Language, Thought and Reality

treat ‘true’ as a predicate of sentences. The third is more apt, namely that according to deflationism the essential features of truth are captured by schemas like (D5 ) It is true that p iff p.

From this Davidson derives a fourth characterization which is again problematic, namely that for deflationists truth has no important connections with concepts such as meaning, belief and reality. Davidson contradicts those who regard Tarski as a deflationist (SCT 291–5). More important to our concerns is whether Quine is a deflationist. The reason for answering in the affirmative is his emphatic insistence that the disquotational account is ‘a full account; it explicates clearly the truth of every clear sentence’, and his resistance to those – whether correspondence theorists or antirealists – who seek something ‘deeper’ (PT 93; see FLPV 138). At the same time, Davidson can point to features of Quine’s position which seem to mitigate this deflationism (PCT 7–8; FDT 271–2; QVT 440). Quine detects a tension in the concept of truth. On the one hand, truth is supposed to be trivial and immanent in the way demanded by the disquotational account. On the other, he accords ‘truth a dignity beyond disquotation’, the dignity of a goal of inquiry, and muses that ‘truth is felt to harbour something of the sublime. Its pursuit is a noble pursuit, and unending’ (RI 500–2; SS 67; see RLS 353; WWD 77–8). Many philosophers have insisted that the idea of truth as the goal of inquiry rules out deflationism. Why should we pursue truth if disquotation is all there is to it? (Bergstrom ¨ 1994: 432). In reply, Quine maintains that we do not pursue or value truths indiscriminately, but only those which gratify our desires or our curiosity (RI 498). This response will not mollify the inflationists. They will point out that even with respect to the most banal topic we value true beliefs over false ones. Opponents have further complained that the disquotational account deprives truth of its status as a property (Alston 1996: 47). ‘To ascribe truth to the sentence is to ascribe whiteness to snow’ (PT 80). But unlike ‘snow is white’, ‘“snow is white” is true’ is about a sentence rather than snow. That the two are logically equivalent does not guarantee that they are synonymous. There are, however, more promising forms of deflationism. They avoid the assumption that truth is a property of sentences, and have not found favour with our protagonists. Nonetheless, they preserve the insight behind both Quine’s disquotational account and Davidson’s endorsement of Aristotle’s truism (FDT 265–8). By contrast to the redundancy theory

Truth

and to disquotationalism, the minimalist accounts I have in mind accept that

r ‘is true’ and its cognates are not redundant; r ‘it is true that p’ does not mean the same as ‘p’; r ‘it is true that p’ says something about what is said by ‘p’.

But they deny that ‘is true’ ascribes a substantial property or relation. It may be necessary to formalize ‘is true’ as a predicate in order to display the validity of arguments like ‘Einstein’s most famous claim is true; Einstein’s most famous claim is that E = mc2 ; ergo it is true that E = mc2 ’. But it does not follow that ‘is true’ ascribes a property to an object denoted by ‘that snow is white’ in the sense in which ‘is white’ ascribes a property to snow. To establish whether snow has the property of being white, we look at snow. But in order to establish whether it is true that snow is white, there is no point in scrutinizing the proposition itself; rather, what we need to examine is what the proposition is about. To be true, the proposition that p need not possess a substantial property – like corresponding to reality, cohering with our evidence/beliefs or satisfying our needs. It only needs to satisfy truistic logical equivalences like (D5 ). Horwich (1998) has produced a sustained version of minimalism. He treats propositions rather than sentences as truth-bearers, and maintains that the content of the truth-predicate is exhausted by the nonparadoxical instances of the following schema: (D6 ) The proposition that p is true iff p.

Horwich concedes that we cannot turn (D6 ) into an explicit definition through substitutional quantification, since the latter is explained by reference to truth. But he urges us to rest content with (D6 ) as an axiom schema; the totality of its instances constitute the axioms of the theory of truth. In response, Davidson has made three points (FDT 273–4; CT 108–9): (A) Horwich’s position violates one of Tarski’s requirements for a theory of truth, since its axioms are infinite in number. The problem cannot be solved by generalizing his schema through objectual quantification over propositions, as in (D7 ) (p) (the proposition that p is true iff p)

because this would treat sentences as singular terms referring to propositions rather than as expressing propositions.

133

134

Quine and Davidson on Language, Thought and Reality

(B) Not just Horwich’s schema (D6 ), but even its instances like (10) The proposition that Socrates is wise is true iff Socrates is wise are strictly speaking unintelligible. For one thing, it is unclear how the referent of a singular term like ‘the proposition that p’ is determined by the referents of their constituent terms. For another, in (10) the sentence ‘Socrates is wise’ must function as the name of an abstract entity on the left-hand side of the biconditional, and as an ordinary sentence on the right-hand side. (C) Minimalism makes it impossible to account for the conceptual connections between truth and notions like meaning and belief. In particular, the account must take for granted the notion of a proposition, and hence of sentence meaning. It thereby ignores the fact that sentence meaning can only be explained by reference to truthconditions, and hence on the assumption of the concept of truth. I confess to finding objection (B) as incomprehensible as Davidson finds (10). ‘Socrates is wise’ is never the name of an abstract entity. On the righthand side of (10) it is a sentence used to say something, and on the left-hand side it is part of a construction that refers to what is said by the sentence. The construction ‘the proposition that . . .’ is not a functional expression that maps an object named by sentences on to another object, a proposition, as Davidson assumes. Like the pronoun ‘that . . .’ it is simply an operator for nominalizing indicative sentences. The semantic features of ‘Socrates is wise’ determine the referent of ‘the proposition that Socrates is wise’ in this way. The sentence has a meaning (which may in turn be determined by the reference of its constituents); as a result, it can be used to say something. What proposition ‘(the proposition) that p’ refers to, depends on what is said by an utterance of the sentence ‘p’. To know what proposition is being referred to by the that-phrase is simply to know what is said by an utterance of the sentence. No competent speaker of English has difficulties here, and any mystery is of Davidson’s own making. (A) is right to balk at the suggestion that our understanding of the concept of truth consists in knowing an infinite number of axioms. To avoid this embarrassment one can insist that our understanding is captured by the axiom schema itself: we know a priori not only that every substitution instance of it expresses a conceptual truth, but also how to construct any such substitution instance (see chapter 8, section 3). The alternative is to generalize equivalences like (D5 ) or (D6 ) through non-nominal

Truth

quantification into sentence position, as suggested most recently by Kunne (2003). ¨ Irrespective of the feasibility of either move, however, there is a perfectly good colloquial way of expressing the idea behind minimalism. Following Wittgenstein and Strawson one can define truth as follows: (D8 ) No matter what one says (believes, etc.) or could say (believe, etc.), what one says is true iff things are as one says they are.

‘Things’ here does not indicate worldly items which make what we say true, it is simply the colloquial equivalent of a sentential variable (in this respect, it resembles the phrase ‘in fact’ of section 4). If Sarah says truly that it is raining, she states how things are, without predicating something of a thing. The same idea can be used to deal with cases in which no proposition is specified. Instead of ‘Whatever the Pope says is true’ we get ‘Whatever the Pope says, things are as he says they are’. Although the function of ‘true’ may not be disquotation, it is analogous to what Quine calls cancelling semantic ascent (PL 10–3). It acts as a denominalizer or re-sententializer. Human beings have beliefs and express these beliefs through assertoric utterances. They can also specify or allude to what other human beings believe or say, and to what they could believe or say. The most straightforward expedient to do so is to turn declarative sentences into noun-clauses by adding ‘that’. The truth-predicate allows one to undo this nominalization by re-sententializing the thatclause. When we do this, we do not simply say how things are. But what we say is whether things are as they have been or could be said or believed to be, and nothing more. Minimalism can also explain why we value true beliefs. We are interested in how things are; and while a true belief tells us how things are, a false belief does not. Why are we interested in how things are? Above all because it is often essential to our projects and actions (Horwich 1998: 44–6; Alston 1996: 60), as the pragmatists recognized. In cases where it is not, we are interested out of sheer curiosity. Whatever the explanation of that curiosity, it does not upset minimalism. Our interest in whether it is true that p is either simply an interest in whether p (which may or may not be puzzling, depending on the topic of ‘p’) or, less directly, an interest in whether those who claim or believe that p are in the know about how things are. Pace Davidson, such a minimalist approach need not deny that there are conceptual connections between truth and other philosophically interesting notions. It can and must allow, for example, that one can know that p

135

136

Quine and Davidson on Language, Thought and Reality

only if it is true that p. The connection Davidson is interested in most is that with meaning. This is the topic of the next chapter. If this chapter is on the right track, a departure from Tarski is not crippling, as Davidson has it, but mandatory for an account of truth. Still, semantics should turn from truth to meaning, just as Davidson wants it to, not because truth is indefinable, but because a minimalist account replaces puzzles about truth by questions about what we say in using sentences. In Ramsey’s words (1990: 38–9), ‘there is really no separate problem of truth’, only a problem ‘as to the nature of judgement or assertion’.

5

Meaning and truth

If the previous chapter is right, the concept of truth is an appendix to our practice of saying how things are. But what is it to say how things are? More generally, what is it to say something meaningful? What is it for words or sentences to have meaning? These are the central problems of philosophical theories of meaning. I start by sketching Quine’s treatment of these questions (section 1). This sets the stage for Davidson’s truth-conditional theory of meaning. Unlike Quine, Davidson rarely tackles the question of what linguistic meaning is head-on. But the assumptions behind his approach, notably his antipathy towards ‘meanings’, his third-person perspective and his holism, derive from Quine. Davidson’s departure consists in his focus on the question of what form a theory of meaning for natural languages should take. Section 2 presents the requirements Davidson imposes on such a theory, and his argument that they are met by a Tarskian truththeory. The core of that argument is the idea that the meaning of a sentence is determined by its truth-conditions. Section 3 discusses various problems with that idea. The final section turns to some of the recalcitrant idioms that pose a threat to the application of truth-theories to natural languages. 1

Quine on reference, meaning and use

Quine radically challenges some of the assumptions about meaning that guided previous analytic philosophy. His ultimate conclusion is that the intuitive notion of meaning that it shares with common sense is vacuous and must be replaced by a behaviourist ersatz – stimulus meaning. But his reflections start out from within the analytic tradition.

[137]

138

Quine and Davidson on Language, Thought and Reality

For one thing, he takes a linguistic turn by making semantic notions central to philosophy. Thus he regards it as the ‘first milestone of empiricism’ to switch attention from ideas to words, and hence to linguistic meaning (TT 67). For another, in reaching what he regards as a scientifically respectable account of meaning, he relies heavily on some of his predecessors. According to an influential picture of linguistic meaning, every meaningful expression refers to an object, the latter being its meaning. Like Wittgenstein, Quine resists this referential conception of meaning. He starts out with a Fregean distinction between meaning on the one hand, naming or referring on the other (FLPV 8–9, 21, 47–9, 62; TT 43). Referential theories of meaning identify the meaning of an expression with the object for which it stands. But many types of expressions (sentences, logical constants, general terms) are meaningful or significant without referring to an entity. Furthermore, even in the case of expressions that do refer, there is a difference between meaning and referring. A singular term like ‘the fifth man’ need not refer to anything to be meaningful; and two such terms can have different meanings in spite of referring to the same object, as with Frege’s ‘the morning star’ and ‘the evening star’. One might add that to identify the meaning of a word with its referent is to commit a category mistake, namely that of reification: one can be in love with the referent of ‘the deceased wife of Prince Charles’, but hardly with its meaning. As a result, Quine distinguishes two parts of semantics, the theory of reference and the theory of meaning (FLPV 130–8). The former is concerned with notions like naming, truth, denotation and extension, the latter with the notions of meaning, significance, analyticity and entailment. The theory of reference has had its troubles, namely the semantic paradoxes, but these have been overcome by Tarski. By contrast, the theory of meaning is in a ‘sorry state’, which Quine seeks to redress. One difficulty lies in the very idea of words having ‘a meaning’. Even if one accepts that a meaningful expression need not refer to anything, one might insist with Frege that their being meaningful consists in their being associated with an abstract object – a ‘sense’. In line with the later Wittgenstein, Quine rejects any such entities called ‘senses’ or ‘meanings’. The idea of mental or abstract meanings is ‘the myth of a museum in which the exhibits are meanings and the labels are words’ (OR 27). That an expression is meaningful is not due to it being associated with an object which is its meaning. Rather, the ‘useful ways in which people ordinarily talk about meanings boil down to two: the having of meanings, which

Meaning and truth

is significance, and sameness of meaning, or synonymy . . . But the explanatory value of special irreducible intermediary entities called meanings is surely illusory’ (FLPV 11–12, see 49; PPLT 5; RHS 73). Appealing to Dewey and Wittgenstein, Quine goes on to claim that in so far as the notion of meaning is legitimate, it must be explained by reference to verbal behaviour and thus to linguistic use (TT 45, 192; OR 26–8). In this spirit we can say that two words have the same meaning – are synonymous – if they have the same use. Quine is quick to stress, however, that the notion of linguistic use is itself vague. He tries to make it more specific through ‘a behavioral doctrine of meaning’ (TT 47), one which avoids both the reification of ‘meanings’ and the use of intensional notions. The first step towards this doctrine takes us to ‘the second milestone of empiricism’, namely the idea that sentences have ‘semantic primacy’ over words (TT 20, 68–70). This slogan stands in the contextualist tradition of Bentham, Frege and Wittgenstein. It combines several ideas, some good – like the recognition of contextual definitions and sentential paraphrase – some bad – like the claim that the meaning of sentences can be understood independently of the reference of their components (see chapter 7, section 4), and some misleading. Among the latter is the claim that sentences are the ‘significant units’ of language, the ‘primary vehicles’ or ‘repositories’ of meaning (FLPV 38–9; OR 72; TT 3; PT 37; SS 75). In ordinary parlance, we ascribe meaning primarily, though not exclusively, to words. Dictionary definitions, the very paradigm of explanations of meaning, paraphrase words or phrases rather than sentences. This is no coincidence. Far from being the units of significance, most sentences are complex signs. Their meaning is at least partly determined by the meaning of their constituents (Dummett 1981: 3, 593). Understanding the components and mode of composition of a sentence is a necessary (though perhaps not sufficient) condition for understanding it. By and large, we master sentences by learning how to construct them from familiar words. However, Quine need not contest these points. What underlies his misleading claim is the idea that words owe their meaning to their uses or roles in sentences (TT 3; PT 37). The sentence is ‘the unit of communication’ (TT 75): only through sentences, not through words, do we say something. The meaning of a word is determined by how it can be used within sentences. By this token, however, Frege (1953: §62) and Wittgenstein (1922: 3.3) were mistaken to insist that ‘a word has meaning only in the context of a sentence’. It is precisely individual words that have a use in sentences. If it is clear what role a word would play in a sentence, it has a meaning

139

140

Quine and Davidson on Language, Thought and Reality

irrespective of whether it actually fulfils that role on a given occasion. That is why it would be absurd to maintain, for example, that a list of irregular English verbs is a string of meaningless signs unless it occurs as part of a sentence. It is equally absurd to maintain that the words printed in bold at the beginning of dictionary entries are meaningless, all the more so since the text that follows specifies what they mean! Fortunately, Quine never claims that individual words lack meaning, but only that their meaning is to be explained by reference to their role in sentences. He explains the synonymy of words by reference to the semantic equivalence of sentences: two words or phrases are synonymous if substituting one for the other yields semantically equivalent sentences. Two sentences are ‘equivalent if their use is the same’, or, ‘less vaguely’, if ‘their utterance would be prompted by the same stimulatory conditions’ (TT 48). Quine’s behaviourist version of the idea that meaning is use differs from the Wittgensteinian version in four respects:

r it repudiates the idea that meaning can be explained by reference to rules/conventions of use;

r it focuses on the use of sentences, disregarding the explanations speakers give of individual words;

r it stresses the conditions under which speakers utter or assent to a sentence, while ignoring what speech acts can be performed by such utterances and how hearers can intelligibly react to them; r it describes the conditions of utterance in causal and, more specifically, in neural terms, namely by reference to the impact – ‘triggerings’ – of external things on the speaker’s ‘sensory receptors’ (TT 50).

As it stands, this conception of use is too wide. No two sentences are uttered under exactly the same conditions, since ‘the motives for volunteering a given sentence can vary widely, and often inscrutably’ (TT 48; see RR 15, 45–9). That a sentence is or is not uttered under certain conditions may be due not to its meaning but to non-linguistic customs (e.g. etiquette) or the speaker’s state of mind. To keep meaning free of semantically irrelevant motives of language use, Quine focuses on ‘cognitive’ meaning and equivalence, ignoring ‘emotional and poetic’ aspects (TT 53). The cognitive meaning of a sentence is that aspect of its meaning which is relevant to its truth or falsity. But beyond that point, Quine has explained cognitive meaning and equivalence in two distinct ways.

Meaning and truth

His original specification is inspired by verificationism. Quine rejects the positivists’ principle of verification, according to which the meaning of a sentence is determined by the method of confirming or infirming it. His reason is the ‘third milestone of empiricism’, namely holism. This move extends the contextualism of the second milestone from ‘sentences to systems of sentences’; it insists that only networks of sentences, not individual sentences, could have empirical content, because only they face the tribunal of experience (TT 70–1; see FLPV 37–42; PLSP 54–5). Nevertheless, in principle the meaning of a sentence remains determined by ‘what would count as evidence for its truth’ (OR 80–1; see RR 38; NNK 80; RHS 155–6). It is just that individual sentences are not associated with a set of confirming evidence. Quine’s second explanation of the cognitive meaning or equivalence is subject to the same holistic caveat. Two sentences are semantically equivalent if they have the same truth-conditions. The meanings of sentences are determined by their truth-conditions, and those of words are abstractions from the truth-conditions of sentences that contain them (WP 89; TT 38, 47–8, 69; MVD 87–8; RR 65, 78–9; SS ch. 7). Or rather, the meanings are so determined in so far as individual sentences have fixed truth-conditions of their own. These two specifications of cognitive meaning are not equivalent. Someone who has mastered disquotation might know that the sentence ‘Wagner died happily’ is true if and only if Wagner died happily, without knowing what evidence would show it to be true. For Quine, however, the two boil down to the same. According to his principle of charity, the truthconditions of a sentence s are identical with the conditions under which a speaker would assent to s, at least if s is an observation sentence. But these conditions of assent are also identical with the speaker’s evidence for s. Accordingly, truth-conditions, assent-conditions and evidence coincide, at least in the basic cases. The result is a behaviourist conception of meaning that Quine deems to be scientifically respectable, namely ‘stimulus meaning’. The ‘affirmative’ and ‘negative’ stimulus meaning of a sentence s for a speaker A are, respectively, the class of stimulations which would prompt A to assent to or dissent from s. Stimulus meaning simpliciter is the ordered pair of the two. The stimulus meaning of sentences is determined by the neural stimulations which would prompt speakers to assent to them (WO 31–4). In summary, then, Quine holds the following views:

141

142

Quine and Davidson on Language, Thought and Reality

i. the meaning of a word is not an object for which it stands; ii. talk about meanings is to be eschewed in favour of talk about significance and synonymy; iii. the significance of an expression is to be sought in its use; iv. sentences are semantically prior to words, and the meaning of words is to be explained by reference to their role in sentences; v. the (stimulus) meaning of a sentence is determined by the conditions under which a competent speaker would assent to or dissent from it; vi. those conditions are to be described in terms of the neural stimuli that impinge on the subject. 2

Davidson’ s truth-conditional theory of meaning

Except for the last, these points have strong echoes in Davidson. Inquiries into Truth and Interpretation begins with the question: ‘What is it for words to mean what they do?’ But the seminal essays collected in that book do not furnish a straightforward answer to that question. Instead, they contemplate ‘what form’ a theory of meaning for a specific language should take (ITI xiii–xiv, 55). Before Davidson, a theory of meaning was supposed to provide an analysis – in a suitably loose sense – of the concept of meaning. Theories of meaning in this analytic sense are, for example, the aforementioned referential theory, behaviourist and causal theories like that of Quine, verificationist theories, Wittgensteinian accounts of meaning as use, speech-act theories influenced by Austin, and Grice’s theory of communication intentions. By contrast, Davidson envisages a constructive theory which differs from analytic theories of meaning roughly in the way in which a Tarskian truththeory differs from traditional theories of truth. Such a theory does not directly explain what meaning is. Instead, it generates for each actual or potential sentence s of a particular language a theorem ‘that, in some way yet to be made clear, “gives the meaning” of s’, and shows in particular how that meaning depends on that of its components (ITI 23). Analytic theories of meaning should be compatible with the way the meaning of particular sentences is specified or explained; but unlike constructive theories they do not prescribe an algorithm for generating such specifications. Like Carnap, Davidson uses T-sentences to specify the meaning of sentences from an object language. By contrast to Carnap, the languages concerned are natural rather than artificial. Like the Tractatus and Chomsky, Davidson explains the workings of natural languages by reference to

Meaning and truth

complex formal systems which are allegedly hidden beneath the grammatical surface. In fact, his original ambition was to provide a semantic analogue and extension of Chomsky’s generative syntax, a formal system capable of displaying the underlying semantic structure or ‘logical form’ of each sentence of a natural language (ITI 22, 30, ch. 4). A Davidsonian theory of meaning for a particular natural language like English is empirical; and actually to construct such a theory is a task for linguistics (see Larson and Segal 1995). The philosophical work consists in establishing ‘what form’ such a theory could or should take (ITI 55), that is, what it looks like, how it can be constructed, what concepts it employs, how it can be empirically confirmed, etc. But although Davidson does not reduce philosophical semantics to empirical linguistics, he effects an important change of perspective. Along with Quine, Davidson straddles both sides of what Strawson (1971: 171–2) called the ‘Homeric struggle’ between formal semanticists, who treat language as an abstract system of complex rules, and those who regard language primarily as a kind of human activity. He explains natural languages by reference to formal calculi, but he also places it in a larger context of human behaviour and human beliefs. Along with Dummett, Davidson turns the theory of meaning into a theory of understanding, and thence into a theory of linguistic communication. To understand an expression is to know what it means. Starting from this truism, Dummett holds that a theory of meaning must be a theory of understanding, a theoretical model of what competent speakers know when they understand an expression (1978: 217; 1993b: 3, 35). As we shall see (chapter 8, section 3), Davidson does not endorse Dummett’s claim that a theory of meaning must articulate the implicit knowledge allegedly possessed by speakers. His project is more modest, namely to provide a theory knowledge of which would enable speakers to understand these sentences, irrespective of how they actually do so. Still, like Dummett he regards a theory of meaning primarily as an instrument of understanding or interpretation. We interpret a bit of linguistic behaviour when we say what a speaker’s words mean on an occasion of use. We know that the words ‘Es schneit’ have been uttered on a particular occasion and we want to redescribe this uttering as an act of saying that it is snowing. What do we need to know if we are to be in a position to redescribe speech in this way, that is, to interpret the utterances of a speaker? (ITI 141, see 125; Evnine 1991: 72–4)

143

144

Quine and Davidson on Language, Thought and Reality

The central question of Davidson’s theory of meaning is not what it is for an expression to mean something, or even what it is for one expression to be synonymous with another, as in Quine, but rather: What is it to understand what a speaker has said on a particular occasion?

A theory of meaning for a natural language L sheds light on this question by meeting the following objective: someone who knows the theory is in a position to understand every utterance (actual and potential) of L. To this end, Davidson contends, the theory must satisfy the following conditions (ITI xiii, 125–8, 215; SCT 312; see Miller 1998: 247–9): I Extensional Adequacy

For each sentence s of L it generates a theorem that specifies the meaning of s. II Empirical Verifiability

The theory is empirically verifiable without any prior knowledge of what the speakers of L mean by their expressions. III Non-Circularity

The evidence on which the theory is based should not employ ‘concepts too closely allied to that of meaning’, which it seeks to explain (ITI xii). IV Finite Axiomatization

The meaning-giving theorems must be generated by a finite number of axioms and rules of inference. Davidson bases the Extensional Adequacy condition on the holistic nature of linguistic understanding. But as it stands, the condition merely states something which should go without saying, namely that a constructive theory of meaning should be comprehensive, that is, cover all the potential and actual utterances of a group of speakers. Genuine semantic holism, by contrast, insists that a partial theory is not even possible, because the understanding of any individual sentence presupposes an understanding of all the other sentences. The requirement of Empirical Verifiability eventually turns into the claim that a theory of meaning must be capable of being constructed and verified under conditions of radical interpretation, in which we try to interpret utterances of a completely unknown language without any

Meaning and truth

knowledge of what the speakers believe or desire. Its rationale is provided by the requirement of Non-Circularity. The empirical evidence on which a theory of meaning rests should be describable in terms which are as remote from the concept of meaning as possible. Davidson insists on finite axiomatization in order to explain what has come to be known as ‘semantic productivity’ or the ‘creativity of language’. As competent speakers of a natural language we can produce and understand sentences which we have never encountered before. Furthermore, the number of these sentences is potentially infinite. According to Davidson, if the theory contained an infinite number of axioms, one for each of the infinitely many sentences of L, it could not be mastered or acquired by finite creatures like us. What kind of semantic theory could satisfy these requirements? In answering this question, Davidson turns to three ideas that derive from Frege and the early Wittgenstein. The first is the principle of compositionality, according to which the meaning of a sentence is determined solely by the meaning of its constituents and by the way in which they are put together (their mode of combination). A compositionalist theory explains the creativity of language because it allows for finite axiomatization. Its meaning-giving theorems display the meaning of a sentence s as a function of the semantic properties of its lexical parts together with its syntactic structure. For this reason, it can get by with a finite set of axioms, namely one for each of finitely many ‘semantical primitives’, roughly speaking words, plus a relatively small number of recursive rules for their combination. This finite stock of axioms and rules is capable of generating a potentially infinite number of meaninggiving theorems. We can understand an unlimited number of unfamiliar sentences, because we are familiar with the words out of which these sentences are composed and with the ways in which the words have been put together. This explains not only how we can learn a language, but also what it is that we learn (ITI 3, 8–9, 18, 55–6, 202). The second idea is that sentences have semantic priority over words. Prima facie this contextualism is incompatible with compositionalism: if the meaning of a sentence is a function of the meanings of its components, the latter should be semantically prior. In fact, however, contextualism is the dialectical flip-side of compositionalism. In Davidson, compositionalism and contextualism jointly provide a rationale for the ‘holistic conception of meaning’, which in turn underlies the requirement of Extensional Adequacy.

145

146

Quine and Davidson on Language, Thought and Reality

If sentences depend for their meaning on their structure, and we understand the meaning of each item in the structure only as an abstraction from the totality of sentences in which it features, then we can give the meaning of any sentence (and word) only by giving the meaning of every sentence (and word) in the language. Frege said that only in the context of a sentence does a word have meaning; in the same vein he might have added that only in the context of the language does a sentence (and therefore a word) have meaning. (ITI 22, see 18)

As we have seen, Frege’s claim is hyperbolical. But it can be glossed in a defensible way which does not run counter to compositionalism. The meaning of a sentence is a function of the meaning of its constituents. But sentences are semantically prior because they constitute the unit of communication. Words in isolation do not communicate anything; hence their meaning consists in the systematic contribution they make to the meaning of the sentences in which they can occur (Strawson 1971: 171). According to Davidson, the meanings of words are abstracted from their role in sentences (ITI 220). One cannot understand an individual sentence without understanding the other sentences in which its components occur. But these other sentences involve new components, which in turn can only be understood by looking at the sentences in which they can occur, and so on. Ultimately, any word and any sentence can be understood only by understanding all words and all sentences. The axioms of a compositional semantics specify the meaning of words. But the theory is tested against empirical evidence in radical interpretation through its theorems. Indeed, that test ultimately applies not even to individual meaning-giving theorems, but to the whole lot. Davidson follows Quine in extending contextualism into a full-blown semantic holism. Both would subscribe to Wittgenstein’s famous dictum ‘To understand a sentence means to understand a language’ (1967a: §199), albeit for their own reasons. The third idea is that the theorems of a semantic theory give the meaning of a sentence s by specifying its truth-conditions. This is the key to Davidson’s truth-conditional semantics: ‘to give truth-conditions [necessary and sufficient conditions for its truth] is a way of giving the meaning of a sentence’; a theory of meaning is ‘a set of axioms that entail, for every sentence in the language, a statement of the conditions under which it is true’ (ITI 24, 56). By the same token, to understand a sentence is to grasp the conditions under which it is true. This idea can be traced back to Frege (1893: §32). It first becomes explicit in the Tractatus: ‘To understand a sentence means to know what is the case

Meaning and truth

if it is true’ (4.024). Yet Wittgenstein’s claim rests squarely on a repudiation of Frege’s idea that sentences are names of truth-values. There is a fundamental difference between names and sentences. To understand a name is to know the object it stands for. By contrast, sentences do not stand for anything. To understand a sentence we need not know whether it is true or false, but we do need to know the conditions under which it is true, that is, what is the case if it is true. We need not know what the world is like, but we must know what the world must be like if the sentence is true. In this respect, Davidson is closer to Wittgenstein than to Frege. He fervently denies that sentences are names and that the meaning of sentences are objects. However, in ‘Truth and Meaning’, he reaches a truthconditional conception of meaning by a route which bypasses both of these illustrious predecessors. Davidson starts out by considering what form the theorems of a theory satisfying his requirements should take. He first considers the suggestion that these theorems should be of the form (M) s means m

where ‘s’ denotes a sentence of L and ‘m’ its meaning. This suggestion reflects a referential account of meaning, in that such a theory pairs sentences with meanings understood as entities. Like Quine, Davidson rejects this account, but for slightly different reasons. He shares Quine’s hostility towards the idea that the meaning of an expression is an entity associated with it, towards properties as denotata of predicates, and towards propositions or states of affairs as denotata of sentences. Yet he insists: ‘My objection to meanings in the theory of meaning is not that they are abstract or that their identity conditions are obscure, but that they have no demonstrated use’ (ITI 21; see IA n3). The next type of meaning-giving theorem Davidson considers has the form (P) s means that p

with ‘s’ being replaced by the name of a sentence of L and ‘p’ by a sentence of the metalanguage which specifies its meaning. This proposal does not reify the meaning of sentences. Nevertheless, Davidson rejects it. Although ‘means that’ is a sentential operator rather than a function mapping objects on to objects, it creates an intensional context in which co-extensional expressions cannot always be substituted salva veritate. As a result, Davidson disparages it as ‘obscure’ (ITI 22–3; see Evnine 1991: 76–82; Platts 1979: 52–8).

147

148

Quine and Davidson on Language, Thought and Reality

Behind this familiar Quinean reaction one can detect three reasons. The first is a preference for purely extensional languages in logic which Davidson shares with Quine and logical positivism. This preference might be defended on pragmatic grounds. An extensional language (first-order predicate calculus) is used in Tarski’s theory, which Davidson hopes to apply to natural languages, whereas a logic that includes intensional constructions threatens to be fiendishly complicated. But this reason will be invalidated if it turns out that an extensional logic cannot do justice to all the intricacies of natural languages. The second reason is that a theory of meaning which employs theorems of form (P) is at odds with the requirement of non-circularity. ‘Means that’ looks suspiciously close to the notion of meaning to begin with. Worse still, it creates an intensional context. In such a context, expressions are substitutable salva veritate only if they are synonymous, that is, have the same meaning. This means that a P-theory tacitly presupposes the concept it purports to elucidate. Furthermore, and this is the third reason, such a provision may be ruled out, namely if the indeterminacy of translation establishes that there are no criteria of identity for intensions (ITI 101). Davidson concludes that a satisfactory theory of meaning should be extensional in two respects. First, it should not postulate intensional entities such as Fregean ‘senses’. Secondly, it should be couched in a language which does not contain intensional constructions or notions. Fregean theories of meaning distinguish between three features of sentences – reference, sense and force. Davidson sticks to reference, which is why he concedes that what he seeks is a theory of meaning in a ‘mildly perverse sense’ (ITI 24, 109). This extensionalism separates Davidson’s project from those truthconditional theories of meaning which follow Carnap in making liberal use of intensional notions like that of a possible world. Intensional semantics agrees with Quine on one point: a weighty concept of meaning can be captured only with the help of intensional notions. Davidson, by contrast, thinks that it is possible to provide a genuine theory of meaning without intensional notions, namely if one combines a truth-conditional conception of sentence meaning with Tarskian truth-theories and a theory of radical interpretation. Though extensionalist, his considered approach to meaning is neither reduction nor elimination but approximation: although meaning is not the same as truth-conditions, a theory of truth-conditions for L can do service as a theory of meaning, if knowing it will enable someone to interpret the utterances of L in the context of radical interpretation (see ITI 56n, 178–9).

Meaning and truth

To get beyond the intensional ‘means that’, Davidson takes two steps (ITI 23). The theorems are supposed ‘to give the meaning’ of the object language sentences. The obvious way of guaranteeing this is to use either the sentence itself, if the object language is contained in the metalanguage of the theory, or a translation of it. The ‘final bold step’ is to replace the intensional ‘means that’ of (P) by an extensional truth-functional connective. Since we are aiming at an equivalence, the biconditional ‘if and only if’ springs to mind. This would yield theorems of the form (Q) s if and only if p

Unfortunately, (Q) is ill formed. On the left-hand side of the biconditional we do not have a sentence, but the name of a sentence of L. Fortunately, we can turn this name into a sentence by attaching a predicate to it. Davidson introduces the dummy-predicate T to fulfil this role, thereby arriving at theorems of the form (T
) s is T if and only if p

with ‘s’ being replaced by the name of a sentence of L, and ‘p’ by that sentence or its translation. At last we are in sight of the Promised Land. It remains to specify the predicate ‘is T’. That predicate must satisfy just one condition, namely to yield all and only the true instances of schema (T
). So what predicate applies to ‘Snow is white’ iff snow is white, to ‘Snow is green’ iff snow is green, etc.? The answer is obvious: it is the truth-predicate. If ‘s’ names a true sentence, it is true that p, if ‘s’ names a false sentence, it is false that p. In either case the biconditional holds. Accordingly, the theorems of a theory of meaning should take the form (T) s is true if and only if p

But this is of course Tarski’s T-schema and its instances are our beloved T-sentences. Equally, the condition that a theory of meaning for L should yield all and only the true instances of this schema is nothing other than Convention T, Tarski’s criterion for the material adequacy of a definition of truth for L. The path to this point has been tortuous, but the conclusion may be stated simply: a theory of meaning for a language L shows ‘how the meanings of sentences depend on the meanings of words’ if it contains a (recursive) definition of truth-in-L . . . the concept of truth played no ostensible role in stating our original problem. That problem, upon refinement, led to the view that an adequate theory of meaning must

149

150

Quine and Davidson on Language, Thought and Reality

characterize a predicate meeting certain conditions. It was in the nature of a discovery that such a predicate would apply exactly to the true sentences. (ITI 23–4)

A Davidsonian theory of meaning for a natural language L yields theorems about the truth-conditions of sentences of L. The idea is that the sentence ‘p’ used on the right-hand side of a T-sentence reveals the meaning of the sentence ‘s’ quoted on the left-hand side. Furthermore, the theory uses a Tarski-style definition of ‘true in L’ to deliver these theorems. Such a definition makes do with an extensional language, and a finite number of axioms and rules of inference. It explains the compositional nature of meaning because it accounts ‘for the meaning (truth condition) of every sentence by analysing it as composed, in truth-relevant ways, of elements drawn from a finite stock’ (ITI 56). Davidson applies Tarski’s theory to natural languages, and more specifically to what their speakers say on a given occasion. As a result, he must contend with two features Tarski could ignore, namely that natural languages are spoken and that they contain indexicals (ITI 33–4, 131). An important consequence of accommodating these features is that he cannot simply speak of the truth-value of type-sentences, but must relativize truth-values to speakers and times. For this reason, Davidson’s semantics does not deliver standard T-sentences but relativized T-sentences like (1) ‘I am tired’ is true in English as spoken by x at t iff x is tired at t. (2) ‘Es regnet’ is true in German as spoken by x at t iff it is raining in the vicinity of x at t.

Originally, Davidson used the phrase ‘spoken (potentially)’ to provide truth-conditions for sentences that have never been uttered. But he dropped ‘potentially’ when he came to regard relativized T-sentences as natural laws, lawlike generalizations concerning some individual speaker or linguistic community which support counterfactual claims. Even without ‘potentially’, (1) and (2) are true generalizations of this sort (RD 21; ITI xiv). The modification of Convention T which accommodates this relativization runs as follows. A theory of meaning for a language L must entail all sentences that conform to the schema (T∗ ) s is true in L as spoken by x at t if and only if p

where x ranges over speakers of L, t over times, s over the metalinguistic names of sentences of L, and p over the translations of these sentences into the metalanguage.

Meaning and truth

3

Meaning and truth-conditions

Mainly as a result of Davidson’s efforts, truth-conditional semantics has acquired the status of an indisputable orthodoxy (e.g. Dummett 1978: xi–xii; Lewis 1983: 169). Davidson has never been so apodictic. He rests content with the claim that specifying truth-conditions ‘is a way of giving the meaning of a sentence’, and that ‘we know of no other way’ of doing so that satisfies his requirements on a theory of meaning (ITI 23–4, also IA 36). This is to his credit. For truth-conditional semantics raises serious questions. Davidson himself emphasizes three such questions (ITI 131): I. Can a Tarskian truth-theory be given for natural languages? II. Can such a theory be verified on the basis of evidence available to someone without prior knowledge of the interpreted language? III. Would knowledge of the theory suffice for interpreting utterances of the speakers of this language?

(II) will feature in the chapter on radical interpretation. (III) is part of a cluster of problems which concern the relation between a Davidsonian theory of meaning and actual linguistic communication, and will occupy us in chapter 9. Question (I) is discussed in the last section of the current chapter. But before we can turn to the question of whether an axiomatic truth-theory is feasible for natural languages, we must first scrutinize the very idea that the meaning of sentences is determined by their truthconditions, and the meaning of words by their contribution to the truthconditions of the sentences in which they occur. Davidson does not face up to this task. He ‘resuscitates meaning by administering the kiss of death’ (Hacking 1975: 179), namely by eschewing the notion in favour of those of truth-condition and interpretation. We are simply told that meaning is ‘what is invariant as between different acceptable theories of truth’ (ITI 225), yet without being told why meaning should be a matter of truth-conditions in the first place. Like Dummett (1993b: 1), Davidson hopes that analytic questions about the general concept of meaning will be resolved indirectly, through reflection on what a constructive theory of meaning for a particular language should look like (ITI xiii). On the one hand, this may help to avoid the deadlock which questions like ‘What is meaning?’ tend to create (see RD 83). On the other, the analytic questions about meaning Davidson ultimately purports to address are in danger of being swept under the carpet. For the ability

151

152

Quine and Davidson on Language, Thought and Reality

to provide a constructive theory of meaning does not ipso facto solve the analytic problem. Thus Heal (1978) has argued that it is one thing to know how to calculate the meaning of complex expressions of a particular language L from that of their components, quite another thing to know what ‘meaning’ means. Heal grants that the meaning of some terms can be specified through a recursive calculus (e.g. ‘natural number’ through Peano’s first two postulates). But she does not consider the possibility that ‘meaning’ or some of its cognates might belong to this group. Davidson might argue that terms like ‘meaningful’ can be defined recursively, namely by specifying the semantic primitives of the language and the operations through which meaningful expressions can be created from these primitives. This lacuna is easily filled, however, by adapting Dummett’s argument against Tarski’s theory of truth (chapter 4, section 2). If we want to know what it is for sentences to be true, it is not enough to be told what counts as being true in different languages. By the same token, if we want to know what it is for expressions to be meaningful, it is not enough to be told what counts as being meaningful in different languages. Indeed, reflection on the shape of constructive theories cannot replace the elucidation of the concept of meaning, but must be informed by it. The meaninggiving theorems delivered by a constructive theory must be such that what they specify is indeed the linguistic meaning of sentences, rather than, for example, their etymology or their aesthetic appeal. And to make sure of this, we need to rely on some understanding or other of the concept of meaning. Davidson’s reticence on this point is also surprising, since there are significant arguments in favour of linking meaning and truth-conditions. The first is Wittgenstein’s contrast between understanding a declarative sentence and knowing its truth-value, and the idea that even the former requires knowledge of what the world is like if the sentence is true. A second consideration is this. Non-synonymous expressions can be distinguished by seeing how their substitution for one another in a sentence can result in a shift from true to false. Conversely, if such substitution cannot lead to a change of truth-value, e.g. with ‘beseech’, ‘implore’ and ‘entreat’, it seems that these terms are synonymous (Rundle 1990: 147–8). The moral would be that substitution of non-synonymous terms alters the conditions under which a sentence is true. This in turn suggests that the meaning of words is at least partly determined by the effect they have on the truth-conditions of the sentences in which they can occur.

Meaning and truth

In pursuing these connections between meaning and truth, we must keep in mind two points from our previous discussion. One is that truth is neither a feature of sentences nor a relation between sentences, speakers and times. Consequently, both standard and relativized T-sentences are misleading, and one cannot identify the meaning of a sentence with its truth-conditions, for the simple reason that there are no conditions under which sentences are true. But, as mentioned above, this is not essential to Davidson’s project (ITI 56n). He has no reason in principle to resist the following reformulation of truth-conditional semantics. We must distinguish between meaning as a feature of type-sentences and truth as a feature of what is or could be said by using tokens of such types on particular occasions. Still, the meaning of a type-sentence might be determined by the conditions under which someone who utters a token of that type says something true. If so, we illuminate the meaning of a sentence by stating under what conditions an utterance of it will say something true (Strawson 1971: 179–80). By this token, the theorems of a truth-conditional theory should take the following form (T#) What s says when uttered by x at t is true iff p

as in (3) What ‘I am tired’ says when uttered by x at t is true iff x is tired at t.

The second legacy concerns the minimalist account of truth. Both proponents and opponents tend to assume that such an account is incompatible with a truth-conditional theory of meaning (PCT 9; Dummett 1978: xxi, ch. 1; Horwich 1998: §§22, 32). They reason as follows. Our knowledge of equivalences like disquotational T-sentences cannot simultaneously constitute both our knowledge of the meaning of the quoted sentence and our grasp of what it is for that sentence to be true. We can use T-sentences to explain the meaning of object language sentences only if we have a grasp of the concept of truth that is independent of these T-sentences. The minimalism I condoned in chapter 4, section 5 avoids this difficulty, since it explains truth through a generalization rather than particular equivalences. But another incompatibility seems to loom. The generalization presupposes the notion of a proposition; yet propositions are supposed to be sentence meanings; and if sentence meanings can only be elucidated by a truth-conditional semantics, then minimalism is circular. Both Quine and Davidson accept the assumption behind this line of

153

154

Quine and Davidson on Language, Thought and Reality

reasoning: propositions, if they exist, will not just be the bearers of truth and the objects of propositional attitudes, but also the meanings of sentences (WO 201–6; PL ch. 1; PT 77; SS 77–8 and SCT 324; UT 10; SIO 45). But that assumption is wrong. Propositions, in the sense of what is or could be said, are what we convey by the use of sentences, not their meanings. Unlike the meaning of a sentence, for example, what is said (believed, etc.) can be true or false, implausible or exaggerated (White 1970: 14; Hacker 1996a: n8). Far from being identical with sentence meanings, what is said on a particular occasion depends on sentence meaning and context of utterance. It may seem that there is yet another circularity here, if the meaning of a sentence is in turn determined by the conditions under which what is said by an utterance of it is true. But this impression is misleading. For sentence meaning might be determined not by the condition under which a specific utterance of the sentence is true, but by a general rule specifying under what conditions utterances of the sentence will be true. Furthermore, if there is a circularity here, it afflicts not just truth-conditional theories, but equally the view that the meaning of a sentence depends directly on what is said by an utterance of one of its tokens. In any event, in this chapter I shall not assume either that there is an irreconcilable conflict between minimalist conceptions of truth and truthconditional conceptions of meaning or that such a conflict would have to be resolved in favour of minimalism. Davidson’s truth-conditional semantics faces a host of more immediate problems. Two of them – the extensionality problem and the triviality of T-sentences – will be discussed later, since they raise issues involving radical interpretation. In this section, I shall instead discuss the immanence of meaning, the concept of truthconditions, and the individuation problem. 4

The immanence of meaning

According to Davidson, the schema (P) s means that p

is obscure because of the intensional nature of ‘means that’. But the real problem with (P) lies somewhere else. As Rundle points out, it is odd to say of a sentence s that it means that p, because ‘means that’ specifies a consequence. The noise coming from my bike may mean that the chain is rusty, but sentences like ‘The chain is rusty’ or ‘Die Kette ist rostig’ do not mean that the chain is rusty (Rundle 1979: §53; 1990: 186–9). If we want to

Meaning and truth

specify the meaning of a sentence, we shall employ different connectives, for example ‘means’ or ‘means the same as’, as in (4) ‘Die Kette ist rostig’ means ‘The chain is rusty’. (4
) ‘Die Kette ist rostig’ means the same as ‘The chain is rusty’.

This is not just a trivial feature. Standard explanations of meaning like (4) and (4
) link two forms of words, rather than a form of words and something extra-linguistic, whether it be a meaning or a truth-condition. Proponents of both P-sentences and T-sentences would protest that ultimately explanations of meaning must go beyond the ‘translational semantics’ on display in (4) and (4
), since otherwise all explanations would presuppose a prior understanding of the sentences quoted on the righthand side, which would engender an infinite regress (Evans and McDowell 1976, vii–xi; Dummett 1993b: 6–7). This objection alerts us to an important difference between (4) and (4
). A monolingual Spaniard can know what (4
) expresses, but she cannot know what (4) expresses. By contrast, a monolingual German cannot know what (4
) expresses without also knowing what (4) expresses, simply because she understands ‘Die Kette ist rostig’. This last point also suggests, however, that (4), unlike (4
), is as direct a specification of meaning as we should hope for. Of course, it will only explain ‘Die Kette ist rostig’ to someone who already understands ‘The chain is rusty’. But any genuine explanation of meaning must presuppose some prior understanding on the part of its recipient, which is why the ultimate entry into language is through training the neophyte to use expressions in appropriate circumstances (see chapter 8, section 2). 5

The notion of a truth-condition

T-sentences like (5) ‘Schnee ist weiß’ is true iff snow is white

have the appropriate form for stating a condition, since they feature a biconditional. Baker and Hacker have argued, however, that the truthconditions they purport to ascribe lack certain features which are essential to the very idea of a condition (Baker and Hacker 1984: 190–7). First, they do not specify conditions for the sentence quoted on the left-hand side having the property of being true, for the simple reason that the righthand side does not even talk about that sentence. The right-hand side of (5) does not state what the sentence ‘Schnee ist weiß’ must be like in order to be true. Secondly, conditions must be general. The conditions for

155

156

Quine and Davidson on Language, Thought and Reality

making a valid will, for example, hold indifferently for all, including those who fail to meet them. By contrast, T-sentences apply only to individual sentences from the object language, namely those named on the left-hand side. These objections are not conclusive. As regards the second, it is true that (5) only states a condition for an individual sentence. But at least that condition, namely snow being white, is also a necessary and sufficient condition for the truth of some other German sentences, e.g. ‘Schnee hat dieselbe Farbe wie Kokain’. The real stumbling-block is the first point. Snow being white may be a condition for the truth of that sentence, as it is for the truth of ‘Schnee ist weiß’. But it is not a condition that either of these sentences themselves could fulfil in order to be true. This contrasts with what I take to be an adequate, if minimalist, statement of the condition under which sentences in general are true: A sentence s is true iff s states how things are. Here we have a condition that a sentence must meet in order to be true. T-sentences do not state conditions in this sense. But there is a less demanding sense. For example, the condition for winning a game by default is not that the game possesses such-and-such properties, but that one of the contestants fails to show up. Similarly, ‘Schnee ist weiß’ and ‘Schnee hat dieselbe Farbe wie Kokain’ are both true on condition that snow is white. 6

The individuation problem

One question facing truth-conditional semantics is whether truthconditions individuate sentence meaning finely enough. Intensional semantics holds that two sentences have the same truth-conditions or meaning iff they are true in the same possible worlds. But the necessary truths of mathematics, for example, are true in precisely the same possible worlds, namely all, and therefore they would all have to mean the same. Davidson can avoid this unpalatable consequence by insisting that two sentences of L have the same meaning or truth-conditions just in case that, in an adequate theory of meaning for L, the right-hand sides of their T-sentences are identical. Since T-sentences reflect both the components and the structure of sentences, mathematical truths about ellipses would not have the same meaning as mathematical truths about prime numbers. This solution may still ascribe identical truth-conditions to some logical truths, for example tautologies involving the same elementary propositions. The T-sentences for ‘Fa → Fa’ and ‘(Fa ∨ ∼Fa)’, for example, contain the same right-hand side, namely ‘a satisfies F or a does not

Meaning and truth

satisfy F’. But, following the Tractatus, it may be possible to argue that these tautologies, at any rate, do have the same content. They all say the same, namely nothing, about the same things, namely a and F. More troubling is the fact that this solution implies that no two sentences of a language can be synonymous if they differ in either structure or constituents (Baker and Hacker 1984: 209, 327–39). For example, it implies that ‘Tully was a Roman orator’ and ‘Cicero was a Roman public speaker’ are not synonymous, because their respective T-sentences differ: (8) ‘Tully was a Roman orator’ is true iff Tully was a Roman orator. (9) ‘Cicero was a Roman public speaker’ is true iff Cicero was a Roman public speaker.

The metalanguage sentences on the right-hand side, though not identical, are synonymous. Yet Davidson could not invoke this fact to account for the synonymy of the object language sentences on the left-hand side, since this would engender a regress. Nor could he insist that they are synonymous because they ascribe the same property to the same individual, since this would drag in the intensional notions he seeks to avoid (ITI 205–6). The problem could be overcome by insisting that an adequate truththeory does not just feature disquotational axioms, but also axioms like ‘Tully’ refers to Cicero x satisfies ‘is an orator’ iff x is a public speaker.

For in that case, one can derive for both of our sentences T-sentences with the same right-hand side, namely that of (9). But that suggestion runs counter to Davidson’s idea that a theory of meaning should be modest (see chapter 8, section 2). 7

Truth-conditional semantics and natural languages

A theory of meaning for a natural language is an attempt to describe a complex practice in terms of an artificial formal system. In Davidson’s case, this system is confined, moreover, to a Tarskian truth-theory, a calculus which is austere in that it avoids intensional notions and is relatively simple in its fundamental structure. However, natural languages are ‘not made to match logic’s ruler’, to use Frege’s words. Tarski himself regarded them as beyond the ken of formal semantics, because they contain semantic notions and are not formally specifiable. In particular, they contain a host of constructions that seem to resist formalization through the first-order predicate calculus of a truth-theory.

157

158

Quine and Davidson on Language, Thought and Reality

Though far from complacent, Davidson invests moderate optimism in the programme of showing that a truth-theory can accommodate these ‘recalcitrant idioms’ after all, and hence serve as a theory of meaning for natural languages (ITI xv–xvi; IA 36). Among these recalcitrant idioms are adverbial modifications and ascriptions of propositional attitudes, which I shall mention later on (chapter 8). In this section I want to deal with two other obstacles to what is known as ‘Davidson’s Program’: the divergence between the logical constants and their alleged equivalents in natural languages, and non-declarative sentences. One should expect that the axioms of a Davidsonian theory, by ‘giving the meaning’ of the object language terms, explain the meanings of those terms. But there is a discrepancy between the standard explanations of natural language terms that one finds, in dictionaries, for instance, and those afforded by a Davidsonian theory. As we have seen, with respect to singular terms and predicates the question arises whether the explanations provided by Tarskian axioms are vacuous. The logical constants create a different problem. Their standard semantic explanations do not fit the propositional connectives and quantifiers of ordinary languages, which they purport to elucidate (Strawson 1952: III.2; Baker and Hacker 1984: 170–4). For example, it seems that the meaning of ‘if p then q’ differs from that of ‘p → q’ in that it requires non-truth-functional grounds for excluding the possibility that p is true yet q false. Since the logical constants indicate the logical form of propositions, this casts doubt on one of the central claims Davidson makes on behalf of a truth-theory, namely that it reveals the underlying logical structure of natural languages. Along with most formal semanticists, Quine and Davidson would reply as follows. This kind of difference is pragmatic rather than semantic; it concerns not the meaning of the logical constants but merely their everyday application, and in particular the conversational implicatures highlighted by Grice (1989). ‘If p then q’, for example, is semantically equivalent to ‘p → q’ and hence to ‘∼p ∨ q’. But to affirm the bare alternative is less informative than a straightforward denial of p or a straightforward assertion of q. So, if someone who utters ‘If p then q’ sticks to the conversational maxim of being as informative as possible, we take it as implied that she is uncertain as to the individual truth-values of antecedent and consequent, yet has non-truth-functional evidence about a connection between them (ML 15–6; NDE 437). This course can be challenged at two levels. Does it do justice to specific propositional connectives? And is there a non-question begging

Meaning and truth

distinction between semantics and pragmatics according to which all those features of use ignored by the predicate calculus fall on the side of pragmatics? (Strawson 1997: ch. 9; Glock 1996a; Travis 1997). Even if these objections are waived, the ‘merely pragmatic’ gambit should not appeal to Davidson. After all, he seeks a theory grasp of which would enable someone to participate in ordinary communication, and in particular to understand what speakers say through their utterances on a particular occasion. From this perspective, relegating the indisputable disanalogies to pragmatics is in vain. If one must grasp the distinctive features of the ordinary terms to be able to communicate, the result will merely be that semantic knowledge falls short of communicative competence, and must be supplemented by pragmatic knowledge. Take an expression which seems relatively favourable to formal treatment, namely ‘and’. Its alleged equivalent ‘&’ is commutative, i.e. ‘p & q’ is semantically equivalent to ‘q & p’. Yet this does not hold of ‘and’. Someone who does not recognize the difference between ‘She kicked him and he fell to the floor’ and ‘He fell to the floor and she kicked him’ will be unable to understand what anglophones say in uttering these sentences. Describing his failure as merely pragmatic does nothing to alter the fact that such a person is not a competent speaker of English. He is ignorant of a general linguistic difference, one that depends neither on the particular circumstances in which these sentences may be uttered, nor on the intentions of the utterer. This point militates against the idea that the meaning of, for example, English connectives and quantifiers and hence the logical form of English sentences are captured by a Tarskian truth-theory. It need not militate against linking the meaning of words to the truth-conditions of sentences in which they occur. One might argue, for example, that the aforementioned English sentences differ in meaning precisely because what people say in uttering the first is true only on condition that the kicking preceded the fall, while what they say in uttering the second is true only on condition that the fall preceded the kicking. The second obstacle, however, challenges the very core of truthconditional semantics. Natural languages feature sentences that are not declarative, for example imperatives, interrogatives and optatives. Evidently, a comprehensive theory of meaning must be capable of specifying the meaning of these sentences. However, since Aristotle it is known that non-declarative sentences do not in general express something which is either true or false. It is absurd to assess ‘Where is today’s Guardian?’

159

160

Quine and Davidson on Language, Thought and Reality

or ‘Turn off the light!’ for their truth or falsity. It is equally absurd to mention these sentences on the left-hand side of a T-sentence. If nondeclaratives lack a truth-value, they also lack truth-conditions. Consequently, the meaning of these sentences cannot simply consist in their truth-conditions, and they seem to defy truth-conditional semantics. Unless a truth theory can accommodate these non-declarative sentences, ‘it is inadequate as a general theory of language’ (ITI 116). The standard reaction to this difficulty is to distinguish two aspects of sentence meaning, and to deal with them in separate divisions of the overall theory of meaning. Following Frege, one distinguishes the sense of a sentence from its force. For example, uttering the three sentences (10) The door is open. (11) Open the door! (12) Is the door open?

effects different linguistic acts: (10) an assertion, (11) a command, and (12) a (yes–no) question. At the same time, these sentences also have a common semantic component, namely the sense or propositional content that the door is open. Like every complete sentence, (10)–(12) can be decomposed into two components: a ‘sense conveying component’ or ‘propositional content clause’ (‘that the door is open’), and a ‘force indicator’ or ‘mood indicator’ which specifies that (10) is declarative, (11) imperative, and (12) interrogative.1 Allegedly, the propositional content does have a truth-value, and hence a truth-condition, even when it occurs in non-declaratives. As a result, the ‘core part’ of the theory of meaning, the theory of sense, can deal with the propositional contents of sentences by specifying their truth-conditions. As the ‘supplementary part’ of the theory of meaning, the theory of force investigates the larger context of linguistic practice, the literal use of nondeclarative sentences, and perhaps the non-literal use of declarative sentences. In the former capacity, it will specify both what it is to perform different speech acts, and what speech acts are performed by sentences of different grammatical form in different contexts. In combination with a theory of sense, a theory of force will allow one to describe any utterance of the language as a particular speech act, that is, as an act with both a 1. Davidson presents this as a difference in mood. However, that term classifies sentences according to the form of the main verb. Yet the main verb in a declarative sentence may be in the subjunctive, and in an interrogative sentence it is often in the indicative (Rundle ¨ 1990: 84; Kunne 1993: 9). Nevertheless, for the sake of brevity I shall not depart from the now-established misuse.

Meaning and truth

particular illocutionary force and a specific propositional content, for example as an act of asserting that the door is open, asking whether the sun is out or enjoining that the door be closed. According to its proponents, this combination manages to systematically relate the sense and the force of a sentence to its use in communication (Dummett 1993b: 38–41, 219–20; Wiggins 1997: 10–11; McDowell 1998: ch. 1). Davidson steers a different course. He has two reasons for doing so. The first is his scepticism about developing the required theory of force. Proponents of such a theory maintain that there is a conventional relation between the mood of sentences and the force of utterances. Roughly speaking, mood plus context determines illocutionary force. For example, an assertion is a declarative sentence uttered under conditions specified by conventions. Davidson, by contrast, believes that only mood, not force, is a ‘conventional feature of sentences’ that contributes to their meaning (ITI 114–15, 117). There is at best a loose connection between them, since there are many utterances whose force belies their mood. (13) Did you notice that Joan is wearing her purple hat again? (14) I’d like to know your telephone number. (15) In this house we remove our shoes before entering.

Although (13) is interrogative, it can be used to make an assertion, and although (14) and (15) are declarative, they can be used, respectively, to ask a question or to issue a command (ITI 110; similarly Wittgenstein 1967a: §§23–4). This objection is problematic. Obviously, there is no one-to-one correlation between mood and force, for the simple reason that there are only a few moods in English, but an indefinite number of different speech acts. However, this leaves open the possibility of a conventional one–many correlation. In denying this, Davidson seems to confound the force of an utterance with its point, especially the information the speaker intends it to communicate (Rundle 1990: 87–9; Dummett 1993b: 209–12). Although (13) implies that Joan is wearing a purple hat (contrast ‘Did you think that . . .?’), it nevertheless remains a question that can be answered by ‘No, I didn’t’. Equally, (14) and (15) remain statements or assertions, even though the first implies a request and the second a command. Though powerful, this response is not entirely convincing either. Someone who responds to (14) by saying ‘How very interesting, are there other things you are curious about?’ has not just missed the extralinguistic point of the utterance; he is not, for example, in the position

161

162

Quine and Davidson on Language, Thought and Reality

of someone who wonders ‘Why is she asking for my telephone number?’ Rather, he is either making a linguistic joke or ignorant of a linguistic point. This ignorance, moreover, concerns a convention, namely that ‘I should like to ’ can be used to make a request, provided the speaker’s ing is something the hearer can facilitate. In my view, therefore, there is a conventional relation between the form of words and the speech acts they can be used to perform, even though it is more complex than sense/force theories tend to assume. Davidson’s second argument against the sense/force distinction appeals to what he calls ‘the autonomy of (linguistic) meaning’ (ITI 113–14, 164–5, 274–5): the ‘ulterior purpose’ of an utterance, the ‘extra-linguistic ends’ to which it is used, is independent of its ‘literal meaning’. This independence, Davidson believes, rules out that uttering a sentence of a certain grammatical form under ‘conventionally specified circumstances’ could ever suffice for executing a particular speech act. For any such purported conventional guarantee would have to be capable of being put into words. It would have to be possible, for example, to introduce a super-assertion sign the use of which guarantees that an assertion is made. But any such sign could be used to perform a different speech act. ‘Every joker, storyteller, and actor will immediately take advantage of the strengthened mood to simulate assertion . . . Mood is not a conventional sign of assertion or command because nothing is, or could be, a conventional sign of assertion or command’ (ITI 113–14; see Stainton 1996: 192–4). This passage rightly stresses that human beings can put words to extralinguistic work in ways that transcend conventional meaning. But it assumes rather than shows that the illocutionary force of an utterance is among the extra-linguistic features of utterances, that is, that it is determined by the speaker’s intentions rather than linguistic conventions. If there is a distinction between the force of an utterance and its point, contrary to Davidson’s first argument, one can insist that an utterance featuring the super-assertion sign ipso facto has the force of an assertion, whatever the utterer’s intentions. The same goes for our actual, and fallible, indicators of illocutionary force. The utterance of a joker is no less an assertion for not being meant in earnest. If her intent is obvious, we may not hold her responsible for its content. Nevertheless, we can and must characterize certain tongue-in-cheek remarks as assertions. Storytelling and acting, for their part, are forms of discourse which are exempt from certain conventions of ordinary speech, but these exemptions are themselves governed by conventions.

Meaning and truth

Davidson’s alternative to the sense/force distinction is fuelled by his ambition of reducing the three aspects of sentence meaning distinguished by Fregean theories – reference, sense and force – to a single one, namely reference (ITI 109). He wants to bring the differences between moods – the only aspect of force he considers semantically relevant – within the scope of a Tarskian truth-theory. At the same time, he wants to honour the fact that sentences in certain moods do not have a truth-value. As a result, he maintains that a satisfactory theory of mood should satisfy three conditions (ITI 115–16): A. It must capture what is common between declaratives and sentences in other moods. B. It must show what is semantically different between sentences in different moods. C. It must deal with non-declaratives through the resources of a truth-theory.

(C) suggests that the elements mentioned in (A) and (B) – content clause and mood indicator – must be combined through a truth-function. However, the result of truth-functional combination is a sentence with a truthvalue, which would exclude non-declaratives. Davidson tries to solve this conundrum by appealing to a paratactic construction. Non-declarative sentences which seem to defy a Tarskian truth-theory are analysed not into a ‘sentential operator’ working on a content clause, but into two complete sentences or sayings. Both of them are amenable to a truth-theory, yet they are combined in a demonstrative rather than truth-functional way. Thus (11) turns into (11
) My next utterance is imperatival in force: The door is open

and (12) into (12
) My next utterance is interrogative in force: The door is open.

The first sentence in these paraphrases is the ‘mood setter’, the second the ‘indicative core’. Thus the common constituent of different moods required by condition (A) is simply the indicative sentence itself, while the mood setter provides the distinguishing feature required by condition (B). Condition (C) is satisfied because each of the two component utterances has a truth-value. We can ‘give the semantics’ of a non-declarative sentence through two specifications of truth-conditions. Understanding (11), for example, amounts to knowing under what conditions its mood indicator is

163

164

Quine and Davidson on Language, Thought and Reality

true and under what conditions its indicative core is true, even though there is no such thing as knowing under what conditions (11) itself is true. Davidson’s proposal is ingenious and attractively simple, since its taxonomy of illocutionary forces is no richer than the taxonomy of grammatical mood. Alas, these advantages are purchased at an exorbitant price. Davidson concedes that, syntactically speaking, non-declarative sentences cannot be viewed as consisting of two declarative sentences. Nevertheless, he is committed to the idea that utterances of non-declaratives ‘must semantically be viewed as consisting in two utterances’. In uttering a nondeclarative sentence, we simultaneously perform two speech acts, just as one can simultaneously rub one’s stomach with one hand and pat one’s head with the other (ITI 119–21). Unfortunately, this analogy fails: one can rub one’s stomach without patting one’s head, but one cannot utter the mood indicator as an independent linguistic act, if only because of its cataphoric dimension (Kunne ¨ 1993: 10). Furthermore, Davidson’s analysis yields intolerable results when applied to his own counter-examples to the sense/force distinction. If the mood of (13) in fact belies its force, as Davidson has it, it cannot be analysed into: (13
) My next utterance is interrogative in force: You noticed that Jane is wearing her purple hat again

since that would mean that (13) is a question after all. Nor can Davidson opt for (13∗ ) My next utterance is assertive in force: You noticed that Jane is wearing her purple hat again

since this is simply not what (13) asserted. Finally, if he opts for (13#) My next utterance is assertive in force: Jane is wearing her purple hat again

this would make (13) indistinguishable from ‘Jane is wearing her purple hat again.’ We would lose sight of the fact that the speaker of (13) uses an interrogative sentence to make an assertion (see Dummett 1993b: 208–9). Next, Davidson cannot simply legislate that by using the mood setter, speakers ‘characterize an utterance as having a certain force’ without asserting it (ITI 120–1). Unlike predication, characterization carries an assertive commitment. In that case, however, the mood setter itself must in turn be decomposed into a core and a mood setter, which would

Meaning and truth

engender an infinite regress. The only way of avoiding this regress is to treat the indicative not as a mood among others, but as the absence of any particular mood. Davidson inclines towards this solution, but it leaves the status of declaratives unexplained and is at odds with the fact that declarative form is a prima facie indicator of assertion (Dummett 1993b: 205–8). Last but not least, it is simply wrong to maintain that the cores which follow the mood setters of (11
) and (12
) are, respectively, imperatival and interrogative (Baker and Hacker 1984: 98–9). These forms of words are declarative, which is precisely why they possess truth-conditions. Davidson would presumably reply that neither mood setter nor indicative core is to be understood in the ordinary way here. But in that case he owes us an explanation of how they are to be understood. Worse, any explanation which rehabilitates the mood setters entails that the cores do not possess truth-conditions after all, and thereby defies the purpose of the whole exercise. It is fair to conclude that Davidson’s paratactic account of moods is untenable. This scuppers his plan of reducing illocutionary force to truthconditions and the theory of force to the theory of reference. Furthermore, if Davidson is right in thinking that the standard sense/force distinction is equally untenable (as do Baker and Hacker 1984: chs. 2–3), then nondeclarative sentences and non-assertoric speech acts mark a stumblingstone not just for his extensionalist version, but for truth-conditional theories of meaning in general. In any event, the first problem facing Davidson’s project, namely the applicability of a Tarskian truth-theory to natural languages, turns out to be as serious as Tarski himself supposed. We have also seen that Davidson’s theory faces problems over the individuation of sentence meaning, and that the explanations of sentence and word meaning it favours differ in form and content from the corresponding explanations given in ordinary discourse. Before returning to these and other difficulties, I shall tackle the idea of radical translation/interpretation, which is supposed to give empirical content to Davidsonian theories of meaning, and to remove some of the obstacles in their path.

165

6

Radical translation and radical interpretation

Quine’s term ‘radical translation’ refers to the translation of a completely unknown language with no historical or cultural links to familiar languages. Radical translation can neither assume any prior understanding of the language in question nor resort to a bilingual interpreter. Similarly, Davidson’s ‘radical interpretation’ is the attempt to understand an alien community without the benefit of any prior knowledge of what its members think or of what their utterances mean. As a result of the work of Quine and Davidson, it is now commonplace to find philosophers discussing exotic anthropological scenarios, cases in which we encounter an isolated and completely alien tribe and try to understand its language and activities. Quine’s first explicit treatment of radical translation appeared in a collection entitled On Translation (1958). This may be responsible for Rorty’s claim that Quine and Davidson develop ‘the philosophy of language of the field linguist’, on whose ‘mundane standpoint’ their whole philosophy is based (1986: 339). Other commentators have complained that radical translation and interpretation have little connection to the actual procedures field-linguists, anthropologists and archaeologists employ in interpreting unknown languages, practices or texts. One might also wonder how such a special phenomenon should come to be the focus of so much philosophical attention and the linchpin of two important philosophical systems. Why base one’s philosophy on the standpoint of the field-linguist, and not on that of the priest, the physicist, the artist or the logician – to mention just a few figures that have inspired philosophers in the past? And if these are not mundane enough, why not take the Oxford exit and stick to the person on the Clapham omnibus?

[166]

Radical translation and radical interpretation

Such questions rest on a mistaken assumption. Quine’s discussion of radical translation does not arise from an intrinsic interest in fieldlinguistic translation, but from his attack on the analytic/synthetic distinction. One prong of this attack (the only one still in the running if chapter 3 is correct) is that there are no criteria of identity for meanings. We cannot answer questions like ‘Does this expression mean the same as that one?’, since we lack principles for individuating meanings. In the absence of such principles, intensional notions are vacuous and intensional discourse is non-factual (OR 19–20). Quine soon realized that this line of reasoning commits him to a discussion of translation. Translation is an explicit and systematic attempt to establish relations of synonymy. Furthermore, unlike establishing synonyms within a language – as in a Thesaurus – it is an empirical enterprise, since it cannot rely on the intuitions of competent speakers (FLPV 60–4). The matter gained urgency when Quine’s ‘dogmatic’ opponents – Carnap (1956: Apd. D), Grice and Strawson (1956: 156–8) – defended the analytic/synthetic distinction by insisting that there are ways of establishing whether two linguistic expressions mean the same. This forced Quine to return to translation in Word and Object, with the aim of discrediting synonymy. He reached the desired result through a discussion of what he came to call radical translation. By considering cases in which we start completely from scratch, he argued that translation is ‘indeterminate’. There is, after all, no ‘fact of the matter’ as to whether two linguistic expressions are synonymous, and hence nothing for a translator to be right or wrong about. Indeterminacy of translation is supposed to furnish the final nail in the coffin of intensional notions. Accordingly, Rorty has things back to front. Far from deriving their philosophies from the perspective of the field-linguist, Quine and Davidson use the idea of anthropological encounters as a ‘thought experiment’ (FM 167; ITA 5; PT 38) or ‘conceptual exercise’ (UT 12; SCT 325). The purpose of the heuristic device is to enforce the austere approach to language and meaning that Quine and Davidson deem proper. Such an approach does not rely on concepts that are semantic and/or intensional; but it does take into account all the physical and behavioural facts which, according to them, underlie linguistic communication and constitute meaning properly conceived. In the absence of any prior information, the radical translator is reduced, according to Quine, to observing the causal connections between the natives’ environment and their dispositions to verbal behaviour. If these ultimate data are too poor to determine what

167

168

Quine and Davidson on Language, Thought and Reality

their utterances mean, then the traditional notion of meaning has no fundamentum in res: ‘it is only radical translation that exposes the poverty of ultimate data for the identification of meanings’ (PT 46). Mutatis mutandis for Davidson. His ambition is to construct a theory of meaning for natural languages which does not assume semantic knowledge. Radical interpretation seems to guarantee this. At the outset the interpreter is in no position to describe the natives’ behaviour in semantic terms, because he is ignorant of what their utterances mean. At the end, it is hoped, he will have a theory that assigns meaning (truth-conditions) to these utterances. Radical interpretation addresses the question: ‘What would it suffice an interpreter to know in order to understand the speaker of an alien language, and how could he come to know it?’ (RD 83; see ITI 179). Quine and Davidson have used ‘radical translation’ and ‘radical interpretation’ in two different ways (see WO 28; OR 45 and RD 77n). One is to denote a task, namely of understanding the actions and utterances of an alien community, without any prior knowledge. The other is to denote their prescribed procedure for solving that task. On the first construal, radical translation/interpretation is interpretation from scratch, a task humans (explorers, conquerors, missionaries and their victims) have faced and mastered numerous times. On the second, radical translation/interpretation is interpretation which relies on the kind of evidence and methods recognized by Quine and Davidson. In fact, one should distinguish three cases. Normal translation/interpretation is the process of establishing the sense of utterances that are either from a foreign language or obscure for other reasons. Translation from scratch is translation without prior semantic knowledge or the aid of a bilingual. Radical translation proper is not just from scratch, but purports to be presuppositionless. It does not rely on any assumptions about what the natives think and do, and it uses only information of an austere (non-intensional and non-semantic) kind. Radical translation/interpretation has acquired an overriding importance even among many philosophers who reject details of Quine’s and Davidson’s approach. The motives are threefold. The first is epitomized by Quine’s slogan that ‘radical translation starts at home’. It is often assumed that all linguistic understanding is based on radical translation: even our understanding of straightforward utterances of our own language relies ultimately on the evidence and methods available to a radical translator. The same is then supposed to hold of infants who acquire a first language.

Radical translation and radical interpretation

Others who set store by radical interpretation/translation are not committed to the idea that we actually engage in it when we speak or learn our mother tongue. Some of them are motivated by Quinean eliminativism. If successful, the indeterminacy thesis would not just remove the analytic/synthetic distinction and its metaphilosophical corollaries, but also the need to accommodate semantic phenomena such as meaning within a naturalistic or more specifically physicalist world-view. If Quine is right, there is simply no fact of the matter as to what our words mean. Therefore it is no embarrassment to naturalism if phenomena like meaning cannot be captured by the methods and concepts of the physical sciences. The final motive is Davidsonian rather than Quinean. It is precisely the hope of accommodating apparently recalcitrant phenomena like meaning and thought within a naturalist framework. If radical translation/ interpretation is possible even in principle, this will show how our rich intensional and semantic concepts and statements can be derived from, albeit not strictly reduced to, something more basic. Semantic and mental phenomena may be real, yet they are ultimately constituted by appropriately complex patterns of non-semantic and non-mental phenomena. In the following three chapters I shall take issue with all these motives. The first ignores not only the differences between translation/ interpretation, ordinary communication and language-acquisition, but also our first-person authority concerning speaker’s meaning. The second motive is flawed because the indeterminacy diagnosed by Quine and Davidson is not intrinsic to meaning, but a result of their approach to it. For one thing, they go wrong in thinking that all semantically relevant features can be gathered from dispositions to assent to sentences, and in sidelining the reference of sentence components. For another, the insistence on a presuppositionless approach is no more legitimate in the case of interpretation than it is in the case of investigating the physical world. The final motive is based on a futile hope. Radical interpretation does not solve the extensionality problem confronting Davidson’s theory of meaning. Nor does it provide even a feasible basis for interpretation from scratch. The procedures of Quine and Davidson are either unsuited to the task, or assume too much to be truly radical. This chapter is mainly devoted to radical translation and interpretation as hermeneutic models for interpretation from scratch. I start by sketching Quine’s radical translation and his celebrated thesis that it is indeterminate. Section 2 goes on to claim that Quinean translation cannot even reach

169

170

Quine and Davidson on Language, Thought and Reality

the meagre results it countenances without tacitly relying on hermeneutic methods, anthropological assumptions and intensional notions – notably that of assent – which he explicitly condemns. Next I describe how radical interpretation differs from radical translation. Davidson abandons Quine’s austere behaviourism in recognizing the intentional nature of linguistic behaviour and resisting the idea that meaning is tied to neural stimulations. But he mistakenly follows Quine in restricting the evidence to the holding-true or preferring-true of sentences (section 3). In the final section, I shall critically examine Davidson’s claim that his principle of charity is a precondition of interpretation. That principle embodies a fundamental insight, namely that all interpretation must take certain things for granted. However, these presuppositions differ from charity in that they include substantial assumptions about human beings, a minimal hermeneutic anthropology. In interpretation from scratch we lack specific information about what our interlocutors believe, mean and do. But we can and must take for granted that they share with us certain needs, cognitive capacities and patterns of behaviour, including certain kinds of speech acts. This militates against the Davidsonian hope of deriving intensional notions from something significantly more basic. In the next chapter, I shall argue that it also undermines the Quinean attempt to debunk intensional notions. Once we abandon the flawed ideal of presuppositionless interpretation, the indeterminacy of translation can be resisted. Neither Quine nor Davidson manages to remove the obstacle of meaning from the path of naturalism. 1

Translation runs amok

Quine’s radical translator translates from a completely alien ‘source language’, sometimes called Jungle, into a ‘target’ or ‘background’ language, usually English (ITA 5). This he does by constructing a ‘translation manual’. Unlike a bilingual dictionary, such a manual translates sentences rather than words. Nevertheless, it is not simply a table that correlates Jungle sentences on the left-hand side with English sentences on the right. Rather it is a function that generates a complete mapping of Jungle sentences on to English sentences. It does so recursively, by specifying, first, English translations of simple Jungle sentences and, second, how complex Jungle sentences are constructed out of these simple ones (WO 27–8, 68–72; PT 48).

Radical translation and radical interpretation

The radical translator describes language in the purely extensional terms of canonical notation, and on the basis of an austere behaviourist methodology. He treats human beings as black boxes whose ‘dispositions to speech behaviour’ are triggered by external stimuli (WO 207). Verbal behaviour is described not in terms of meanings or intentions, but only in terms of statistical regularities obtaining between movements, sounds and the environment. All the radical translator has to go by are the causal connections between ‘the forces that he sees impinging on the native’s surfaces and the observable behaviour, vocal and otherwise, of the native’ (WO 28). At the same time, the radical translator does not simply note what circumstances cause or inhibit the utterance of a sentence. For all manner of reasons, speakers of Jungle may refrain from uttering a sentence s even in a situation in which s would be true, for example because s is too obvious, because uttering s would be impolite, or simply because they are not in the mood. Pace Chomsky (1969: 57–8), Quine escapes this difficulty by focusing not on the natives’ disposition to utter sentences, but on their disposition to assent to sentences. The radical translator arranges the circumstances and ‘volunteers the sentence himself’, ‘asking only for a verdict of true or false’ (PT 39; TT 48). He first applies this procedure to ‘observation sentences’, that is, ‘occasion sentences’ of a particular kind. Occasion sentences are sentences which are true on some occasions, false on others – and which hence command assent only if queried in an appropriate situation. Thus ‘The Times has come’ is an occasion sentence, while ‘Water boils at 100 o C’ is a ‘standing sentence’, indeed, one of an eternal kind. In addition to being occasional, observation sentences are such that natives assent to them independently of collateral information; hence they are assented to by most competent speakers outright, and without further reflection (WO 42–4; OR 86–7; TI 2; PT §§2–4). Observation sentences report features of the environment that are directly observable on a particular occasion, as with ‘It rains’ or ‘This is red’. According to Quine, they provide the ‘entering wedge’ for both radical translation and language acquisition. The reason is two-fold. First, they are keyed to present events conspicuous to both linguist and native. Secondly, these sentences can be treated ‘holophrastically’, that is, as unstructured one-word sentences. A rabbit scurries by: the native seems to notice it and says ‘Gavagai’; the translator notes ‘Rabbit’ or ‘Lo, a rabbit’ as a tentative translation. To check that translation, he later repeats ‘Gavagai’

171

172

Quine and Davidson on Language, Thought and Reality

to his native informer in the conspicuous presence of a rabbit, and notes his assent or dissent (WO 29–30; TI 2; PT 39). The evidence for radical translation is exhausted by the evidence for the stimulus-meaning of native sentences. It concerns the conditions which prompt the native to assent to or dissent from a queried sentence. Quine further reasons that it is ‘stimulations’ or ‘stimulatory circumstances’ rather than macroscopic objects that prompt assent. Consequently, the conditions of assent/dissent, the external input into the black box, are patterns of neurophysiological events, ‘physical irritations of the subject’s surfaces’ such as the pattern of radiation on the retina (WO 31, 235; TT 48). Quine’s discussion of radical translation seeks to explain how a ‘meagre input’ of sensory stimulation gives rise to a ‘torrential output’ of verbal theorizing (OR 83). Its master-problem is: ‘how much of language can be made sense of in terms of stimulus conditions?’. The answer is: very little. Beyond certain limits, radical translation involves ‘a certain systematic indeterminacy’ (WO 23–6). What we can establish, according to Quine, is (WO 68): 1. the translation of native observation sentences, because in their case stimulus meaning approximates at least to meaning ‘properly so-called’ (WO 38); 2. the truth-functional connectives of Jungle (WO 57–8); 3. whether a native sentence is ‘stimulus-analytical’, that is, accepted under any circumstances, come what stimulation may (WO 55, 66); 4. whether two native sentences are ‘stimulus-synonymous’, that is, assented to under the same circumstances by all speakers (WO 46–7).

In order to get even this far, we need more than a description of the native language in terms of stimulus and response. We also require the famous ‘principle of charity’. According to this principle, our translationmanuals should minimize the ascription of false beliefs, especially as regards observation sentences and truth-functional connectives. For, Quine argues, it is ‘less likely’ that the interpretees hold obviously silly beliefs, such as contradictions, than that our translation is wrong (WO 59; see OR 46). Accordingly, we can assume that they are capable of noticing salient features of the environment and that they do not suffer from the ‘prelogical mentality’ occasionally diagnosed by anthropologists. Even with the assistance of the principle of charity, when she goes beyond (1)–(4) the translator is confined to ‘analytical hypotheses’, equations of Jungle words with English words that must not contravene the

Radical translation and radical interpretation

behavioural data, but cannot be verified by them either. Radical translation is indeterminate in ways that far exceed the familiar vagaries of translation (see Kirk 1986: ch. 1.1): manuals for translating one language into another can be set up in divergent ways, all compatible with the totality of speech dispositions, yet incompatible with each other. In countless places they will diverge in giving, as their respective translations of a sentence of the one language, sentences of the other language which stand to each other in no plausible sort of equivalence however loose. (WO 27)

For any language L, there are at least two translation manuals M1 and M2 which satisfy two conditions (WO 68–72; PT 47–8): i. they are behaviourally equivalent in that they fit the facts about the linguistic behaviour of the speakers of L equally, and unimprovably, well. ii. they are semantically incompatible in that the English sentences which they assign to a given sentence of L do not intuitively speaking mean the same, and hence are not interchangeable in English contexts.

This indeterminacy has two aspects or roots, which Quine distinguishes as the arguments ‘from above’ and ‘from below’ (RIT 183; see OR 67). The argument from above is a corollary of the thesis that scientific theory is underdetermined by evidence. In translating Jungle sentences we have a choice which mirrors the choice we have in adjusting our own web of beliefs to the empirical evidence. We may translate a given utterance differently by making compensating adjustments in the translation of other alien utterances. For example, if the native assents to the sentence ‘gavagai’ whenever I assent to ‘there is a rabbit’, it does not follow that both have the same stimulus-meaning. The native may report the presence not of rabbits but of a fly of the kind that always hovers over rabbits. Conversely, if she assents to ‘gavagai’ while I dissent from ‘there is a rabbit’, this may be due not to a difference in meaning, but to different collateral information. It might be that, in contrast to me, she has detected a rabbitfly or knows that these flies accompany rabbits. In the case of observation-sentences like ‘Gavagai’ this problem can be overcome in principle (ITA 8–9; PT 51; cf. WO 37–9). We can exclude collateral information, for example by disassociating rabbits and rabbitflies. But this procedure cannot be extended to theoretical sentences. Since these sentences do not report a feature of the environment observable on a particular occasion, we have no way of singling out the relevant features.

173

174

Quine and Davidson on Language, Thought and Reality

Any guess would depend on a variety of optional conjectures concerning the meaning of other sentences, which means that different overall translations can account for the same behavioural data (WO 76–7; OR 89; RIT; EEW 314). The argument from below aims to establish a second dimension of indeterminacy which is supposed to afflict even observation sentences, namely the ‘inscrutability of reference’ (WO §12). Even in the case of native sentences to which we can assign an objective (stimulus) meaning, we cannot establish what the terms occurring in these sentences refer to, since that depends on how we translate certain other native expressions. The basic idea is simple: ‘divergent interpretations of the words in a sentence can so offset one another as to sustain an identical translation of the sentence as a whole’ (PT 50). We can assign the same stimulus meaning to a sentence s by interpreting its components in different ways. Nimtz has shown that Quine takes two different routes to this conclusion (2002: 88–114). The first purports to show that in radical translation specific terms can be interpreted in incompatible ways that nevertheless preserve stimulus meaning. Assume that we have established that the stimulus meaning of the native sentence ‘gavagai’ is identical with that of our ‘There’s a rabbit’ or ‘Lo, a rabbit’. It still remains impossible to tell what the extension of ‘gavagai’ is, whether it refers to a rabbit or to an undetached rabbit part, the temporal stage of a rabbit, rabbit fusion (the scattered physical aggregate of all rabbits), or what not. We cannot even tell whether it is a concrete general or an abstract singular term which refers, for example, to a recurring universal, namely rabbithood. These various predicates are not co-extensional (‘rabbit’ is not instantiated by an undetached rabbit part); but if any one of them is instantiated, so are all the others, and by something which is in the same spatiotemporal region. The only way of removing these uncertainties is to ask in Jungle questions like ‘Is this the same gavagai as that?’ But that presupposes a prior translation of what Quine calls ‘the apparatus of individuation’, or of ‘objective reference’: expressions like ‘the same’, articles, pronouns, numerals, singular and plural endings, quantifiers, etc. (WO 53, 70–2). Yet we cannot establish the translation of these expressions independently of establishing the translation of expressions like ‘gavagai’. Once more there are different ways of construing the overall behavioural data. Suppose that we have established that (1) Dohab gavagai roles dahab gavagai

Radical translation and radical interpretation

is assented to only when a rabbit that has formerly been seen reappears. It would seem natural to translate (1) as (1
) This rabbit is the same as that rabbit.

But other alternatives are equally compatible with the behavioural evidence. For example, if we translate ‘roles’ not as ‘is the same as’ but as ‘is a part of the same rabbit as’ we can translate ‘gavagai’ as ‘undetached rabbit part’ and render (1) as (1∗ ) This undetached rabbit part is a part of the same rabbit as that undetached rabbit part.

Quine’s second, so-called ‘proxy function argument’ is more abstract (FM 165; TT 18–20; PT 31–2). A proxy function is a one-to-one mapping f which assigns to each object x in a domain a different one, its correlate or proxy fx. We can then reinterpret every sentence as being a sentence about the proxies fx of the objects x, namely by also reinterpreting each predicate as being true not of objects, but of their proxies. On our standard interpretation, a predicate ‘x’ means that x is a , on the proxy function interpretation, it means that fx is an f of a . For example, we can map every physical object on to its ‘cosmic complement’, that is, the universe minus the object (SS 71–2). By this token, the word ‘rabbit’ denotes not each rabbit but the cosmic complement of each, and the predicate ‘furry’ denotes not each furry thing but the cosmic complement of each. We can then interpret ‘Rabbits are furry’ as saying not that rabbits are furry, but that the complements of rabbits are complements of furry things. A systematic permutation over the reference of sentence components will not alter the stimulus meaning of sentences (WO 27).

2

Quine’ s closet hermeneutics

One way of attacking radical translation and the indeterminacy thesis is to argue that Quine’s method of translation cannot yield even the meagre results it is supposed to, without tacitly smuggling in either a prior understanding of the natives, or hermeneutical methods and intensional notions which he disowns. Such a strategy must avoid armchairanthropology by keeping apart genetic and conceptual issues. The Quinean translator might return from his safari with a perfect grasp of the native tongue, however austere his procedures. The question is whether

175

176

Quine and Davidson on Language, Thought and Reality

radical translation can make such success intelligible, and whether it provides an adequate account of just what has been achieved when a completely alien language has been learned. This anti-genetic lesson applies to a prima facie plausible misgiving about Quine’s method. Even a sympathetic commentator like Hookway has suggested that Quine and Davidson tend to ignore the interaction between translator and native (1988: 172–3). As it stands, this complaint is unjustified. The radical translator is not confined to observation. As we have seen, he ‘takes the initiative’ (WO 29) by trying to elicit assent or dissent to native sentences he himself volunteers. Furthermore, contrary appearances notwithstanding, interaction is not an essential precondition for successful translation. Imagine a translator along the lines of H. G. Wells’ Invisible Man, who can move freely among the natives, without interacting with them or being noticed. There is no a priori reason why such an invisible translator should not be able to pick up Jungle, for example by observing the linguistic instructions given to native children. If they can do it, why shouldn’t he? Of course, his learning would be facilitated if he could ask questions and have his mistakes corrected. In principle, however, the invisible translator could learn through mere observation, just as prodigies have learned chess in this way. Consequently, Quine cannot be accused of ignoring the need for interaction between translator and native. What he can be accused of is mischaracterizing this interaction. He is committed to describing it as an experiment rather than a conversation. That is why he contrasts radical translation proper with ‘going bilingual’ (WO 47, 71; cf. SS 78–82). The translator does not strictly speaking ask the native a question. He operates ‘in a more causal vein’, that is, he provides a certain acoustic input into a verbal machine and notes what neural ‘stimulations . . . will prompt the native to assent or dissent’ (WO 30). But actual translation from scratch is not based on theorizing, let alone theorizing governed by the strictures of Quine’s austere behaviourism; rather, it is based on learning, a particular form of human interaction. In translation from scratch we communicate and cooperate with the natives in a multitude of ways, before we have come up with a complete translation manual. This is not just a minor failing, since it is connected to a crucial assumption of the indeterminacy thesis. Both the argument from above and the argument from below presuppose that a translation is automatically indeterminate if it is based on assumptions to which we can conceive an alternative. This stands in stark contrast to what is known as the

Radical translation and radical interpretation

hermeneutical circle, the idea that in understanding a remote text or culture we have to start by making certain prima facie plausible assumptions about specific passages or actions, the validity of which is then checked against the plausibility of the overall interpretation to which it leads, which in turn is modified by reference to the rectified understanding of the specific passage, and so on. If, unlike the philologist or invisible interpreter, we can interact with the subjects, this procedure is even more potent. We can check our tentative translations through mutual querying, explanation and correction. Quine is committed to ruling out this procedure as inadequate. He insists on a presuppositionless method in his arguments for indeterminacy. But, as I hope to show, Quine himself is a closet hermeneutist in the translations he licenses. Hence, we can either translate much more than he allows, or nothing at all. There are three prominent points at which this objection might apply, namely the move beyond observation sentences, the translation of the truth-functional connectives, and the identification of assent/dissent. According to Quine, in order to get beyond the threshold of observation sentences like ‘it is raining’ to which all speakers assent in the same situations, independently of their background information, we do need to ‘go bilingual’, that is, to learn the native language as children do, and then to translate ‘by introspected stimulus-synonymy’ (WO 46–7). Unfortunately, it is unclear what this phrase means, or even whether it is meant seriously. The most plausible gloss on it is that the bilingual translator makes observations like ‘Whenever I would assent to “Ich bin ein Berliner” I would also assent to “I am a doughnut”’. But in that case the translator does not really introspect anything. Rather, he observes his own behaviour and makes inferences about his dispositions, just as he has previously done with the natives. Yet it is highly implausible to suggest that a bilingual needs to establish his behavioural dispositions in order to explain the words of one language in terms of the other. Quine himself acknowledges that ‘going bilingual’ transcends his method of translation. But even if Quine cannot account for the move beyond observation-sentences, this need not disturb him excessively, since that move is not listed among the results of his method. By contrast, univocal translation of the truth-functional connectives is.1 At first blush this 1. No longer in Roots of Reference (§20), though for reasons different from the ones canvassed here.

177

178

Quine and Davidson on Language, Thought and Reality

seems justified. We can translate, for example, the native ‘blip’ as ‘and’ if for any two Jungle sentences ‘p’ and ‘q’ the native assents to ‘p blip q’ if and only if he will assent to ‘p’ and to ‘q’. However, this account already relies on assumptions which seem no less problematic than those Quine deplores vis- `a-vis the inscrutability of reference. Just as distinguishing between rabbits and rabbit-stages presupposes prior translation of the apparatus of individuation, translating truth-functional connectives presupposes that i. all uses of ‘blip’ are truth-functional, which does not hold of our ‘and’ (see WO 58); ii. sentences can be distinguished from words or other components of sentences. This will create particularly severe problems if, for example, the conditional is expressed by a term like ‘if . . . then’, which is not only compound, but both parts of which play roles other than that of a conditional (‘She is clever, if superficial’, ‘Then she left’); iii. declarative sentences can be distinguished from questions, commands, subjunctives, etc. But the ability to make these distinctions goes hand-in-hand with the translation of interrogative pronouns and a rough grasp of speech act patterns.

The final breakdown of Quine’s method, concerning the identification of assent/dissent, is the most fatal, since it would prevent Quinean translation from even getting off the ground. For reasons rehearsed above, radical translation cannot proceed by simply observing what situations prompt what utterances. It is therefore essential to Quine’s approach that we can identify the dispositions to assent or dissent. He suggests that the radical translator can identify assent/dissent as follows: in asking ‘Gavagai’ . . . in the conspicuous presence of rabbits . . . he has elicited the responses ‘Evet’ and ‘Yok’ often enough to surmise that they may correspond to ‘Yes’ and ‘No’, but has no notion which is which. Then he tries echoing the native’s own volunteered pronouncements. If thereby he pretty regularly elicits ‘Evet’ rather than ‘Yok’ he is encouraged to take ‘Evet’ as ‘Yes’ . . . However inconclusive these methods, they generate a working hypothesis. (WO 29–30; see PT 39)

Note in passing that it is unclear what sort of evidence could lead one to surmise that ‘Evet’ and ‘Yok’ are expressions of assent or dissent without indicating which one expresses assent and which one dissent. Assent is linked to behaviour in strikingly different ways from dissent, namely to acceptance or agreement rather than to rejection or disagreement.

Radical translation and radical interpretation

The more immediate trouble is that Quine’s procedure is not presuppositionless. For a start, it presupposes that the translator has correctly translated the observation sentence – in our case ‘gavagai’ – which he uses to elicit assent or dissent from the native. If he has not, he will be led to confuse assent and dissent. But that translation is subject to the aforementioned risk that the natives utter or fail to utter ‘gavagai’ for reasons other than the presence or absence of rabbits. Moreover, Quine’s procedure presupposes a mutual understanding between native and translator to which he is not entitled. It is assumed that the native understands that the translator’s ‘Gavagai’ is meant as a question concerning the meaning of that expression, and not as a religious ritual involving rabbits, a challenge to his hunting-rights, or simply a dumb repetition. In that last case, an ‘Evet’ in response to the translator’s echoing the native’s utterances might not be a sign of assent, but a rebuke for parroting. These possibilities of misunderstanding are illustrated by the apocryphal etymology of the English term ‘kangaroo’. Allegedly, one of the first Europeans pointed at a kangaroo and asked ‘What is this?’, to which the aborigines replied ‘kangaroo’, which in their language means ‘I haven’t a clue what you are talking about!’ (RR 44; Hacking 1975: 150). A more reliable tale involves the first French translation of Tongan. That language does not contain numerals above twenty. But when the French translator Labillardi `ere persisted in asking for such numerals, he received expletives in reply, which he solemnly noted as Tongan numerals (Wood 1938: 24). Such possible and actual misunderstandings demonstrate that Quine’s procedure presupposes that interpreter and native engage in a specific kind of dialogue, that is, perform certain types of speech acts. Quine takes for granted that the native tries to teach his language to the translator, which (among other things) means that he will apply words in paradigmatic situations, and correct the translator’s fledgling attempts to speak Jungle. Quine’s approach cannot account for this mutual understanding, because that understanding can only be described in intensional and intentional terms, for example by saying that the natives take the translator’s ‘Gavagai’ to be a yes/no question. But the only alternative to taking this kind of understanding for granted is to assume that the native knows that the radical translator is trying to establish the stimulus meaning of her words. For Quinean translation to work, the natives had better read a translation of Word and Object!

179

180

Quine and Davidson on Language, Thought and Reality

One might object that Quine acknowledges that his method is ‘inconclusive’ and merely generates a ‘working hypothesis’ (WO 30; see RIT n2). However, without assuming a framework of interaction, identifying assent/dissent would not just be hypothetical, a reasonable if inconclusive guess, it would be completely arbitrary. There would be no reason to suppose that the informant’s reaction is at all relevant to assent or dissent. Moreover, any concession that the identification of assent/dissent is not presuppositionless means that Quine is here applying different standards from those at work when he propounds the inscrutability of reference. If there is no fact of the matter as to whether ‘gavagai’ refers to rabbits, then, by parity of reason, there is no fact of the matter as to whether the native assents to or dissents from the translator’s ‘gavagai’. Within Quine’s framework, that would remove the possibility of translating anything. There is yet another and equally severe problem with the method of radical translation. In characterizing the native’s reaction as assent and dissent, Quine tacitly describes the output of the behaviourist experiment in richer terms than the input. The former is held to consist of surfaceirritations, more specifically, of patterns of stimulation at the surface of the perceptual organs. Assent and dissent, by contrast, are not mechanical reactions or mere bodily movements, but forms of intentional (linguistic) behaviour. If the native screams ‘Yok’ because he was stung by a hornet, he has not dissented from the anthropologist’s ‘gavagai’. They are also intensional. One assents to or dissents from what is said, namely that things are thus-and-so. Quine insists that he is talking not of assent in the ordinary sense, but only of ‘surface assent’, a mere noise or bodily movement (MVD 91). But this manoeuvre is futile. He himself states that to assent to a sentence is to pass a verdict on its truth which may be mistaken, and that the subject believes what is uttered (TT 48). This in turn implies that assent and dissent are not mechanical reactions, but responses to something the native has understood, namely the anthropologist’s utterance. Consequently, the concept of assent which Quine actually deploys is intimately interwoven with epistemic and intensional notions. Furthermore, he could not settle for a behaviourist ersatz like surface assent. Unless it expresses what the native believes to be true, assent (and the notion of stimulus meaning defined by reference to it) becomes irrelevant not just to questions of meaning, which Quine might happily accept, but also to epistemology. There would be no point in trying to avoid ascription of patently false beliefs, as the principle of charity bids us to do. More generally, Quine’s discussion of radical

Radical translation and radical interpretation

translation would lose its point, which was to explain the link between our utterances and the evidence on which they rest. Notions like surface assent, stimulus meaning and stimulussynonymy are not supposed to be more than a behaviourist ersatz of the discarded intensional notions. At the same time, they are not supposed to be less than Carnapian ‘explications’ of these notions (see WO 66; RHS 427–8). That is to say, they are alternatives which avoid the drawbacks of the originals while still serving their cognitive purposes. In our case this means that the notions of stimulus-meaning and stimulus-synonymy should capture the ideas of cognitive significance and of cognitive equivalence, respectively (TT 47–51). This in turn means that they must have a cognitive dimension: they must concern evidence that the speaker understands, and which is therefore relevant to his beliefs. Unlike our intensional notions, a consistent behaviourist ersatz would lack the conceptual connections with epistemic concepts like belief, knowledge, etc. Quine repudiates our intensional concepts in the name of an austere behaviourism. But he must tacitly rely on these concepts, if his discussion is to have the implications he assigns to it. Equally, his procedure of identifying assent and dissent must rely on assumptions which he condemns in his arguments for indeterminacy. For both reasons, the answer to the question ‘How much of language can be made sense of by Quinean translation?’ is not ‘very little’ but ‘none whatsoever’. It is tempting to regard this result as a reductio ad absurdum of Quine’s position. He aimed to provide a behaviourist theory of language; but it transpires that such a theory cannot make sense of meaning and understanding to even a modest degree. He himself has been driven to the disarming admission that ‘radical translation is a near miracle’ (ITA 9). Quine has nonetheless rejected the charge that the indeterminacy of translation is a reductio of his behaviourism. His equanimity rests on two convictions. First, he finds it ‘reassuring to reflect that any discrepancies of translation within the limits of indeterminacy will fall innocuously between objective checkpoints’ (RHS 75), in that the behavioural evidence does preclude certain translation manuals. Secondly, he holds that ‘the behaviorist approach is mandatory’, because in learning a language ‘we depend strictly on overt behavior in observable situations’ (PT 37–8). However, if my arguments are sound, the determinacy Quine claims was purchased by surreptitious violations of his behaviourist strictures. Had he stuck to his avowed method, no checkpoints would have remained and even stimulus meaning would have turned out to be completely

181

182

Quine and Davidson on Language, Thought and Reality

indeterminate. Furthermore, although there is no alternative to learning the natives’ language on the basis of what they say and do, there is an alternative to describing what they say and do in Quine’s behaviourist idiom of neural stimulus and verbal response.

3

On safari with Davidson

When Davidson coined the term ‘radical interpretation’ in 1973, he paid tribute to Quine. But he also declared that the two thought experiments serve different purposes (ITI 126n, 129n; see SS 80–1). Quine’s radical translation is part of his flight from intensions. Its purpose is to establish the extent to which behavioural data determine the semantic features of a language. Davidson’s radical interpretation is a continuation of truthconditional semantics by anthropological means. It is meant to elucidate what kind of knowledge would enable a hearer to understand the utterances of a completely unknown language. Radical interpretation and radical translation are different but compatible – that is the official line. In fact, however, they are both closer in their ambitions and wider apart in their edifices than this suggests. Although the projects are distinct, they are in turn fuelled by a common aim. Both Quine and Davidson want to show how semantic and psychological notions can be accommodated within a naturalistic framework. This programme of philosophical austerity is more strident in Quine, but it is equally important to Davidson. While Quine seeks to eliminate all intensional notions from science, Davidson seeks to extract rich semantic concepts and statements from evidence which he considers to be more basic, because it can be described in non-semantic terms. But this process is not meant to be one of straightforward definition, analysis or reduction. I am concerned with what I take to be, historically at least, the central problem of philosophy of language, which is how to explain specifically linguistic concepts like truth (of sentences or utterances), meaning (linguistic), linguistic rule or convention, naming, referring, asserting, and so on – how to analyse some or all of these concepts in terms of concepts of another order. Everything about language can come to seem puzzling, and we would understand it better if we could reduce semantic concepts to others. Or if ‘reduce’ and ‘analyse’ are too strong (and I think they are), then let us say, as vaguely as possible, understand semantic concepts in the light of others. (ITI 219, see also 137)

Radical translation and radical interpretation

Less vaguely, the idea is to elucidate and tame the concept of meaning by showing how a ‘specifically semantical’ theory can be derived from ‘non-semantical evidence’ about linguistic behaviour (ITI 142). Less vaguely still, a Davidsonian theory for a language L must satisfy two conditions: First, it must be interpretative, in that its theorems give the meaning of the sentences of L. Secondly, it must be radical, in that the evidence on which it is based must be available to someone who as yet knows neither what the speakers of L mean by their utterances, nor what they think (believe, desire, etc.) or do.

Because of this last condition, radical interpretation shares with radical translation the anthropological scenario of understanding an alien community entirely from scratch. At the same time, there are important differences between radical interpretation and radical translation (see Hookway 1988: 167–74; Evnine 1991: ch. 6.2). One of them simply reflects the different goals. The radical interpreter seeks to provide not just a translation manual that correlates Jungle and English sentences, but a theory of meaning that states what the former mean by specifying under what conditions they are true. Other differences between radical interpretation and radical translation, however, arise from incompatible views about mind and language. Although Davidson follows Quine in adopting a thirdperson perspective on thought and meaning, he disowns Quine’s austere behaviourism. This gives rise to several notable contrasts. The first is that Davidson acknowledges the ‘frankly intensional’ and intentional nature of assent (ITI 231, see 167). Like Quine, he insists that the only empirical evidence available to radical interpretation concerns what sentences the natives assent to under what circumstances. Instead of talking of assent, Davidson talks of ‘holding a sentence true, accepting it as true’. Prima facie, this marks a difference, since assent is a form of behaviour, whereas holding a sentence true is a type of belief (ITI 135). But Davidson no less than Quine insists that belief or holding a sentence true is manifest in a form of behaviour, namely ‘prompted assent’, which is essentially Quine’s suggestion (SCT 314–15; SIO 147). The real difference is that Davidson does not subscribe to Quine’s behaviourist account of assent. Assent is intentional, since we do it for a reason. And it creates an intensional context. A speaker may assent to ‘Cicero was a Roman orator’ without assenting to ‘Tully was a Roman public speaker’. Consequently, there

183

184

Quine and Davidson on Language, Thought and Reality

is an ‘irreducibly intensional element’ to radical interpretation. We can neither avoid intensional notions nor reduce them to something more ‘scientific’, ‘behaviouristic, neurological, or physiological’ (UT 5; ITI 175–6; STL 25).2 What Davidson seeks to avoid are ‘linguistic concepts’ or ‘unwanted semantic terms’. Not all semantic terms are unwanted, moreover. As we have seen, he is happy to rely on an irreducible concept of truth. What he eschews, it seems, are notions that are both intensional and semantic or linguistic (see ITI 128, 205–6, 219; SCT 296–7+n). This applies to the notion of meaning, which the theory seeks to elucidate, as well as its conceptual relatives (interpretation and synonymy), friends (rule, convention) and servants (reference and satisfaction). Davidson regards these concepts as problematic. They are conspicuous by their absence from his recent list of concepts that cannot be reduced to something more fundamental (FDT 263–5, 275–6). A related difference is that unlike Quine, Davidson does not reject our ordinary psychological concepts. Consequently he describes the task of interpretation as one of assigning meaning to the aliens’ utterances, attributing mental states to them, and understanding their actions. As a result, radical interpretation concerns not just meaning and the content of utterances, but also the content of thought, what we believe, desire, etc., especially since for Davidson thought and language are intimately connected. The tasks of assigning meaning and attributing thoughts – beliefs and desires – hang together holistically (ITI 27, 127, 240; SIO 146–7; SCT 316–18). We could translate Jungle sentences if we knew what thoughts they express, and we could ascribe thoughts to the natives if we could interpret their sincere utterances. Alas, at the start of radical interpretation we know neither the meaning of their words nor their thoughts. This complication leads to further differences between Quine and Davidson, concerning the principle of charity. Both entreat the anthropologist to avoid the ascription of false or inconsistent beliefs to the natives. But Davidson treats charity not just as a heuristic maxim that enhances the prospects of interpretation, but as a principle which is essential to the correctness of an interpretation. An interpretation which fails to make the natives’ beliefs come out as largely true, and their desires as largely intelligible, is not just ‘less likely’ to be adequate, it must be inadequate. 2. Alas, he is not consistent on this score. On occasion, he suggests both of holding true and of preferring true – the notion that underlies his unified theory of thought and action – that they are extensional (SIO 211; SCT 323).

Radical translation and radical interpretation

Davidson’s rationale for taking this line is that our only way of breaking into the holism of meanings, beliefs and desires is to maximize agreement with the interpretees. If interpretation is impossible because we cannot construe the aliens’ behaviour as largely rational, we shall have to conclude that they do not speak a language to begin with. This introduces a normative element into linguistic understanding. We can make sense of others only in so far as we can treat them as rational agents who abide by certain norms of rationality. Furthermore, unlike Quine, Davidson applies charity ‘across the board’, that is, indiscriminately to all types of belief. This more generous dispensation of charity in turn leads him to maintain that the indeterminacy of translation/interpretation is less extensive than Quine assumes (see section 4 and chapter 7, section 4). The most important difference, however, concerns the conditions that prompt the natives to assent to a sentence. For both Quine and Davidson, these conditions determine sentence meaning, and hence constitute the evidence for radical translation/interpretation. But their accounts of these conditions differ sharply. Quine condones a ‘proximal’ theory, that is to say, he describes them in terms of neural stimulations on the subject’s surface. By contrast, Davidson describes the conditions of assent in terms of macroscopic objects and events. The ‘locus of shared meaning’ are ‘distal’ phenomena rather than ‘proximal’ neural firings (IA 41). For Quine, the conditions that prompt assent to an observation sentence determine its meaning, because they constitute the sensory evidence on which the assent rests. By contrast to previous empiricists, he does not equate this evidence with ideas or sense-data. His reason is that the empirical foundations of knowledge and meaning must be intersubjectively accessible and hence physical rather than mental (WO 234–5; OR 82, 87; TT 39–40). Quine grants that observation sentences are, by and large, about macroscopic objects (RR 40; TT 40; TI 2; WWD 74). At the same time, he insists that the evidence on which speakers assent to observation sentences consists of neural stimulations or triggerings. The native might continue to assent to ‘Gavagai’ even if the animal were replaced by a counterfeit; conversely, she might dissent even in the presence of an animal, pending her sensory stimulations. In spite of later protestations, Quine originally treated translation as a matter of ‘neurologizing’ (RI 502; PTF 159): ‘In experimentally equating the uses of “Gavagai” and “Rabbit” it is stimulations that must be made to match, not animals’ (WO 31). This is why he defined both stimulus meaning and observation sentences in neural terms.

185

186

Quine and Davidson on Language, Thought and Reality

For Davidson, this theory of meaning and evidence is based on a ‘third dogma of empiricism’, the dualism of conceptual scheme and empirical content, which implies that ‘epistemic intermediaries’ intervene between the world and our beliefs and utterances (ITI ch. 13; SIO ch. 10; MTE). He in effect accuses Quine of succumbing to the empiricist ‘myth of the given’, the idea that conceptually unmediated sensory stimulations provide the foundations of knowledge and of meaning (similarly Rorty 1979: 171–2; Dilman 1984: ch. 6). Quine for his part has accused Davidson of confusing truth and belief. While intermediaries between world and language are superfluous in accounting for the former, they are essential to specifying the evidence on which our beliefs rest (TT 39; TI 3; Lauener 1990: 223–4). In response, Davidson denies that neural stimulations can play either the epistemic role of providing the evidence for our beliefs or the semantic role of determining the meaning of our sentences. He concedes that neural firings feature in the causal chain between objects and events on the one hand, beliefs and assent on the other. But to think of them as evidence is to confuse the causes of our beliefs with the ‘reasons’ or ‘justification’ on which they rest. Unlike beliefs, neural events cannot stand in relations of either logical or probabilistic support (SIO 141–6, 169–70). Another disadvantage of Quine’s switch from sense-data to triggerings of nerve endings is that the subjects are no longer aware of the alleged evidence for their beliefs. But whereas people are by-and-large ignorant of the neural causes of their beliefs, they cannot be entirely ignorant of the evidence on which they rest. Finally, Davidson has accused the proximal theory of being unduly ‘Cartesian in spirit and consequence’ (MTE 76), because it makes conditions of assent and hence meaning unacceptably private. Quine admitted to problems on this score, and he has shifted his position over the years (see PT 40–4; TI 2–3; PTF). In Word and Object he maintained that the correct translation of the native ‘Gavagai’ is an English sentence with the same stimulus meaning. He also defined observation sentences intersubjectively, namely as those occasion sentences which have the same stimulus-meaning for all speakers of a linguistic community (WO 42–3; OR 86–7; RR 23–4). Since stimulus meaning is determined by the neural triggerings that would prompt assent, both claims assume at least an approximate neural isomorphism from one individual to another. As Quine later recognized, however, such an isomorphism is not to be expected. Neural firings, though intersubjectively accessible, are not shared between different speakers.

Radical translation and radical interpretation

As a result, Quine privatized his definition of observation sentences. A sentence is observational for an individual speaker if on all occasions it elicits assent or dissent ‘when the same total set of receptors is triggered’ (TT 25); a sentence is observational for a group if it is observational for each of its members. Alas, this manoeuvre allows for the possibility that a sentence is observational in a linguistic community, although each of its members associates it with a different neural stimulation. Every individual might assent to ‘I just felt a nibble’ given the same stimuli, but these could differ from individual to individual (PT 41). Furthermore, Davidson mentions the case of a native and a linguist confronted with a warthog. The situation prompts the native to assent to ‘Gavagai’. But because of suffering from astigmatism, he has the same pattern of neural triggerings which the linguist has when confronted with a rabbit. Going by stimulus meaning, the linguist would have to translate ‘Lo, a rabbit’, even if the native knew from past experience that no rabbit is in sight (MTE 74). In response, Quine continued to locate stimulation and stimulus meaning at the neural input rather than the external objects of reference, while abandoning intersubjective comparisons of stimulation (CBG 80; see PT 41–4; Gibson 1994: 485). Their place is taken by ‘empathy’. ‘Gavagai’ has its stimulus meaning for the native, ‘Lo, a rabbit’ has its stimulus meaning for the translator. The latter translates the former not because the two are linked to isomorphic triggerings, but because the translator conjectures that she would have uttered it had she been in the native’s position. Quine retains his definition of observation sentences for individuals and defines a sentence as observational for a group if it is individual for each member and if they all assent to it on ‘witnessing the pertinent occasion. What counts as witnessing the occasion will be judged, as in the case of translation, by projecting oneself into the witness’s position’ (TI 3–4; see PT 43). This appeal to empathy does not defuse the threat of neural solipsism. According to Quine, the linguist has the ability of ‘projecting himself into the native’s sandals’, that is, to imagine himself in the native’s position and to consider ‘what English sentences it would prompt of him’. Let us grant that we possess an ‘uncanny knack for empathizing another’s perceptual situation’ (PT 42; IT 4; CBG 292). It would be downright spooky, however, if we had the ability to put ourselves in someone else’s neural sandals, given that we do not even know what our own neural stimulations are. Empathy can only mean to imagine what an objective situation looks like for others, not to imagine what it triggers like.

187

188

Quine and Davidson on Language, Thought and Reality

As the parable of the trimmed bushes (WO 8) already suggested, neural homogeneity or heterogeneity does not matter to meaning and translation, as long as speakers are disposed to assent to observation sentences in the same objective situations. In Davidson’s words, what matters to meaning does not ‘occur within the speaker’s skin’ (MTE 77). In his last statements Quine came to accept that meaning and translation are keyed to external objects of which both native and linguist are cognizant (WWD 74; see PT 44; PTF 160). This concession pays homage to his own insight into the intersubjective nature of meaning. But it also necessitates a retreat from a central claim of his earlier work, namely that stimulus meaning is a scientifically respectable explication of the notion of meaning, indeed one that matches the explanandum as far as observation sentences are concerned (WO 42; RHS 427–8). Davidson moves away from Quine’s austere behaviourism to a considerable extent. I shall argue, however, that this move is half-hearted and does not make interpretation from scratch intelligible. Davidson combines an ontological monism or physicalism with a conceptual dualism. He accepts that ‘linguistic phenomena are nothing but behavioural, biological or physical phenomena’, while insisting that the ‘exotic vocabulary of meaning, reference, truth, assertion, and so on’ in which we describe these phenomena cannot be reduced to the vocabulary of physics. The task, therefore, is ‘bridging the gap’ between the semantic and intentional on the one hand, and the physical on which it supervenes on the other (SCT 315; ITI 126–7). Radical interpretation is supposed to solve this task by showing how theories of meaning can be derived from ‘more primitive data’, purely physical descriptions of human behaviour. Accordingly, interpretation is a matter of ‘theory-construction’ (ITI xx). ‘Everyday linguistic and semantic concepts are part of an intuitive theory for organizing more primitive data, so only confusion can result from treating these concepts and their supposed objects as having a life of their own.’ The starting point of this theory is ‘non-semantical evidence’ ‘that can be stated without essential use of such linguistic concepts as meaning, interpretation, synonymy, and the like’ (ITI 142–3). The ‘interpreter has nothing to go on but the pattern of sounds the speaker exhibits’, his initial data are ‘non-linguistic goings on’ like movements of the lips and larynx (SIO 13–14; ITI 126). These passages strongly suggest that Davidson himself succumbs to the myth of the given. Interpretation is the derivation of statements about what a speaker says, thinks and does from descriptions of him ‘as a

Radical translation and radical interpretation

physical system’ (Lewis 1983: 108). This picture is awry. Even interpretation from scratch cannot be a matter of inferring meanings and beliefs from a physical description of mere sounds and bodily movements, no matter whether macroscopic or neural. While we find it easy to describe human behaviour, even that of a completely alien community, in ‘rich’ mental terms, we are ignorant of the austere physical descriptions that radical interpretation treats as evidence. We can describe a person’s features as ‘sad’, ‘radiant’ or ‘bored’, yet do not know how to describe his face in purely physical terms (Wittgenstein 1967b: §225). And even a complex philosophical lecture is easier to understand than to describe in terms of its physical or phonetic features. In other words, we are able to state the conclusions of the alleged inference, without even being able to understand the premises from which they are supposedly derived. One might reply that the inference is unconscious. However, while human speech involves complex causal processes of which we are unaware, the neurophysiological causes of our linguistic behaviour are not pieces of evidence from which we derive the meaning of what has been said. Since Davidson himself urges this point against Quine, this line of defence is foreclosed to him. Perhaps we could provide physical descriptions of the natives’ surroundings, phonetic descriptions of their utterances, and biomechanical descriptions of their facial expression and bodily movements. Yet this is not the kind of evidence Davidson requires. For he rightly insists that the behavioural evidence that matters to meaning must be available not just to scientists with fancy equipment, but to ordinary communicators (see chapter 7, section 2). Furthermore, we literally have no clue about how to derive semantic and mental characterizations from this kind of evidence. In human interaction we do not encounter mere sounds or movements. We no more infer what people say from mere sound-patterns or their physical descriptions than we infer what is before our eyes from mere shapes and patches of colour. The capacities to perceive objects, to discriminate human speech, and to acquire linguistic competence given suitable training are simply part of our innate cognitive capacities. They are not mysterious, since social and linguistic primates like ourselves could not survive without them. On occasion, Davidson comes close to granting these points (e.g. RH 330). But Wittgensteinians have long insisted that a more thoroughgoing rejection of empiricism is inimical to the project of radical interpretation: semantic properties and facts are not theoretical or inferred, but observational, and the idea of a non-semantic starting-point for understanding

189

190

Quine and Davidson on Language, Thought and Reality

is spurious (Heal 1989; Hacker 1997: 302–4; McDowell 1998: chs. 4–5, 15, 17). The trajectory of this reasoning is correct: the derivation of the semantic and intentional from the physical is futile. Meaningful speech is not merely a physical phenomenon, in that it is embedded in distinctively human capacities and forms of interaction. Furthermore, within this context, it does have a life of its own, and our understanding of it is not derived from physical facts. The bottom line is not physics but praxis. But this insight does not by itself solve the problem of interpretation from scratch. In normal communication our knowledge of meaning may be observational: what other members of our linguistic community say is, by and large, evident, because we have all been socialized into the same linguistic practice. Even to utterances in a foreign tongue we do not, under normal circumstances, react as mere noises. A crucial difference remains, however. The radical interpreter ‘does not already know how to interpret utterances the theory is designed to cover’ (ITI 128). Although he interacts with the natives, that interaction is often guided by what Quine calls ‘analytic hypotheses’. Davidson could further insist that the ‘non-semantic’ or ‘nonlinguistic’ evidence on which he bases such hypotheses consists not of arcane descriptions in the vocabulary of physics or biomechanics, but of common-or-garden descriptions of macroscopic events and of human action. The situation is described, for example, as a rabbit scurrying by or a ship sailing down the stream. And the evidence concerning native utterances involves not a phonetic description but simply a reproduction of those utterances (EAE 50–2). The radical interpreter is capable of providing such a reproduction, which he can put to the native for assent or dissent. The aforementioned passages notwithstanding, Davidson’s considered view should be that interpretation from scratch starts not from physical descriptions but from observations about what reproduced sentences native speakers accept in what circumstances. This last point reminds us that Davidson’s radical interpreter, unlike Quine’s translator, avails himself of evidence concerning an intentional and intensional form of behaviour – assent. At the same time, Davidson follows Quine in setting aside any other anthropological assumptions or behavioural evidence: all the evidence for or against a theory of truth (interpretation, translation) comes in the form of facts about what events or situations in the world cause, or would cause, speakers to assent to, or dissent from, each sentence in the speaker’s repertoire. (ITI 230; my emphasis)

Radical translation and radical interpretation

This restriction is implausible. There is no reason to assume that assent or dissent must be easier to identify than other speech acts such as questions or demands for explanations like ‘What does this word mean?’ (Putnam 1975b: 257–8). Indeed, pace Davidson and Quine (PT 39), there is some reason to assume that ‘the entering wedge’ for the anthropologist will not be declarative sentences to be tested for assent/dissent, but greetings, orders, exclamations and requests. In other passages, Davidson advisedly makes a weaker claim, namely that in radical translation holding a sentence true is ‘a good place to begin’, because it is an attitude an interpreter may plausibly be taken to be able to identify before he can interpret, since he may know that a person intends to express a truth in uttering a sentence without having any idea what truth. (ITI 135, see 144–5, 152; SIO 210–11)

The trouble is that prompted assent does not provide the unassuming starting point Davidson requires. For one thing, he presupposes that the radical interpreter is capable of telling whether the natives’ assent is sincere. I agree that ‘even the compulsive liar and the perennial kidder may be found out’ (ITI 144–5, see 135, 161–2; UT 5). But I am less sanguine that this can be done in advance of understanding what they say. This is not a major problem, however. At the beginning of radical interpretation we cannot assume that we can detect insincerity; but I see no alternative to assuming that the natives are sincere and willing to cooperate. We may be unable to interpret a community that consistently deceives us, but this is only a useful reminder of the fact that interpretation from scratch involves human cooperation. Two more serious difficulties emerged in our discussion of Quine. First, to test for assent, the interpreter not only needs to pick out declarative sentences, she must also rely on a hypothesis as to what beliefs they express. Secondly, even if it is sometimes possible to know that a person assents to a sentence without knowing what claim she thereby assents to, this does not show that it is possible to identify assent without prior assumptions. Identifying assent presupposes that anthropologist and native engage in a certain type of communication, and hence a certain mutual understanding. It does not, therefore, provide interpretation with a truly radical starting point. Perhaps such qualms have become obsolete. In his recent attempts to integrate the theory of meaning into a unified theory of mind and action,

191

192

Quine and Davidson on Language, Thought and Reality

Davidson has altered the evidential basis of radical interpretation from holding a sentence true to preferring the truth of one sentence over the truth of another sentence (UT; ITI 145–9; SCT 309–28). A unified theory ascribes meaning, belief and desire in one fell swoop. It combines a theory of interpretation influenced by Quine with Bayesian decision theory in the form bequeathed by Ramsey (1990: ch. 4) and Jeffrey (1983). Both projects start out with readily observable behaviour – assent in the former case, choices in the latter. Both face the difficulty that this behaviour is the result of ‘two unknowns’, meaning and belief in the former case, probabilistic belief (subjective probability) and desire (subjective value) in the latter. Both negotiate this hurdle by holding one factor steady and solving for the other (SCT 317–19). The theory of interpretation holds belief steady by assuming that the natives share most of our beliefs; decision theory holds subjective probability steady by specifying a proposition that we can assume to have the same probability for the subject as its negation. Davidson insists that the unified theory includes rather than replaces the old project of radical interpretation, and he often reverts to the latter for ease of presentation. But he also maintains that the theory of interpretation and decision theory must be combined. Interpretation can only advance beyond observation sentences by taking into account the ‘relations of evidential support’ between them and more complex sentences. Such relations are probabilistic. Their exploration requires more than the ascription of ‘flat-out beliefs’, namely the ascription of degrees of belief. But whereas holding true is directly manifested in assent, there is no behavioural manifestation of probabilistic belief. The only way of getting at the latter is through the preferences explored by decision theory. Unfortunately, decision theory presupposes that the experimenter understands the options between which the subject is choosing. Such understanding, however, cannot be taken for granted in radical interpretation (SCT 321–2; UT 6–9; RH 331, 431). Davidson tries to overcome this difficulty by taking as basic an attitude which is a function of what the agent takes the sentence to mean, the value he sets by various states of affairs, and the probability he attaches to those states assuming the truth of the sentence. It is ‘the attitude an agent has toward two of his sentences when he prefers the truth of the one to the truth of the other’, where the sentences are supposed to be uninterpreted, in order to avoid the assumption that the subjects understand the descriptions of alternatives in the same way the experimenter does (SCT 322, 325n, Apd.).

Radical translation and radical interpretation

Davidson’s synthesis of radical translation and decision theory is ingenious and sophisticated. But it does not solve the problem of finding a truly radical, that is, presuppositionless, starting point of interpretation. Davidson insists that preferring the truth of s1 to that of s2 is a ‘simple attitude’ that an interpreter can recognize before he has ‘detailed knowledge of any of the agent’s propositional attitudes’. What the interpreter has to go on, then, is information about what episodes and situations in the world cause an agent to prefer that one rather than another sentence be true. Clearly an interpreter can know this without knowing what the sentences mean, what states of affairs the agent values, or what he believes. (SCT 322; see UT 9)

But it has not been so clear to some, especially since Davidson does not specify experimental designs for ascertaining what a subject prefers true (see Lorenz-Meyer 1993: 261–2; Hansen Soles 1999). Holding true is a two-place relation that holds between a speaker and sentences; and it is displayed in acts of assertion and assent. By contrast, preferring true is a three-place relation between the speaker and any two sentences. It is not an attitude we are familiar with, and it is not manifested in any particular speech act. Davidson tries to allay such worries. He writes that ‘every utterance that can be treated as a sincere request or demand may be taken to express the utterer’s preference that a certain sentence be true rather than its negation’, and insists that orders may be more easily detectable than acts of assent because of their closer connection to action (SCT 325n; RH 331). But without knowing what the speaker thinks or means, it will simply be unclear what preferences she is expressing by an order. Furthermore, this procedure focuses on the speech acts the agent herself performs. However, like assertions, requests and orders can be influenced by non-semantic considerations. Although my real preference is for my guest to stop smoking, out of politeness I ask him to open the window. There is no easy remedy for this problem. Whereas there is an established act of querying for assent, it is not even clear what a speech act of querying for preferring true could look like. Of course we can confront the subject with alternatives and observe how she chooses. But this does not provide a way of testing preferences between sentences. If the native chooses a first-aid box over a knife, this may express a preference for peace, for medical assistance, for a red cross on white background, etc. Normal interpreters from scratch can narrow down the possibilities. But they rely on a prior understanding that

193

194

Quine and Davidson on Language, Thought and Reality

would be unavailable if preferring true were the entering wedge of interpretation. Therefore the move to preferring true exacerbates rather than solves the problem that Davidson presupposes more than radical interpretation is strictly entitled to. 4

Charity is not enough

So far we have dealt with the initial evidence of radical interpretation. Now we must turn to the method for deriving from this evidence a theory of meaning that delivers T-sentences for all sentences of L. Radical interpretation involves three steps which are closely related to the steps of radical translation (ITI 136): 1. establishing the logical structure of L; this involves projecting on to L not just the propositional connectives, as in Quine, but also quantificational structure and hence the first-order predicate calculus (see SCT 319–20); 2. translating indexical sentences held true or false depending on discernible features of the environment (Quine’s observation sentences); 3. translating remaining sentences, which are neither logical nor observational; it is here that the patterns of probabilistic reasoning stressed in Davidson’s unified theory come into play.

All three steps revolve around the principle of charity. Davidson speaks of charity as a ‘method of interpretation’, a ‘methodological advice’, a ‘methodological presumption of rationality’ and a ‘basic methodological precept’, namely of maximizing agreement between native and interpreter (ITI 136–7, 152, 159, 168–9, 196–7). Charity provides the bridge by which the radical interpreter proceeds from his evidence – observations about the conditions under which natives assent to a sentence s – to T-sentences about the conditions under which s is true. It does not amount to a rigid set of rules, but to a collection of ‘rough maxims and methodological generalities’ that the interpreter must supplement by intelligent guesswork in particular cases (NDE 446; see RI 25). Still, within these general and rough maxims, Davidson has distinguished two main strands (SIO 211; see ITI 21, 62, 200–1; EAE 221; UT 6–7; SIO 146–7; RH 343): ‘the Principle of Correspondence’ concerning truth: the conditions which cause speakers to hold a sentence true are by and large identical with the conditions under which the sentence is true;

Radical translation and radical interpretation

‘the Principle of Coherence’ concerning (logical and probabilistic) consistency: the statements and thoughts of speakers must display a minimum degree of coherence, since thoughts are at least partly identified by their logical connections to other thoughts.

For Davidson charity is neither a virtue nor one hermeneutic procedure among others. There is no alternative to assuming that the natives have beliefs and desires that are largely true and rational, because the norms of rationality codified in logic and decision theory ‘have constitutive force’; they are constitutive of the very idea of interpretation. Charity is ‘not an option’ but ‘forced on us; whether we like it or not, if we want to understand others, we must count them right on most matters’ (ITI 197; see Ramberg 1989: ch. 6; Kunne 1993: 6). Whereas radical interpretation is sup¨ posed to be sufficient for interpretation from scratch, charity is supposed to be necessary. Furthermore, it is necessary for a priori reasons. ‘What makes interpretation possible, then, is the fact that we can dismiss a priori the chance of massive error’ (ITI 168–9; see RH 342). Davidson’s main argument for this claim trades on the interplay of meaning and belief in the explanation of assent. Assuming that the native believes what we do, he assures us, is the only way of solving this equation with two unknowns (ITI 137, 167). This can be taken to imply that we must presume every native utterance to be true, and at first Davidson seems to have taken it that way. In its original formulations, the principle required the interpreter to ‘maximize agreement’ with the natives, with respect to both truth and coherence (ITI 27, 101). Davidson soon realized, however, that this is a ‘confused ideal’ (ITI xvii, see RH 82). For one thing, the number of sentences is infinite, which means that there is no maximum of agreed sentences (ITI 136). For another, as Grandy (1973) and Lewis (1983: ch. 8) pointed out, it would be wrong to ascribe to the natives beliefs we take to be correct even in cases in which there is no explanation of how they could have acquired these beliefs. Quine later struck a similar note: I would maximize the psychological plausibility of my attributions of belief to the native, rather than the truth of the beliefs attributed. In the light of some of the natives’ outlandish rites and taboos, glaring falsity of their utterances is apt to be a psychologically more plausible interpretation than truth. (WWD 76)

In the wake of such difficulties, several proponents of radical interpretation have modified the principle of charity. All of these modifications are based on the idea that we should ascribe to the natives beliefs that it is

195

196

Quine and Davidson on Language, Thought and Reality

psychologically plausible for them to have, whether or not these beliefs coincide with ours. Orthodox Davidsonians have dismissed these modifications out of hand; and they have derided the psychological assumptions invoked as ‘labour saving heuristic aids’ which may guide actual interpretation but are irrelevant to the project of radical interpretation (Ramberg 1989: 73–4). This reaction is myopic. Contrary to Davidson’s claims, the original principle of charity cannot be a necessary precondition of interpretation from scratch, given that actual cases of such interpretation proceed without it. Furthermore, it is not even sufficient, since, as Grandy and Lewis demonstrated, it would often lead to misinterpretation. More recently, Davidson has counselled optimizing rather than maximizing agreement. We should ‘make native speakers right when plausibly possible’ (ITI 169, 137). At the same time he recognizes that it is difficult to say what ‘the right sort of agreement is’ (ITI xvii). Yet without such a specification, charity becomes vacuous (as in Lukes 1982; Wiggins 1997: 17–18). We should avoid ascribing unintelligible error, but equally unintelligible insight. As regards the question of what beliefs we can intelligibly ascribe, this takes us back to square one. Davidson also offers us something more substantial. Even after giving up on maximizing agreement, he has continued to insist on applying charity ‘across the board’ (ITI xvii, 136n). This contrasts with Quine, who restricts charity to the translation of logical constants and observation sentences, that is, to truths which are either logical or empirically evident. It suggests that radical interpretation starts by projecting all of our beliefs on to the natives, allowing for the ascription of error only in a second step (e.g. ITI 197). It is glaringly obvious, however, that this is not how we actually interpret from scratch. The Spanish conquistadors or the Australian explorers of the central valley in Papa Guinea did not start out on the assumption that the natives shared all of their beliefs, including, for example, the belief that ships can sail against the wind, or that infectious diseases are caused by micro-organisms. Consequently, charity across the board is no more a precondition of interpretation from scratch than maximizing agreement. Nor is it a promising procedure for such interpretation, since it leaves open the question of what kind of divergence can be tolerated. Grandy blazes a more promising trail with his ‘Principle of Humanity’: instead of crediting the natives with the beliefs and desires we actually have, we start out by crediting them with the beliefs and desires we would have had, if we had been in their position. This idea has echoes in

Radical translation and radical interpretation

Davidson. On the one hand, he allows that differences between interpreter and natives are acceptable in certain areas, and that observed or inferred anomalies can be taken into account (ITI 152). On the other, he insists that we need to view the natives as ‘nearly enough, as like ourselves’, and that this similarity involves not just their beliefs but also their desires (EX 18). Our interlocutor is presumed to be ‘consistent, a believer of truths, and a lover of the good’ (EAE 222). Unfortunately, even the principle of humanity faces an obstacle. How can we establish what the natives’ position is, in advance of understanding any of their beliefs and desires? According to Heal, the principle of humanity bids us to imagine the natives’ ‘life experiences’ (1997: 187). Yet this is not something we can do in the case of total strangers. The real preconditions of interpretation from scratch can only be brought into focus by two moves beyond charity. First, we can ascribe certain beliefs and desires without first having to interpret sentences, because they are manifest in non-linguistic behaviour. Contrary to Davidson’s argument in favour of charity, with respect to some utterances we can solve for meaning by holding belief fixed, without assuming their truth. Furthermore, to do this we need not rely on presumptuous hypotheses about what the natives might have learned, or on what we would have been led to believe in their place – as the principle of humanity would have it. In specific circumstances, certain thoughts are evident from particular forms of behaviour. If the natives insist on eating and drinking from leaden vessels, we cannot ascribe to them our belief that lead is highly poisonous; if they refuse to accept our offer of food and water during the day but accept it happily after sunset, we can ascribe to them the desire to fast during the daytime, a desire we may not share; and so on. This goes against Davidson’s claim that we must assume ‘general agreement of beliefs’ because ‘knowledge of beliefs comes only with the ability to interpret words’ (ITI 196). But it chimes with his own unified theory: understanding Jungle is embedded in understanding the native practice as a whole. It also leads on to the second move, which was anticipated by Wittgenstein: ‘The common behaviour of humankind is the system of reference by means of which we interpret an unknown language’ (1967b: §206). Attributing thoughts prior to linguistic interpretation is possible because of the common behaviour of humankind. Interpreting from scratch relies on convergence not just or even primarily of beliefs, but of patterns of behaviour, which in turn presupposes a framework of shared cognitive capacities and conative propensities (needs, emotions, attitudes).

197

198

Quine and Davidson on Language, Thought and Reality

Some of these shared features are part of our animal nature. This includes our needs for drink, food and shelter, our sexual drives, our inclination to protect the young and weak, our reactions to physical danger like predators or disease. It also includes perceptual capacities. We take for granted that the natives can survey the scene around them and are aware of what goes on within their perceptual range. These preconditions are interdependent. We could not recognize them, for example, as avoiding dangerous things, unless we could assume that they know that they are confronted with a knife rather than a piece of fruit. There are also more specific behavioural universals. As a matter of empirical fact some features of human behaviour – concerning gestures, facial expressions and demeanour – seem to have transcultural significance. What is more, these include features of our communicative behaviour, such as the intonation of utterances (see Izard 1969; Beldoyen 1984). Quine mentions the idea that radical translation might be based on characteristic forms of behaviour such as gestures, only to object that gestures ‘are not to be taken at face value; the Turks’ are nearly the reverse of our own’ (WO 29). This dismissal is precipitous. Although the Turkish gesture of dissent involves a vertical movement of the head, it is not nodding; and it can be recognized as a gesture of rejection, since it also involves a sound which is clearly dismissive. I cannot provide a neat transcendental deduction of the necessary preconditions of interpretation from scratch. But I hope to have shown that they evolve around anthropological assumptions that are both more specific than charity, and less demanding. We need not project most of our beliefs on to the natives. This makes room not just for counting them wrong. It may transpire that on some issues the natives not only hold different views, but that they are right, and we are wrong! In approaching a foreign text or culture, we must keep in mind the possibility that we might have something to learn. That is one lesson of the hermeneutic tradition which its Davidsonian admirers have yet to assimilate. By the same token, Davidson’s celebrated attack on the idea of a conceptual scheme (ITI ch. 13; SIO 39–41) combines insight with illusion. As Hacker (1996a) has shown, that attack involves ideas that are problematic in their own right, notably that a conceptual scheme is constituted by a set of inter-translatable languages, and that Tarski’s theory of truth establishes that there cannot be languages we are incapable of translating into our own. Freed from these trappings, the central plank seems to be that our standards of rationality are constitutive not just of the idea

Radical translation and radical interpretation

of interpretation, but also of the very ideas of thought and meaningful speech. If we cannot find a way to interpret the utterances and other behaviour of a creature as revealing a set of beliefs largely consistent and true by our own standards, we have no reason to count that creature as rational, as having beliefs, or as saying anything. (ITI 137)

We may grant Davidson that any intelligible form of behaviour must be similar to ours in some respects: ‘what is too far out of line is not thought’ (RI 25). But we have seen that the requisite similarity is confined to basic cognitive and conative features. In that case, however, there is no reason to deny that human beings might, and indeed have, employed conceptual schemes that differ from ours at least partly. In certain cases, we may be incapable of translating some of their terms, as with ancient Greek colour vocabulary (Dancy 1983), and in others translation requires that we enrich our own language, just as the ancient Greeks would have had to enrich their language in order to translate our contemporary scientific theories. Taking into account the anthropological dimension of interpretation from scratch also weakens the project of deriving the semantic and intentional from something closer to physics. There is no hope of an algorithm that will churn out interpretations of Martians and giant octopuses from physical descriptions of them or of their behaviour, irrespective of what they share with us. In understanding people we must adopt a perspective that is neither physical, nor even distal, but anthropological, or perhaps just human. Finally, if we take such a perspective for granted, the indeterminacy of interpretation can be defused. That, at any rate, is what I shall argue in the next chapter.

199

7

Indeterminacies

According to Quine, the indeterminacy of translation shows that there are no criteria of identity for meanings, and hence that our intuitive notion of meaning is vacuous (PL 1–2, 67–8; OR 23; PT 37, 52–3). This in turn spells ruin not just for analyticity, but for intensional notions (proposition, attribute, necessity, etc.) in general. It also undermines the ‘traditional semantics’ which revolved around these notions. Quine variously characterizes this semantics as ‘mentalistic’, on account of its commitment to meaning as something beyond behaviour, as ‘absolutist’, on account of its presupposing separate and distinct meanings, and as ‘introspective’, on account of its uncritical acceptance of our intuitive notion of meaning (RR 36; MVD 86; RHS 493; ITA 9). Indeterminacy also threatens the idea that propositional attitudes have a determinate content. What we believe and desire is expressed by sentences. If it is pointless to ask whether a sentence the native assents to means ‘p’ or ‘q’, it is also pointless to ask whether she believes that p or that q (RIT 180–1). Since beliefs and desires are individuated by their content, this will demolish what is known as ‘intentional psychology’, our pervasive practice of explaining the behaviour of human beings by reference to their beliefs and desires. Quine has welcomed this consequence. He accepts the ‘Brentano thesis’, according to which our intentional and intensional statements cannot be reduced to the purely extensional statements of the physical sciences. To him, however, this demonstrates not ‘the indispensability of intentional idioms and the importance of an autonomous science of intention’ but rather the baselessness of intentional idioms and the emptiness of a science of intention . . . If we are limning the true and ultimate structure of reality, the canonical scheme for us is the austere scheme that knows . . .

[200]

Indeterminacies

no propositional attitudes but only the physical constitution and behavior of organisms. (WO 221; see WP ch. 21; MVD)

There are various ways in which these momentous conclusions might be resisted. In this chapter I argue that some of them fail, but that indeterminacy can nevertheless be avoided, partly by adopting the less austere and more realistic perspective on linguistic understanding recommended in the last chapter. In section 1, I deny that radical translation and hence indeterminacy starts at home and defend our first-person authority concerning speaker’s meaning and reference. Section 2 defends the indeterminacy thesis against three objections, that it is self-refuting, that it confuses ontological and epistemological questions, and that it wrongly presupposes a third-person perspective on meaning. But although meaning cannot transcend observable behaviour, it transcends radical translation and interpretation, since the latter needlessly restrict the semantically relevant features of linguistic behaviour and impose unwarranted methodological restrictions. The next two sections attack the inscrutability of reference derived from the argument from below by questioning two of these restrictions. Section 3 argues that gavagai reinterpretations fall foul of anthropological facts about human thoughts and perceptual capacities. Section 4 maintains that the proxy function argument unduly restricts the semantic evidence to the conditions under which speakers assent to whole sentences. The final section argues that the argument from above fails, because the indeterminacy of translation either reduces to a harmless underdetermination of theory by evidence or relies on a dogmatic physicalism. Not only are there facts of the matter concerning meaning and reference (contra ontological indeterminacy), mutual communication with the natives allows us to establish what they are (contra epistemological indeterminacy).

1

Does radical translation start at home?

According to Ontological Relativity, the whirlpool of semantic uncertainties does not end with the indeterminacy of meaning and the inscrutability of reference. There is yet another dimension of indeterminacy, namely ‘ontological relativity’. Understanding a language – determining its meanings and ontological imports – is relative not only to one of several possible translation manuals, but also to the choice of one of several possible languages to translate into. We are forced to project the ontology of

201

202

Quine and Davidson on Language, Thought and Reality

some ‘background language’ or ‘theory’ on to the native language (OR 49, 67–8; PL 81–2). Finally, ‘radical translation begins at home’ (OR 46; see WO 59, 78; RR 83–4). Even understanding one’s own language, indeed one’s own utterances, involves a process of translation. To be sure, we usually adopt a ‘homophonic translation’ in such cases. For example, we translate ‘The cat is on the mat’ by ‘The cat is on the mat’. But occasionally the principle of charity bids us to translate English utterances ‘heterophonically’, to avoid ascription of absurd beliefs. I radically translate even my very own words. As a result, it is indeterminate what my utterances mean, or what I am talking about. In his early writings, Davidson toed a similar line.1 He treated the terms ‘understanding’ and ‘interpreting’ as interchangeable (e.g. ITI 125, 157, 169). Furthermore, he stated explicitly that the ‘problem of interpretation is domestic as well as foreign: it surfaces for speakers of the same language in the form of the question, how can it be determined that the language is the same?’, and he concluded that ‘all understanding of the speech of another involves radical interpretation’ (ITI 125, see 239, 276; SIO 12–13). If you say ‘The door is shut’, I have to interpret your utterance no less than one from an alien language. It is just that the outcome is likely to be a homophonic interpretation expressed in a disquotational T-sentence, namely ‘“The door is shut” is true iff the door is shut’. There is a striking parallel here between Quine and Davidson on the one hand, and the philosophical hermeneutics of Gadamer and Apel on the other. Both positions start from an exceptional situation that poses special difficulties to linguistic understanding – the encounter of an alien tribe in the former case, the reading of remote texts like the Iliad or the Bible in the latter. This special problem is then taken to be general, to be part of the human condition. In both cases the result is a striking picture of human beings as constantly engaged in complex semantic activities like translation and interpretation. Understanding, even in 1. Two caveats. First, Davidson later accepted first-person authority, and presented radical interpretation not as our actual method of communication, but only as a feasible one. Secondly, while he stated explicitly that all understanding involves radical interpretation, the claim that he identifies the latter with translation is controversial. Unlike radical translation, radical interpretation does not just correlate quoted sentences of the source language with quoted sentences of the target sentence, but specifies the truth-conditions of the former. At the same time ‘interpretation is . . . essentially translation’ because it is ‘interpretation in one idiom of talk in another’ (EAE 253; ITI 126, see also 57, 238; SAL 4), namely of the object language of a truth-theory in its metalanguage, although this fact may be disguised if the object language is part of the metalanguage (see Glock 1993: 200–5; Glock 1995; Alvarez 1994 and chapter 8, section 5 below).

Indeterminacies

the domestic case, and indeed one’s own case, is supposed to rest on translation/interpretation. In their ordinary employment, the terms ‘understanding’ and ‘translating’ or ‘interpreting’ are not equivalent. Translating means expressing what the sentences or words of one language mean in terms belonging to another. Similarly, linguistic interpretation is a matter of expressing obscure sentences or passages in a different and, one hopes, clearer way. Quine and Davidson tend to ignore ordinary parlance on this issue, because they hold that there is a sense in which understanding and meaning presuppose translation/interpretation. Quine subscribed to what may be called the thesis of the indispensability of translation: ‘It makes no sense to say what the objects of a theory are, beyond saying how to interpret or reinterpret that theory in another’ (OR 50; see Romanos 1983: 92). Any sense we can make of the idea that the expressions of a language L1 have linguistic meaning and reference derives from the possibility of translating them into a background language L2 . The indispensability thesis would explain why all understanding, even that of my own utterances, must be based on radical translation. But it also invites a vicious regress that was first spotted by Davidson. He agreed with Quine as regards the indeterminacy of translation and the inscrutability of reference, yet he did not accept ontological relativity. However, the reason is not that Davidson detected semantic bedrock here. On the contrary, he urged us to ‘stand firm’ against the tempting thought that, vagaries of translation notwithstanding, ‘at least the speaker knows what he is referring to’ (ITI 235). His complaint was that, for once, Quine has understated his paradoxical case. The point should not be that we can ascribe an ontology to a language only relatively to a background language. Rather, given that we can understand the background language itself only through translation into yet another language we are stuck with a vicious regress. This means that ‘any claim about reference, however many times relativized, will be as meaningless as “Socrates is taller than”’ (ITI 233–4). Davidson is right to draw this conclusion, but wrong to think that Quine has shied away from it. Quine also concludes that even in our own case ‘there is no fact of the matter’ as to what our words refer to (OR 47). Consequently, we end up not with some form of relativism, but with a referential nihilism. However, Quine cannot sustain such nihilism, because it concerns an extensional notion he continues to employ. For this reason, he tries to escape the ‘quandary’ of ontological relativity. Allegedly, the regress of translation comes to an end by ‘acquiescing in our mother

203

204

Quine and Davidson on Language, Thought and Reality

tongue and taking its words at face value’ (OR 48–9). It is unclear what Quine means by ‘acquiescing’. If it is a matter of simply employing the terms of one’s own language, it will not get him out of the quandary. The very point of his doctrine of ontological commitment was that ‘reflective persons unswayed by wishful thinking . . . have cause to wonder what, if anything, they are talking about’ (WO 242). But the point of ontological relativity is that this question is not only unanswerable but vacuous. In more recent writings, Quine tried to rectify matters. For one thing, he disowned the distinction between inscrutability of reference and ontological relativity. Reference is not doubly relative, to a translation manual and to a background language, but only to the former. For another, he suggested that acquiescence is a matter not of simply using our home language, but of furnishing disquotational explanations of reference. If, instead of making mischief through proxy functions, we choose as our manual of translation the identity transformation, thus taking the home language at face value, the relativity is resolved. Reference is then explicated in paradigms analogous to Tarski’s truth paradigm; thus ‘rabbit’ denotes rabbits, whatever they are, and ‘Boston’ designates Boston. (TI 6; see PT 51–2; TT 19–20, 54; RHS 367)

But this quote also shows that Quine’s change of mind was half-hearted. It still suggests that we do not really know what the things are that our term ‘rabbit’ denotes. Yet we do know; it is just that this knowledge is expressed not in denotational truisms but in genuine explanations, such as ‘“rabbit” refers to a species of rodents’ or ‘The animal over there is a rabbit’. Moreover, Quine’s recantation does not go to the root of the regress, namely the indispensability thesis. That thesis is incoherent. Understanding does not presuppose translation, but the other way around. One cannot translate something one does not understand. To translate e1 from L1 is to find an expression from L2 that means the same, and this can be done only by someone who knows what e1 means. That is why translations must be done by people who already understand the language of the text to be translated, and why one cannot translate a text in cipher without having deciphered it. It is equally absurd to maintain that meaning presupposes translation. Translating an expression e1 from L1 by an expression e2 from L2 is legitimate only if e2 means roughly the same in L2 as e1 does in L1 . Consequently, the very notion of translation presupposes that the expressions to be translated are meaningful independently of translation, namely by

Indeterminacies

virtue of being used and explained in their home language. If one denies that words have meaning prior to translation, talk of translation becomes unintelligible. The indispensability thesis puts the cart (translation or interpretation) before the horse (meaning). Far from translation being a precondition of meaning and reference, the latter are preconditions of the former. This argument also militates against the interpretationalism one occasionally finds in Davidson: ‘it is understanding that gives life to meaning, not the other way around’ (SAL 12). Taken literally, this is mistaken. If A uses e1 , there is something for B to understand only if e1 is meaningful, either because it has a literal meaning, or because A endows it with a particular speaker’s meaning on this occasion. On a more charitable interpretation, Davidson’s claim is part of his increasingly vociferous support for the idea that meaning is use (IA 41; RH 80; LL 298). Linguistic expressions do not have a meaning independently of being used within a process of communication, and that process essentially involves understanding. On this interpretation, he can acknowledge that B’s understanding must pay heed to something independent, if not to the conventional meaning of e1 , then at least to the kind of speaker’s intentions invoked by Grice (see SCT 310–11; SIO 111–12; SAL 11–12 and chapter 8, section 4 below). Like Quine, Davidson maintains that understanding must involve interpretation: ‘speaker and hearer must repeatedly, intentionally, and with mutual agreement, interpret relevantly similar sound patterns of the speaker in the same way’ (ITI 277). But his rationale is different. He thinks that this is required by another feature of understanding, namely its defeasibility – it is always possible that our initial understanding needs to be corrected: ‘however easily, automatically, unreflectively and successfully a hearer understands a speaker, he is liable to general and serious error. In this special sense, he may always be regarded as interpreting a speaker’ (FPA 110). This is a very special, and indeed a misleading sense. That understanding always involves interpretation now means that it may come to require it. Obviously this does not show that understanding involves interpretation in the ordinary sense to which the other passage seems to appeal. To suppose otherwise would be like inferring from the fact that a machine is always liable to break down that when it is running smoothly it must continuously be under repair. Furthermore, Davidson mischaracterizes the exceptional cases in which the need for interpretation does arise. Take his famous example:

205

206

Quine and Davidson on Language, Thought and Reality

I see a ketch sailing by, and my companion says ‘Look at that handsome yawl!’ Certainly, some remedial action is called for here. But it is not what Quine calls ‘heterophonic translation’ or the ‘off the cuff interpretation’ based on a ‘viable theory of belief and meaning’ Davidson envisages (ITI 196, see 100–1). The way to achieve understanding is to demand an explanation from the speaker. In our case I might ask him whether by a yawl he meant a two-masted boat. And his explanation will have an authoritative status. I may point out that such a vessel is usually called a ketch, or ridicule him for mistaking a two-masted boat for a one-masted one at such close range. But it makes no sense to override his verdict on what he meant by ‘yawl’ on this occasion. Consequently, the way to clarify this type of obscure utterance is not interpretation, let alone translation, but speakers’ explanations. Genuine problems of interpretation typically arise when and only when the speaker or writer is unavailable for comment. Speaker’s explanations are crucial because we possess first-person authority concerning speaker’s meaning and speaker’s reference. I cannot be ignorant or mistaken about what I mean by my words, or about how I understand the words of others, in the way in which I can be ignorant or mistaken about the material world, or about what others mean by their words, or about the literal meaning of words in a particular language. For this reason I sympathize with Searle’s claim (1987) that the denial of this first-person authority constitutes a reductio ad absurdum of Quine’s behaviourism. Quine might respond by dismissing this authority as part of the traditional concept of meaning he is repudiating (Malachowski 1988). Precisely because it is a constitutive part of our semantic concepts, however, he would then no longer be talking about meaning as commonly understood. Davidson is committed to elucidating rather than eliminating our established concepts, and has therefore started to defend firstperson authority. The real question is whether this authority can be used to undermine indeterminacy not just in the first-person case, but in general. Searle answers this question in the affirmative. Because I know there to be a difference between my referring to rabbits and my referring to undetached rabbit parts, I can infer that there must also be such a difference when it comes to what the native means by ‘gavagai’. In my opinion, the matter is more complex. As I have argued, there are fundamental differences between the first-person case, the domestic case, and genuine translation. We have no authority concerning either the utterances of others or conventional meaning, especially in the case of an alien tongue. So why should the

Indeterminacies

difference I draw between rabbits and rabbit parts carry over to meaning in general? Worse still, Quine and Davidson suggest with some plausibility that our semantic concepts, speaker’s meaning included, ultimately derive their content from the practice of linguistic communication. If that practice left indeterminate what others mean, and consequently what expressions mean, this would provide a rationale for dismissing the whole cluster, and would therefore undermine the appeal to first-person authority. For these reasons, the fact that radical translation and indeterminacy do not start at home does not defuse the general threat of indeterminacy.

2

Unsuccessful objections to indeterminacy

As regards conventional meaning and the speaker’s meaning of others, Quine’s strategic line of reasoning can be reconstructed as follows: (P1 ) All the semantically relevant facts are captured by radical translation, that is, by a complete account of how the natives’ dispositions to verbal behaviour depend on their environment. (P2 ) These behavioural facts leave underdetermined what expressions mean; that is to say, they do not discriminate between translation manuals that assign different meanings to one and the same expression. (C1 ) There are no criteria of identity for meanings/intensions. (P3 ) No entity without criteria of identity. (C2 ) There are no meanings/intensions.

This section probes criticisms of the argument that I consider to be ultimately uncompelling. The first is the charge of incoherence. Although Quine’s argument is supposed to undermine intensional notions, (P2 ) employs such notions. And how could different translation manuals that fit the behavioural evidence equally well be incompatible, except by ascribing different senses to one and the same sentence of the target language? Quine suggests that the English sentences s
and s ∗ prescribed as translations of a single Jungle sentence s by incompatible manuals ‘stand to each other in no plausible sort of equivalence however loose’ and are ‘not interchangeable in an English context’ (WO 27; PT 48; see RDH 296–7). However, this only solves the problem if this difference between s
and s ∗ can be captured in purely extensional terms. If s is an observation sentence, s
and s ∗ must have the same truth-value and indeed the same

207

208

Quine and Davidson on Language, Thought and Reality

truth-conditions, because both must be accepted by English speakers under the same conditions as those under which the natives accept s. This holds for ‘Lo, a rabbit’ and ‘Lo, an undetached rabbit part’. At the same time, Quine has maintained that if s is theoretical, s
and s ∗ could differ in truthvalue (see WO 73–4; SS 75). Harman (1999: 131–4) has tried to illustrate this possibility. The settheoretic translations of number theory by von Neumann and Zermelo are equally compatible with the evidence, since true sentences of the source language are translated into truths of the respective background languages. Yet they are incompatible with each other: von Neumann’s translation of, for example, ‘3 < 5’, namely ‘3  5’, is true in his system but false in Zermelo’s. As Quine himself pointed out, however, this example fails (RIT 183). It is not indeterminate whether ‘3  5’ is an adequate translation of ‘3 < 5’, because most people who accept the latter would rate the former as nonsense. In my view, therefore, the most auspicious response is to concede that the indeterminacy thesis uses the traditional semantic notions it debunks, while insisting that it does so only negatively. It purports to spell out a consequence of intensional notions that is unacceptable to the intensionalist. We think that s
and s∗ differ in meaning, but from a proper scientific perspective that relies exclusively on behavioural evidence this difference turns out to be spurious. If an incoherence lurks here, it concerns C2 , the conclusion that there is no such thing as meaning. This semantic nihilism appears to be selfdefeating in a straightforward way: if it is true it must be meaningless. This charge deserves to be taken more seriously than is fashionable at present. Semantic nihilists must not be spared the awkward question of what on earth they think they are doing in enunciating their position. Indeed, Quine himself accepts that self-refuting statements are to be avoided. His resolution of the ‘Platonic riddle of nonbeing’ takes for granted that negative existential statements must not be understood as self-defeating (see chapter 1, section 1 ). But is it fair to characterize Quine as a semantic nihilist? The label correctly indicates that he is not a semantic sceptic, who holds that semantic properties, though real, transcend our knowledge. Nor need it be weakened to ‘intensional nihilism’, since the inscrutability of reference impugns an extensional property. On the other hand, the ‘critique of meaning leveled by my thesis of indeterminacy is meant to clear away misconceptions, but the result is not nihilism. . . . Indeterminacy means

Indeterminacies

not that there is no acceptable translation, but that there are many’ (ITA 9; see LAP 117; RHS 429). Still, these translations are acceptable in that they have the same stimulus meaning as the original. As Quine himself never tires of stressing, stimulus meaning is only a ‘behavioristic ersatz’ of meaning as normally conceived (WO 66). Meaning proper is indeterminate and hence, given the dictum ‘no entity without identity’, unreal. Quine is a nihilist about key semantic notions employed by both ordinary speakers and traditional semantics. ‘The theory of meaning . . . strikes me as in a comparable state to theology – but with the difference that its notions are blithely used in the supposedly most scientific and hard-headed branches of philosophy’ (SAO 92). These are strong words, and they had better be backed by knock-down arguments. Yet they are not automatically self-refuting. Quine can protest that the notion of meaning he rejects is not an indispensable one which is abandoned at the cost of intelligibility, but a fantastic if venerable figment of the imagination. The charge of self-refutation is compelling only if it can be shown both that Quine’s extensional ersatz is inadequate to account for meaningful discourse and that there is nothing wrong with genuine meaning. My objections against his behaviourist approach to language and translation from scratch (chapter 3, section 4 and chapter 7, section 2) support the first conjunct. But in order to support the second, we must refute the arguments in favour of indeterminacy. One attempted refutation is directed against (P3 ). It maintains that postulating intensions does not require criteria of synonymy. Even if there is no way of establishing whether two expressions mean the same, this epistemological result is irrelevant to the ontological question whether meanings exist (e.g. Katz 1990: 182–3). By contrast, Quine insists that indeterminacy is an ontological rather than an epistemological thesis. It concerns the question whether a naturalist can accept meanings as part of reality. As indicated in chapter 3, section 1, both Quine and his opponents are wrong in so far as they assume that terms like ‘the meaning of x’ or ‘what x means’ would have to refer to abstract entities with which competent speakers are acquainted (though they are objects of discourse in the innocuous sense explained in chapter 2, section 5). Yet Quine’s position does not depend on this assumption, and it can be rephrased as follows. If the scientifically respectable methods leave translation indeterminate, then there is simply ‘no fact of the matter’ as to what expressions mean, and hence no fact for the translator to be right, wrong or ignorant about.

209

210

Quine and Davidson on Language, Thought and Reality

Intensional discourse lacks ‘factual content’ or ‘factuality’ (FLPV 63; WO 71–3; OR 47, 103; FM 168; TT 23; RHS 187, 429). For naturalism the question whether two expressions are alike or unlike in meaning has no determinate answer, known or unknown, except insofar as the question is settled in principle by people’s speech dispositions, known or unknown. If by these standards there are indeterminate cases, so much the worse for the terminology of meaning and likeness of meaning. (OR 29)

Our discussion of criteria of identity supports Quine. Traditional semantic discourse may not require formal or cast-iron principles for counting ‘meanings’. But it is without content, if the semantically relevant facts leave open entirely what an expression means or whether two expressions mean the same (where these notions are understood in the weighty traditional sense). Ontological table-thumping is futile here. Quine’s opponents cannot reply: ‘We don’t care whether it makes sense to talk about meanings; we want to know whether meanings exist!’ Ex hypothesis, that question would be vacuous; an answer to it would be neither feasible nor required. This defence invites an objection to (P1 ), namely that the facts which underlie meaning (properly conceived) are not exhausted by the behavioural facts radical translation reckons with (Chomsky 1969; Chomsky and Katz 1974). This complaint is ultimately justified. But it is important to pinpoint precisely where Quine’s behaviourist methodology and the associated conception of meaning go astray. Quine stresses both ‘the behavioural and the physiological’ (FM 168). On his view, the facts relevant to meaning must, firstly, be accessible to observation and, secondly, describable in the language of physics. The Chomskians believe that he is wrong on the first and right on the second. I shall adopt the opposite view, and defend Quine’s and Davidson’s third-person perspective. Quine and Davidson are right to reject the idea that the meaning and reference of our words could completely transcend our powers of recognition. Semantic facts cannot be verification-transcendent in a strong sense, namely such that it is logically or conceptually impossible to ascertain them. The idea of a Ding an sich may be nonsense; the idea of a Bedeutung an sich is definitely nonsense on stilts. Any recognizable notion of linguistic meaning is inextricably linked to what Davidson calls ‘the communicative content of words’. The meaning of an expression is what competent speakers understand, and what they explain to neophytes; it also determines what they say in using the expression. Consequently, ‘the correct

Indeterminacies

interpretation of one person’s speech by another must in principle be possible’ (RH 80; SCT 315; see SAL 12–13). But even if meaning must be accessible in principle, it might be hidden from a radical translator whose evidence is confined to potential linguistic behaviour and excludes, for example, psychological or physiological findings about ‘things that go on in people’s heads’ (Fodor and Lepore 1992: 80–1). Quine and Davidson resist this suggestion by appeal to the ‘social’ nature of language. Language is a social art which we all acquire on the evidence solely of other people’s overt behavior under publicly recognizable circumstances. (OR 26; see WO ix; TT 192) Language is intrinsically social. This does not entail that truth and meaning are defined in terms of observable behavior, or that it is ‘nothing but’ observable behavior; but it does imply that meaning is entirely determined by observable behavior, even readily observable behavior. That meanings are decipherable is not a matter of luck; public availability is a constitutive aspect of language. (SCT 314)

Several commentators have maintained that Quine derives his behaviourist strictures against hidden meanings from his naturalism (Friedman 1975; Nimtz 2002: ch. 5.4–6). They rightly point out that this argument assumes that the only scientific facts about language concern behavioural dispositions, and that this assumption is unsubstantiated and implausible. But Quine’s defence of a third-person perspective relies not just on an unwarranted restriction of the evidence available to naturalists, but also on a genuine feature of linguistic meaning. In defending the claim that meaning must be publicly observable, Quine and Davidson often run together two distinct arguments. One is that observable behaviour is all we have to go by in acquiring language. The other is independent of such genetic considerations. It concerns not how we learn language, but what we learn. In psychology one may or may not be a behaviorist, but in linguistics one has no choice. . . . As long as our command of our language fits all external checkpoints, where our utterance or our reaction to someone else’s utterance can be appraised in the light of some shared situation, so long all is well. Our mental life between checkpoints is indifferent to our rating as a master of the language. There is nothing in linguistic meaning beyond what is to be gleaned from overt behavior in observable circumstances. (PT 37–8; see TT 192; ITA 5)

211

212

Quine and Davidson on Language, Thought and Reality

Whether we have learned a language depends not on unobservable mental or neurophysiological processes, but only on our dispositions to verbal behaviour. As regards communication, any mental accompaniment ‘is as may be’, an idle wheel, as Wittgenstein put it, which can be bypassed (RR 35; see WO §2; SS 6). What we communicate is independent of ‘any idiosyncrasy in an individual’s imagery or associations that is not discovered in his behavior’. The semantics of a speaker is not determined by ‘his mind beyond what might be implicit in his dispositions to overt behavior’ (OR 27; see OR 81; PPLT 4). This view sits uneasily with some other views advanced by our protagonists, notably Quine’s neural definition of stimulus meaning and Davidson’s flirtation with externalism (chapter 7, section 3 and chapter 8, section 5). But it is essentially correct. Both the concept of literal meaning and the concept of speaker’s meaning are tied to linguistic use rather than to its mental or physiological accompaniments. Even if the image of a chocolate cake were to cross my mind whenever I use the word ‘grandmother’, it would not follow that ‘grandmother’ means chocolate cake, or even that I mean chocolate cake whenever I use the word. Similarly, even if my utterances of ‘grandmother’ were accompanied by the kind of brain activity that others display when they say ‘chocolate cake’ or think about chocolate cakes, it would not follow that I mean chocolate by that term. Conversely, even if my utterances of ‘grandmother’ were accompanied by the same brain activity as those of others, I could mean chocolate cake by that term, if I applied and explained it accordingly. Nevertheless, many contemporary philosophers reject the idea that meaning must be manifest in an individual’s observable behaviour by appeal to externalist theories of the kind pioneered by Putnam and Burge. Like the third-person perspective, externalism denies that the content of an individual A’s thoughts and utterances is exclusively determined by her intrinsic (mental or physiological) properties. But it appeals to facts beyond A’s current behavioural dispositions, in particular to the history of A’s causal interaction with her physical and social environment. A sustained discussion of externalism is beyond the present purview (see Glock and Preston 1995). But in so far as externalism is incompatible with the manifestability of meaning, it is also incompatible with the everyday notion of linguistic meaning, and hence cannot be invoked to protect that notion against indeterminacy. If the external facts are supposed to be hidden from ordinary observers, the consequence will be semantic scepticism: competent speakers are ignorant of what familiar

Indeterminacies

English expressions mean; indeed, they are even ignorant of what they mean by these expressions. This last point is incompatible with our first-person authority concerning speaker’s meaning; and the first is once more at odds with the idea of meaning as the content of understanding. It implies that, for all we know, genuine communication and understanding may never take place, since even competent speakers can be largely ignorant of what their words mean. Yet, as Davidson reminds us, ‘all of us do understand some speakers of some languages’. Consequently, the facts which determine meaning must be ‘available, not only in principle, but available in practice to anyone who is capable of understanding the speaker or speakers of the language’ (SCT 301; see RD 82). It is simply absurd to suggest that our term ‘teddy-bear’ has a real meaning, and might refer to a type of supernova, if the latter stood in a certain causal relation to our utterances. Externalists might insist that they are not bound by these truisms, since for them meaning is a technical notion unrestricted by ordinary parlance. But in that case their stipulations are irrelevant to any notion of meaning that retains a connection with communication and understanding. Quine and Davidson are concerned with meaning in that sense, which is also the sense in which we speak about meaning, both in everyday life and in linguistics. They are right to insist that meaning in this sense must be manifestable in observable behaviour and in adopting a third-person perspective. As a matter of principle, then, meaning, and by its connection with meaning, belief also, are open to public determination. I shall take advantage of this fact in what follows and adopt the stance of a radical interpreter when asking about the nature of belief. What a fully informed interpreter could learn about what a speaker means is all there is to learn; the same goes for what the speaker believes. (SIO 147–8, see 318; SIO 99–100; MTE 78; RI 14)2

But the manifestability of meaning does not entail that all the relevant facts are captured by radical translation/interpretation. This would only follow if the radical translator/interpreter had access to all of the publicly 2. This passage is incompatible with Davidson’s vehement insistence that he never maintained that the evidence allowed by radical interpretation exhausts all of the legitimate evidence for interpreting a language (SR 12–14; RD 81). Unless, that is, there are ‘fully informed interpreters’ who proceed on public evidence other than that available to the ‘radical interpreter’. This might well be Davidson’s position, since he maintains that radical interpretation is only a possible way of establishing what speakers mean, rather than the one we actually employ (see chapter 8, section 3).

213

214

Quine and Davidson on Language, Thought and Reality

observable evidence relevant to meaning, and to all the proper methods. I have already indicated that this is not the case: a. Quine conceives of linguistic behaviour in purely mechanistic rather than intensional and intentional terms; b. He also assumes, mistakenly, that translation must be either presuppositionless or indeterminate; c. Both Quine and Davidson ignore that radical translator and native engage in communication and mutual instruction; d. They also ignore that this kind of interaction presupposes more than charity, namely a common human nature between those involved; e. Finally, they go wrong in holding that only the assent to or dissent from sentences is semantically relevant, and that the reference of sentence components is a mere epiphenomenon.

As argued above, Quine and Davidson cannot furnish any interpretation from scratch without tacitly violating these restrictions. Now I want to show how indeterminacy can be avoided by shedding these restrictions, starting with (d) in the next section, turning to (e) in section 4, and to (b) and (c) in section 5.

3

Scrutinizing reference I: gavagai

Davidson agrees with Quine that translation/interpretation is indeterminate. For any natural language L, ‘a number of significantly different theories of truth will fit the evidence equally well’ (ITI 62, see 230). Again like Quine, he regards this indeterminacy not just as an epistemological matter, but as showing that ‘there is no fact of the matter as to what someone means by his or her words’ (SIO 169). Davidson detects two sources of indeterminacy. The first is the inscrutability of reference. The second is the ‘blurring of the distinction between the analytic and the synthetic’ (IA 6) that results from the fact that we can alter translations through ‘trade-offs between the beliefs we attribute to a speaker and the interpretations we give to his words’ (ITI 139). For instance, if someone points at a whale shark with the words ‘There’s a whale’, should we assume that she believes the ostended animal to be a mammal, or are we to conclude that by ‘whale’ she means something like ‘very large fish’? ‘Having a language and knowing a good deal about the world’ cannot be sharply separated (EAE 257; see ITI 100–1, 196–7, 199–201; SIO 146–7).

Indeterminacies

At the same time, Davidson thinks that indeterminacy is less pronounced than Quine concludes. As regards the inextricability of linguistic understanding and factual knowledge, applying charity ‘across the board’ reduces our leeway in translating even theoretical sentences: it fixes the native’s beliefs by assimilating them to ours as far as possible (ITI 228, but cf. 27). Davidson’s unified theory of meaning would further facilitate translation of theoretical sentences by allowing us to trace their evidential connections to observation sentences (IA 6; RH 82). I have suggested that neither universal charity nor the unified theory can be taken for granted. But my assumption of a common human nature also rules out the ascription of certain wildly divergent beliefs and desires. Once we can communicate with the natives, moreover, linguistic and factual aspects can be extricated in the same way as in the domestic case, namely by asking the speaker what she means, and by noting what uses of a term she regards as intelligible. What remains is not a threat to determinacy, or even the analytic/synthetic distinction, but only the possibility that the status of some statements may be left open in particular contexts (see chapter 3, section 4). This leaves the inscrutability of reference. For Davidson, inscrutability is complete with respect to assigning reference to sentence components (words and phrases), yet it does not extend to logical form and in particular to the quantifiers: ‘correct theories will agree on the whole about the quantificational structure to be assigned to a given sentence’. Davidson suggests two reasons for this optimism. The first is that ‘the semantic constraints’ on interpretation, namely that it must take the form of a ‘Tarskistyle theory of truth’, forces Jungle into the ‘procrustean bed of quantification theory’ (ITI 151, 136n, 228). This justification is uncompelling. Even if a theory of meaning can be furnished by constructing a truth-theory, there is no reason for thinking that this is the only possible way. Worse still, if Quine’s case for inscrutability is correct, by imposing logical form a Tarskian truth-theory will simply exceed both the available evidence and the matters of fact. Davidson’s second reason may weaken that case. Quine regards quantificational logical form as inscrutable, because what ‘we can pin down to behavioural criteria’ concerns sentences as a whole, namely their stimulus meanings and their combination through truth-functional connectives. Other syntactic features of Jungle, notably the apparatus of reference, including the distinction between singular and general terms and the quantifiers, are established through analytic hypotheses and

215

216

Quine and Davidson on Language, Thought and Reality

hence indeterminate (WO 61, see 52–5, 68–70). Davidson accepts that the ‘constraints must deal only with sentences and truth’, yet he believes that quantificational structure can be established on the basis of ‘observed patterns of inference’ similar to those that reveal sentential connectives (ITI 151; RH 81–2). Assume that the natives assent to ‘Roha gavagai’ if and only if they assent to ‘Gavagai’ with respect to all the animals in sight. In that case, aren’t we entitled to provisionally translate that sentence as ‘All of them are rabbits’ and to equate ‘roha’ with our universal quantifier? Quine would not accept this procedure. His doctrine of ontological relativity implies the possibility of radically different ways of parsing simple sentences. This precludes any determinate mapping of parts of their sentences onto ours, with the result that we cannot even be sure that the natives are referring to or committed to certain entities (OR ch. 2; Romanos 1983: 43–8). But, Davidson remonstrates, in that case there is no reason to think that the natives are talking about anything, and hence to regard their verbal behaviour as a language to begin with (RH 81–2; see ITI 228–9). As Davidson realizes, however (ITI 152), even if sound, this argument leaves intact a weaker yet no less disturbing inscrutability, namely that the evidence licenses the ascription of radically diverse ontologies or schemes of reference. We must therefore take a closer look at the two lines of reasoning Quine advances in favour of this kind of inscrutability. The first one concerns the possibility of producing alternative interpretations of specific terms like ‘gavagai’. According to Quine’s alternatives (undetached rabbit part, temporal stage of a rabbit, part of a rabbit fusion, instantiation of rabbithood, etc.), ‘gavagai’ applies to something within the same spatio-temporal region as rabbits. Whenever there is a rabbit in the vicinity, there is also an undetached rabbit part. According to Quine, this ensures that the term cannot be disambiguated through ostension. Any attempt to establish precisely what the native points at hinges on a prior understanding of the apparatus of individuation, which in turn presupposes a translation of terms like ‘gavagai’ and thus leads to indeterminacy (WO 52–3; OR 30–5, 39–40). Evans (1985: ch. 2) has raised a sophisticated objection to this reasoning. Quine only considers cases in which ‘gavagai’ is combined with the apparatus of individuation, thereby creating the option of combining non-standard translations of ‘gavagai’ with non-standard translations of that apparatus. Evans urges us to consider instead how ‘gavagai’ interacts with other predicates and with negation. Assume that we have established that the term

Indeterminacies

‘orith’ has the same stimulus meaning as ‘white’. In that case, the evidence can favour a standard axiom like (I) an object x satisfies ‘gavagai’ iff x is a rabbit

over the non-standard (I
) an object x satisfies ‘gavagai’ iff x is an undetached rabbit part

in conjunction with (II) an object x satisfies ‘orith’ iff x is white (I
) leads to the following theorem (III) an object x satisfies ‘orith gavagai’ iff x is a white undetached rabbit part.

But that theorem is excluded if the natives assent to ‘orith gavagai’ only in the presence of a rabbit that is white all over, while dissenting to it in the presence, for example, of an otherwise black rabbit with a white paw. Quine cannot rectify matters by proposing an alternative axiom for ‘orith’ such as (II
) an object x satisfies ‘orith’ iff x is part of a white animal

since this is excluded if the natives apply ‘orith’ to, for example, white sheets of paper or white flags. Hookway has rushed to Quine’s defence here. He points out that the use of ‘orith’ in conjunction with other predicates might differ systematically from its use in isolation (1988: 154–5). All the evidence considered so far is accounted for by the following axiom (II∗ ) an object x satisfies ‘orith’ iff

either or

‘orith’ occurs together with ‘gavagai’ and x is part of a white animal ‘orith’ occurs in some other context and x is white

Still, Evans’ argument shows that a reinterpretation of ‘gavagai’ necessitates a reinterpretation of countless other terms, and of the ways in which the terms and sentences of Jungle can be combined. One may well doubt that such a universal reinterpretation could remain faithful to the assent conditions of all Jungle sentences. In fact, no one has ever provided a sustained alternative to the established translation manuals between two

217

218

Quine and Davidson on Language, Thought and Reality

natural languages.3 Quine excuses this failure by stating that there is no motive for such a duplicate translation (WO 72; ITA 9). But surely, vindicating the inscrutability of reference and hence the indeterminacy of translation ought to be incentive enough. Indeed, the enterprise should be easier than translation from scratch, because it can rely on bilinguals. I suspect that it has not been done because it cannot be done, at least not without accruing absurd ad hoc modifications. At this point an even more damaging criticism emerges. By comparison to a standard translation like (II), an alternative like (II∗ ) is much more complex and arcane, because it has to postulate a variety of syntactic roles and ambiguities. Why is this not a reason for rejecting it as false? In reply, Quine acknowledges that (I) provides a ‘better translation’ of ‘gavagai’ than (I
), and that a field linguist would have no patience with his contrived alternatives. But, he adds, indeterminacy is a ‘philosophical point’ (RBG 292; OR 34; see ITA 7–8; WO 74–5; PT 43, 46–7; Hookway 1988: 155– 9; Kirk 1986: 56–63). The superiority of (I) and (II) is of a merely pragmatic kind; it does not show that they capture a fact of the matter distorted by (I
) and (II∗ ). This reply is inadequate. The problem is not that Quine’s behaviourism precludes him from distinguishing an interpretation being simple from it being true, or that he has a pragmatic conception of truth (Wright 1997: 410–12; cf. Nimtz 2002: ch. 5.7). It is rather that he has an instrumentalist conception of knowledge. Quine conceives of science instrumentally, as a way of making sense of experience. His crucial counter to Carnap was precisely that pragmatic considerations like simplicity are no less central to this enterprise than they are to the adoption of linguistic frameworks. On what grounds, then, could he deride the difference between standard and contrived interpretations as merely pragmatic rather than factual, without drawing the same conclusion for the difference, for example, between Newtonian and Aristotelian mechanics? But what about non-pragmatist proponents of inscrutability? Here a different and even more important contrast between standard and contrived interpretations comes into play. The latter are at odds with the anthropological and psychological facts. Evans urges this point in the context of insisting that radical interpretation should not just provide a translation manual that correlates Jungle and English, but a psychological explanation of linguistic behaviour that invokes the kind of unconscious 3. Kirk (1986: chs. 5.2 & 8) shows that various alleged cases, including Quine’s example of the Japanese classifiers (OR 35–9) are either innocuous or spurious.

Indeterminacies

rules prominent in cognitive science. But the relevant facts need not be hidden in this way. For example, the native can display a desire to have what she calls ‘gavagai’ as a pet. This desire is perfectly intelligible on the standard translation (I), but becomes utterly mystifying on the alternative translation (I
). Similarly, if the natives explain ‘orith’ to infants or interpreters without making special provisions for its combination with ‘gavagai’, an interpretation like (II∗ ) can be excluded (see Unger 1984: 18; Hookway 1988: 157). More generally, human beings are by nature and necessity interested primarily in material objects and living creatures rather than their parts, to say nothing of temporal slices or universals. It is to objects and creatures that our perceptual capacities, our desires and our actions are cued. Unless there are specific reasons to suggest that the natives’ interests and activities are extremely exotic, interpreters must stick to the standard translations. Quine recognizes that our interpretations are guided by considerations of ‘psychological plausibility’ according to our ‘practical’ or ‘folk psychology’. Yet for him psychological plausibility is not a genuinely factual constraint, since these constraints are exhausted by stimulus and response (RBG 158, 292; PT 46–9; IA 7). Once more, however, this demotion of anthropological assumptions misfires. There is no reason to deny that these are genuine facts about human beings, their desires and activities, etc. Furthermore, as I have argued throughout, without recognition of such facts, interpretation from scratch could never get off the ground.

4

Scrutinizing reference II: proxy functions

Davidson should be sympathetic to this line of reasoning, since he regards certain beliefs and desires as prerequisite for rational agents. This may explain why he does not officially endorse the ‘gavagai’ argument. At the same time, he enthusiastically supports the argument from proxy functions. Such a function assigns to all singular terms a new reference and to all general terms a new extension, in such a fashion that the two changes offset one another. The apparent change is two-fold and sweeping. The original objects have been supplanted and the general terms reinterpreted. There has been a revision in ontology on the one hand and of ideology, so to say, on the other; they go together. (TT 19)

219

220

Quine and Davidson on Language, Thought and Reality

For proponents of inscrutability, proxy functions have notable advantages (Nimtz 2002: ch. 4.5). First, they avoid the onerous task of devising a step-by-step reinterpretation, since all reinterpretations are automatically coordinated. Secondly, unlike some ‘gavagai’ reinterpretations (e.g. treating it as a singular term like ‘rabbithood’ rather than a predicate), they do not alter the syntactic category of the terms. Thirdly, they do not affect the cardinality of general terms, unlike, for example, the switch from rabbits to undetached rabbit parts. As a result, proxy functions do not necessitate a reinterpretation of the apparatus of individuation. Finally, they leave intact the evidential and logical relations between sentences of Jungle. ‘The observation sentences remain associated with the same sensory stimulations as before, and the logical interconnections remain intact. Yet the objects of the theory have been supplanted as drastically as you please’ (PT 32; see TI 7). It is on account of proxy functions that Quine declares the inscrutability of reference to be ‘trivial and indisputable’, by contrast to the sentential indeterminacy engendered by the argument from above (PT 50; see TI 6). Davidson takes a similar line (ITI ch. 16; SIO ch. 5). Unlike Quine, he thinks that inscrutability poses no threat to the reality of propositional attitudes and their contents. But his reason is that there is no genuine difference between standard and proxy interpretations. Whether we interpret someone as saying that (1) Rome is a city in Italy

or that (2) The area 100 miles south of Rome is an area 100 miles south of a city in Italy

is no more semantically relevant than whether we state the air temperature in Fahrenheit or Celsius. That we can choose freely between these alternatives merely indicates that ‘certain apparent distinctions are not significant’ (ITI 154, see SIO 64–6). Davidson rejects what he calls a ‘building block’ approach in favour of a ‘holistic’ or contextualist position. The senses (truth-conditions) of sentences are not strictly speaking determined by the reference of their components, because the latter is merely an abstraction from the former. ‘Words have no function save as they play a role in sentences: their semantic features are abstracted from semantic features of sentences.’ Although a truth-theory employs notions like reference and satisfaction, these do not capture genuine semantic facts;

Indeterminacies

they are merely ‘theoretical constructs whose function is exhausted in stating the truth conditions for sentences’ (ITI 220–1, 223, see also 127–8; PCT 13–15). Davidson’s instrumentalist conception of reference is echoed by Quine’s recent ‘defusion of ontology’ or ‘global structuralism’. ‘Sentences, in their truth or falsity, are what run deep; ontology is by the way’ (FM 165). Therefore, ‘what matters for any object, concrete or abstract, is not what they are but what they contribute to our overall theory of the world as neutral nodes in its logical structure (SS 74–5; see PT 33–6; SN 9; TT 1, 20). The basic, indeed the only genuine semantic facts, we are told, concern the truth-conditions of sentences rather than the reference of their components. There is an obvious objection to this defusion of reference, and to its corollary that (1) and (2) are semantically equivalent. ‘Rome’ does not mean the same as ‘area 100 miles south of Rome’ and it refers to something different. Davidson retorts: ‘individual words don’t have meanings’ but only ‘a role in determining the truth conditions of sentences’, a role which is ‘captured’ by (1) ‘as surely’ as it is by (2) (SIO 79). But as we have seen (chapter 6, section 1), it is precisely individual words which have a meaning, because it is they which have a role in sentences. And the role of the words and phrases that make up (1), the contribution they make to the meaning of English sentences, is objectively and definitely different from the role of the words and phrases that make up (2). Why do Quine and Davidson turn a blind eye to this fact? The answer lies in their conviction that only the conditions under which speakers assent to or dissent from the utterances of sentences are semantically relevant. Thus Davidson insists that the only way of checking a semantic theory is to look at how the T-sentences it entails square with the assentconditions of native sentences. ‘The point behind this is that an utterance counts as a speech act only if it can be treated as sentential’ (RD 39; see ITI 133, 222; SCT 300; SIO 77–8, 108–10). If this were true, it would license the conclusion that reference is a theoretical semantic property, not one that can be directly observed. But it would not license the instrumentalist conclusion that reference is in any sense unreal, a mere fiction entertained for the benefit of testing truth-theories (Nimtz 2002: ch. 6.4). In any event, the premise is mistaken. Once more Quine and Davidson go astray, not because they insist that semantic phenomena must be displayable in behaviour, but because their conception of such behaviour is unduly narrow. There are perfectly accessible speech acts other than

221

222

Quine and Davidson on Language, Thought and Reality

uttering or assenting to whole sentences, such as ellipsis or invocation (Perry 1994; Rumfitt 1995). Ostension constitutes an even more important exception. In the course of interpretation from scratch, we typically try to explain individual words, by uttering them and pointing out what they stand for. This counts not just against the contextualist premise behind inscrutability, it also upsets the proxy function argument. Unlike ‘gavagai’ reinterpretations, some popular proxy functions do not yield referents that are located in the same vicinity as the originals. Objects and their cosmic complements are not in the same place. The same holds for Davidson’s example of mapping all objects on to their shadows (ITI 230). This creates space for ostensive disambiguation. If the natives point at rabbits rather than their shadows or cosmic counterparts, that’s what they are talking about. The reinterpretations that engender indeterminacy are ruled out by objective facts about linguistic behaviour. Quine tries to resist this conclusion by appeal to ‘deferred ostension’, as when we point at the gauge instead of the tank to show that there is gasoline (OR 39–40). By the same token, we can reinterpret the native by assuming that his ostension is a deferred one to shadows or cosmic complements rather than a direct one to rabbits. This response runs up once more against the anthropological dimension of interpretation from scratch. Deferred ostension is a complex phenomenon. It may not require the explicit introduction of a convention (pace Welch 1984). But it certainly requires a mutual understanding between those involved, and is not a ploy by which either we or the natives would initiate communication. Here the focus is on what we perceive in the ostended space, and that is objects rather than their proxies. This might still leave in place proxy functions that map objects on to other things in the same vicinity, such as their surfaces or their centres of gravity. But there is a second and more general threat to the proxy function argument. It concerns the idea that reinterpretations of terms preserve the assent-/truth-conditions of observation sentences (SS 75; FM 165 and ITI 229–30). Prima facie, this idea holds. ‘Lo, a rabbit’ is true in precisely the same situations as ‘Lo, an undetached rabbit-part’. Similarly for proxy functions. In all possible situations in which it is true that a is an F it is also true that the proxy f of a is the f of an F, and vice versa. Necessarily, Harvey is a furry thing iff the cosmic complement of Harvey is the complement of a furry thing.

Indeterminacies

On second thought, however, the assurance is precarious. What counts, according to Quine and Davidson, are the conditions that (typically) cause speakers to assent to/dissent from utterances (RI 21; SIO 201–2; FDT 275). But it is far from obvious that what causes speakers to assent to (3) Lo, a rabbit

is the same as what causes them to assent to (4) Lo, the centre of gravity of a rabbit.

Davidson would insist that the causes are identical, namely the appearance (i.e. the coming on to the scene) of a rabbit, which is ipso facto also the appearance of a centre of gravity of a rabbit. According to him, the only causes of our behavioural dispositions, as of anything else, are events. By contrast, many contemporary philosophers insist that the primary relata of causal relations are facts. If so, truth-conditions can be facts as well. Yet the fact that there is a rabbit differs from the fact that there is the centre of gravity, which puts paid to the proxy function argument (see Mellor 1995; Nimtz 2002: ch. 7.2). I agree with Davidson that facts cannot be causes, not because of accepting his slingshot argument, but because facts have no spatio-temporal location. In my view, however, this does not salvage the proxy function argument. Unlike Quine, Davidson appreciates the intentional character of assent. He has to accept, therefore, that assent- or truth-conditions are determined by the reasons for which speakers assent to utterances. But the reasons for which speakers assent to (1) and (3), respectively, differ from their reasons for assenting to (2) and (4). Indeed, only someone who possesses the concept of a centre of gravity can assent to ‘gavagai’ for the reason that the centre of gravity of a rabbit has appeared. In maintaining that the difference between (1) and (2) or between (3) and (4) is semantically insignificant, Quine and Davidson subscribe to a dogma of intensional semantics, and one that is incompatible with their own stress on conditions of assent. Being true in the same possible situations is not sufficient for two sentences being synonymous. In particular, (3) and (4) are not synonymous, since they are not feasible explanations of one another, and since someone who lacks the concept of a centre of gravity can believe what (3) says, but not what (4) says. An acceptable criterion of sentence synonymy must shed contextualist prejudices and mention sentence components.

223

224

Quine and Davidson on Language, Thought and Reality

To be synonymous, two sentences s1 and s2 must be what Frege (1979: 197) calls ‘equipollent’ or cognitively equivalent: nobody who understands them both can accept what the one says without also accepting what the other says. But they must also be capable of being parsed in such a way that for every component of ‘s1 ’ there is a synonymous component of ‘s2 ’. Our pairs satisfy the first condition but not the second. Neither ‘rabbit’ and ‘undetached rabbit part’ nor ‘Rome’ and ‘the area 100km south of Rome’ are synonymous, if only because they cannot be used to explain one another. A plausible definition for two terms t1 and t2 being synonymous is that they should be interchangeable in all declarative sentences (except for special cases, e.g. when t1 or t2 are mentioned rather than used) in such a way that the results are cognitively equivalent (see Rundle 1979: §54; Kunne 1983: ch. 6.4). This is obviously not the case for our pairs: ‘Un¨ detached rabbit parts are prolific breeders’ and ‘The area 100 km south of Rome is the seat of the Vatican’, for example, are unlikely to carry favour. By rejecting the contextualist dogma we can meet Quine’s longstanding demand for a criterion of synonymy. We can also resist the central ontological claim of the inscrutability thesis: there is a fact of the matter as to whether the natives speak about rabbits, their centres of gravity or their surfaces. Finally, we are in a position to deal with the epistemological problem. The native’s ‘gavagai’ can only be interpreted as ‘centre of gravity of a rabbit’ if there are reasons to ascribe to her the concept of a centre of gravity, which in turn presupposes fairly sophisticated scientific knowledge. 5

Indeterminacy and underdetermination

The argument from below rests on mistaken semantic assumptions. I shall now argue that the argument from above either rests on mistaken methodological assumptions or begs crucial metaphysical questions. Quine assumes that translation is either presuppositionless or indeterminate. That one can provide alternative translations of a given sentence by making holistic adjustments to the translation of other sentences, however extravagant and implausible they may be, is supposed to show that the initial translation was indeterminate. But, I have argued, this insistence on a presuppositionless procedure precludes even the ascription of stimulus meaning that is crucial to Quine’s behaviourist account. That meaning turns out to be indeterminate by these hyperbolical standards should not

Indeterminacies

count against intensional notions, therefore, but against the quest for the unconditioned. That quest is also at odds with Quine’s own anti-foundationalist holism concerning our knowledge of the physical world. He recognizes that the latter is not presuppositionless, because there are alternative ways of construing the overall data. Yet he does not draw the conclusion that our claims about the physical world are indeterminate, that is, that they do not capture an objective fact of the matter. Both Chomsky (1969) and Rorty (1972) have deplored this double standard. According to them, translation and physics are on a par. Either both of them deal with objective facts, though the evidence may leave underdetermined what these facts are, or both are indeterminate and hence not factual. In response, Quine has insisted that the indeterminacy of translation is not just a special case of the underdetermination of theory by evidence (namely the underdetermination of linguistics); rather, it is ‘parallel but additional’ (RDH 303). This is crucial to the argument from above, because it provides the only rationale for supposing that intensional discourse, unlike physical theory, does not capture a fact of the matter. That there is a fundamental difference between underdetermination and indeterminacy is therefore an article of faith among Quineans. But to the question ‘Why?’ there are more answers than Quineans. On one rendition of the argument, indeterminacy follows directly from combining the thesis of underdetermination with a verificationist theory of meaning (Føllesdal 1973). If there is no evidence which counts in favour of an individual sentence, and if the meaning of a sentence would have to be identical with the evidence in its favour, then individual sentences do not have a specific meaning. However, this line of reasoning weakens rather than strengthens Quine’s case. For one thing, it does nothing to substantiate the claim that in translation there is no fact of the matter. Its conclusion is not that meaning is in any way unreal or non-factual, but that its proper locus is at the level of theories rather than of individual sentences.4 For another, on this construal indeterminacy is additional to underdetermination only in the unfortunate sense that the argument in its favour relies on additional assumptions. It takes for granted not just the 4. Indeed, once this is granted, meaning can even be defined for individual sentences: s1 and s2 are synonymous iff they are intersubstitutable salva confirmatione within a cluster of sentences, that is, iff substituting s1 for s2 does not alter the evidence that counts for or against the whole cluster (Grice and Strawson 1956: 156; cf. WO 64–5).

225

226

Quine and Davidson on Language, Thought and Reality

contentious thesis of underdetermination, but also a verificationist theory of meaning. Quine always set store by the latter (RHS 155–6), but he was pretty much on his own. Therefore indeterminacy cannot be demonstrated simply by combining Quine’s holistic epistemology with his verificationist semantics. A second rendering of the argument from above runs as follows: Insofar as the truth of a physical theory is underdetermined by observables, the translation of the foreigner’s physical theory is underdetermined by translation of his observation sentences. If our physical theory can vary though all possible observations be fixed, then our translation of his physical theory can vary though our translations of all possible observation reports on his part be fixed. Our translation of his observation sentences no more fixes our translation of his physical theory than our own possible observations fix our own physical theory. The indeterminacy of translation is not just an instance of the empirically underdetermined character of physics. The point is not just that linguistics, being a part of behavioural science and hence of physics, shares the empirically underdetermined character of physics. On the contrary, the indeterminacy of translation is additional. Where physical theories A and B are both compatible with all possible data, we might adopt A for ourselves and still remain free to translate the foreigner as believing A or believing B. (RIT 179–80; see WO 78; RDH 303)

This argument is equally straightforward. Because of underdetermination, we have a choice in interpreting Jungle utterances even if, heeding the principle of charity, we assume that the natives must have come up with a theory of nature that optimally fits all the possible evidence. But it is uncompelling. What it shows is this: if our only constraint on translation is that the physical theory we ascribe to the natives should optimally fit the empirical evidence, and if there is more than one physical theory that optimally fits the evidence, then translation will be indeterminate. Both assumptions are problematic. It is far from obvious that the totality of conceivable evidence is incapable of favouring one physical theory as the most rational (Wright 1997: 413–17). More importantly, Quine has given us no reason for excluding other constraints on translation, notably that the beliefs we ascribe should fit the overall practice and culture of the natives. Once more, no objection to the factuality of intensional discourse materializes.

Indeterminacies

According to the most promising version of the argument from above, indeterminacy is on a par with underdetermination epistemologically or methodologically speaking, but not ontologically (see RHS 74, 155; RDH 303; PT 102–5; Gibson 1986). As far as the evidence is concerned, claims about meaning are no more precarious than claims about physical reality, yet they do not concern a fact of the matter. The only facts of the matter are physical. But physical facts do not determine what our expressions mean. Even if we had settled on a complete physical theory of nature, it would not entail an interpretation of Jungle. For different translation manuals can be ‘compatible with all the same distributions of states and relations over elementary particles’ (TT 23). ‘I speak as a physicalist in saying there is no fact of the matter. I mean that both manuals are compatible with the fulfilment of just the same elementary particle states by space-time regions’ (FM 167; see RHS 187; ITA 10). Ascriptions of meaning do not capture a fact of the matter, because different such ascriptions are equally compatible with the same physical state of the world. Like its predecessors, this argument succeeds in making indeterminacy additional to underdetermination. However, it also amounts to an astonishing, if hitherto unnoticed, volte face. Originally, that there are no facts of the matter about meaning was supposed to be the conclusion of Quine’s argument: intensional discourse is not factual because translation is indeterminate. Now the claim that there are no such facts appears as a premise: translation is indeterminate rather than merely underdetermined because intensional discourse is not factual. As a result, Quine’s case against intensions once more turns out to be circular. As we saw in chapter 3, section 1, he needs to show that translation is indeterminate, and that the notion of synonymy is therefore vacuous, without presupposing the illegitimacy – obscurity or non-factuality – of intensional notions. All is not yet lost, however. Quine simply needs to provide an independent justification of the claim that there are no intensional facts of the matter. In the aforementioned passages, he seems to derive that conclusion from two premises. The first premise is physicalism, more precisely, the idea that the only facts of the matter are physical: ‘where positions and states of bodies do not matter, there is no fact of the matter’ (FM 162). The second premise is the Brentano thesis: ascriptions of meaning cannot be reduced to physical statements. This line of reasoning is at odds with Quine’s repeated insistence that his flight from intensions is not based on an anti-mentalist, physicalist

227

228

Quine and Davidson on Language, Thought and Reality

ontology (FM 166–7; RHS 74). But it also suffers from more serious shortcomings. I accept the second premise. Not everybody does, though. According to Quine, to reduce a theory T to another T
requires translating the statements of T by statements of T
and to analyse or define the terms of T in terms of T
(FLPV 20, 38; OR 76–7; MVD 92–5; SS 86–7). It is now generally recognized that there is no prospect for such an analytic reduction of mental and semantic discourse to physics. However, there is also a less demanding kind of reduction. Roughly speaking, it demands only that the statements of T be derivable from those of T
by means of so-called ‘bridge laws’, laws that express identities or at least nomic correlations between properties. The paradigm of this scientific reduction is the one of classical thermodynamics to statistical molecular mechanics, and in particular the reduction of temperature to mean kinetic energy. Some opponents of indeterminacy have pointed out that the failure of analytic reductionism does not entail the failure of scientific reductionism (Kirk 1986: 18–22, 44–6; Nimtz 2002: ch. 5.5). But they have not provided any clues as to how statements about meaning could be derived from the statements of physics. Even proponents of the ‘naturalization’ of semantics admit that so far all attempts have ended in failure (e.g. Loewer 1997). Finally, the parallel discussion of mental properties has turned up strong arguments against the enterprise. Mental properties are multiply realizable in psycho-chemical terms, not just in principle (in mentally gifted robots and Martians), but in fact, and not just across species. For instance, when different test persons solve one and the same simple problem, slightly different parts of the brain are activated (Horgan 1992; Beckermann 2001: 137–41). Semantic properties, even of a simple kind, pose even greater hurdles. There is every reason to think that different parts of the brain are active when speakers of different natural languages utter sentences with the same meaning. When I say ‘More light!’ I am not in the same neurophysiological state as Goethe when he said ‘Mehr Licht!’ on his death-bed, but the two utterances have the same literal meaning. The same holds of synonymous sentences uttered by speakers of a single language, e.g. ‘It is a drake’ and ‘It is a male duck’. The literal meaning of an expression depends on its role within a natural language. And it is difficult to see how the relevant features of such a complex social and historically evolving practice could be correlated neatly with physical properties of the brains of individual speakers.

Indeterminacies

A scientific reduction of mind and meaning will also be ruled out if Davidson’s argument against type/type identity theories is sound. Quine has always rejected scientific reductionism. More recently, he openly embraced anomalous monism, thereby combining a substance monism with a property dualism. A predicate like ‘thinking about Fermat’s last theorem’ applies to physical objects, namely bodies, rather than minds, but it cannot be correlated with a physical predicate (PT 71–2; SS 87–8; see also WP 243; RR 8–15; RHS 533). At the same time, even as regards properties Quine clings on to a non-reductive physicalism. It is not a utopian dream of our being able to specify all mental events in physiological or microbiological terms. It is not a claim that such correlations even exist, in general, to be discovered; the groupings of events in mentalistic terms need not stand in any systematic relation to biological groupings. What it does say about the life of the mind is that there is no mental difference without a physical difference. (FM 162; see RR 15; RHS 74–75)

While rejecting psychophysical parallelism, Quine and Davidson hold on to supervenience. Identity of physical state entails identity of mental state, but not vice versa. ‘If a man were twice in the same physical state, the physicalist holds, he would believe the same things both times, he would have the same thoughts, and he would have all the same unactualized dispositions to thought and action’ (FM 162; see RHS 74–5). By counterposition, there is no difference in matters of fact and no change without some redistribution of microphysical states, which is why physics provides ‘full coverage’ of all events (TT 98; FM 166). This non-reductive physicalism is supposed to support the Brentano thesis, the second premise of Quine’s argument against semantic facts of the matter. But it militates against the conclusion of that argument. If semantic properties supervene on physical properties, then they are multiply realizable and there can be a physical difference without a semantic difference. But there cannot be a semantic difference without a physical difference. By this token, diverse ascriptions of meaning are precisely not equally compatible with the same physical facts, even though we may be forever incapable of deriving the semantic from the physical. Consequently, there is no reason to deny that they concern a fact of the matter, contrary to Quine’s argument. Supervenience does not render the mental and semantic non-factual. It just means that for any mental or semantic difference or change there must also be a physical one. Natural sciences like biology

229

230

Quine and Davidson on Language, Thought and Reality

also supervene on physics: while there is no biological difference without a physical one, biological phenomena like metabolism or evolution are realizable in diverse physical ways. Yet Quine does not and cannot deny that biology is factual. In principle, Quine could abandon supervenience, thereby embracing a stronger version of the Brentano thesis.5 In that case he could insist that semantic attributions are non-factual precisely because they do not even supervene on the physical, and are entirely free-floating. But such a move would weaken his first premise, namely that the only facts of the matter are physical. If the mental does not even supervene on the physical, why should this not show that there are sui generis facts of a non-physical kind? Quine insists that the very notion of a fact of the matter is internal to our theory of nature, and hence to physics (TT 23; RHS 155; Gibson 1986: 147–52). Unfortunately, this begs the question against those who maintain that there are facts of the matter other than physical ones. Quine assumes that natural science has a special ontological authority (N 257; OR 84; TT 21; WO 22). But this is a metaphysical claim about the comprehensiveness of physical science, and not itself a scientific finding. Many philosophers nowadays assume, with Quine, that the physical sciences are causally complete, that is to say, that all events have a complete physical cause. But even if this is true, it does not entail that all events are physical, because some events with a physical cause might be irreducibly mental epiphenomena (Crane 1992). Furthermore, even if all events are physical, it does not follow that all facts are, as Quine maintains. He himself accepts that there are objects of a non-physical kind, namely classes. And there is a clear case for insisting on something analogous for facts. (P1 ) It is true that p iff it is a fact that p. (P2 ) It is true that there is no greatest natural number. 5. At a local level, supervenience arguably fails. Both Wittgenstein and externalists adduce plausible cases in which two physically identical individuals have and express different thoughts, due to a difference in their causal or social contexts. For example, Nofretete could not intend to cash a cheque, even if she were in the same neurophysiological state and comported herself in the very same manner as I do when I go to the bank, because the relevant institutions did not exist (Glock 2000: 16–19). Similarly, a monolingual American and a monolingual German could be in the same neurophysiological state when they utter ‘Empedokles liebt’ (see ITI 98), yet their utterances will mean something different. At a global level, supervenience is more compelling (see SIO 62). How could two individuals mean something different by one and the same utterance if everything were physically equal, including their history and social context? But such global supervenience lacks bite, because everything is never equal. We could never infer what an utterance means by observing that the overall physical situation is precisely the same as for another utterance with an established meaning.

Indeterminacies

(C1 ) It is a fact that there is no greatest natural number. (P3 ) That there is no greatest natural number is not a physical fact. (C2 ) There are facts other than physical ones.

(P1 ) is a truism. One might try to avoid it by impugning the very notion of a fact, in the style of Davidson. But then one could not even formulate Quine’s physicalist thesis that all facts of the matter are physical. (P2 ) will be granted by anyone who accepts that mathematical statements are capable of being true, which includes most naturalists. (P3 ) has been established by Frege in his devastating critique of the ‘gingerbread or pebble arithmetic’ (1953: VII) beloved by empiricists and naturalists like Mill. That there is no greatest natural number does not mean, for example, that one can forever continue to physically add objects to a collection of objects, or that there is an infinite number of objects in the universe. Pace Quine, there are facts of the matter ‘where the positions and states of bodies do not matter’. And why should semantic facts not be among these non-physical facts? Semantic facts have a connection with the activities of human beings that is absent in the case of most physical facts, and may be absent in the case of mathematical facts. But that is no bar to them being factual. This result blocks a direct argument to the effect that intensional statements are not factual. It also undermines the most significant attempt to turn the argument from above into something more vicious than the underdetermination of theory by evidence. Without unwarranted verificationist or physicalist assumptions, the indeterminacy of translation reduces to the underdetermination of theory by evidence. Quine’s concrete illustrations of indeterminacy confirm this impression. They all trade on the idea that a given sentence can be translated in a non-standard way by making adjustment in the translation of other sentences. Since, as Quine rightly assumes, such underdetermination is no obstacle to factuality, the argument from above does not establish that there is no fact of the matter as to what our expressions mean. This removes another ontological threat to intensional notions and statements. But what about the epistemological threat to translation from scratch posed by the sceptic or relativist? If our translation manual for Jungle is underdetermined by the evidence, it seems that we can never be certain that it captures the semantic properties of Jungle, and that radically different translation manuals are equally justified.

231

232

Quine and Davidson on Language, Thought and Reality

Fortunately, as in the case of our theories concerning physical nature, this threat is always defused in practice. Underdetermination is a logical possibility rather than a genuine epistemological problem. Even Quine admits that our standard translation manuals are the most rational ones. He also acknowledges that the ‘fluency of conversation and the effectiveness of negotiation’ with the natives is ‘utterly factual’ (PT 43). But this holds not just of the behaviourist elicitation of responses, but also of our rich and diverse hermeneutic interactions with the natives. The difference between rabbits and their undetached parts, for example, could not remain hidden in our mutual conversations and cooperations. None of this is proof against a hyperbolical doubt that we might have gone systematically wrong right at the start. But as we have learned from Quine, among others, such hyperbolical doubt may be nonsensical, and it is certainly irrational. If we abandon the tempting but misbegotten idea that translation is either unconditional or indeterminate, the threat of semantic scepticism evaporates. As we have seen, the actual history of translation from scratch suggests that there is no fundamental indeterminacy: definite mistakes have been made, and these have been definitely rectified.

8

Meaning and understanding

Davidson’s discussion of radical interpretation combines his own truth-conditional semantics with Quine’s naturalist account of language. The idea already appears in ‘Truth and Meaning’, where it serves to illustrate how a theory of meaning can pass the requirement of empirical verifiability, that is, how it can be established whether a Tarskian truththeory fits a given natural language. If we construct a truth-theory for a language other than our own, we cannot assume any insights into equivalences between object- and metalanguage and therefore operate under conditions of radical translation (ITI 27). In subsequent writings, radical interpretation comes to serve the additional function of defusing objections to truth-conditional semantics. There is also a gradual change from treating radical interpretation merely as a way of verifying claims about meaning to treating it as constitutive of meaning: only those features of a language accessible to a radical interpreter can be relevant to the meaning of its sentences. In the final instance, the thought experiment takes on a life of its own. It is supposed to yield interesting corollaries in epistemology, as well as a picture communication and understanding which is in some respects incompatible with the initial project. The latter is the subject of this chapter. I start by arguing that radical interpretation cannot solve the extensionality problem and transform a theory of truth-conditions into a theory of sentence meaning (section 1). Section 2 addresses the question of whether Davidson is right to insist that a theory of meaning should be what Dummett calls ‘modest’, and section 3 the issue of whether a truth-theory is tacitly known by competent speakers of natural languages. In section 4, I critically discuss Davidson’s attack on the idea of linguistic conventions, and in the final section his account of first-person authority.

[233]

234

Quine and Davidson on Language, Thought and Reality

1

The extensionality problem

Davidson’s abiding ambition is to show that an empirically verifiable theory of meaning for a natural language L can be provided by constructing a truth-theory for L. As we have seen, that project faces numerous difficulties. The one which has attracted most attention is the problem of extensionality. The theorems of such a theory state conditions under which a sentence s of L is true, yet this by itself does not guarantee that these theorems capture the meaning of s. The problem arises at two levels: the avoidance of circularity in the conceptual foundations of the theory and the exclusion of rogue T-sentences. Since Tarski was concerned with truth, he could take for granted that the metalanguage sentences used on the right-hand side of T-sentences translate the object language sentences quoted on the left-hand side. Davidson cannot make that assumption. A correct translation is one that preserves meaning, hence a theory relying on translation would take the notion of meaning for granted and thereby violate the Non-Circularity condition. On the other hand, if Davidson drops the condition that ‘p’ be replaced by a translation of the sentence named by ‘s’, the ‘means that’ of (P) s means that p

can no longer simply be replaced by the ‘is true iff’ of Davidson’s favoured (T) s is true iff p.

Consequently, there is no reason to suppose that a theory of truth can serve as a theory of meaning (Platts 1979: 56–7). The extensional ‘is true iff ’ also opens the door to rogue T-sentences. These are T-sentences which are true, yet do not capture the meaning of the quoted object language sentence. The true T-sentences of German, for example, include not just the desired (1) ‘Schnee ist weiss’ is true iff snow is white

but also a deviant case like (2) ‘Schnee ist weiss’ is true iff grass is green.

On occasion, Davidson seems to bite the bullet of reducing synonymy to material equivalence (ITI 26; PCT). For the most part, however, he admits that truth-conditions are not to be equated with meanings. Nonetheless, he insists, it is possible ‘to extract a rich concept’ like translation

Meaning and understanding

(or meaning) from ‘thin bits of evidence’ concerning truth (ITI 74). A truththeory can specify what a hearer understands, provided that it is subjected to additional constraints (ITI 27 n10, 56 n3, 173, 178–9; SCT 312). Any theory meeting these constraints is supposed to be ‘interpretive’, that is, yield meaning-giving theorems. Davidson has suggested three types of constraint (see Fodor and LePore 1992: ch. 3), which I shall discuss in turn. (A) The Compositional and Holistic Nature of Language (B) The Nomological or Causal Nature of T-Sentences (C) The Constitutive Constraints of Radical Interpretation

(A) An extensionally adequate theory of German must generate a true T-sentence for all of its potentially infinite sentences. It can do so only by displaying the meaning of the sentence as determined by the semantic properties of its components and its syntactic structure. This ‘holistic constraint’ proscribes rogue T-sentences, because theories of meaning must pay tribute to the structural relations between sentences (ITI 26 n10, 61, 138–9, 225). To entail (2), for instance, the theory would have to include two axioms like the following: (I) ‘Schnee’ denotes grass (II) an object x satisfies ‘ist weiss’ iff x is green.

But such axioms allow the derivation of T-sentences that are downright false. Together with suitable rules for demonstratives (ITI 34), they entail relativized T-sentences like (3) ‘Das ist weiss’ is true in German as spoken by x at t iff the object or stuff demonstrated by x at t is green.

To be meaning-giving, a T-sentence must not just be true, it must also be entailed by an overall theory which entails true T-sentences for all sentences of L. On the conceptual side, this solution to the extensionality problem has provoked the objection that ‘entails’ is an intensional notion, and that hence ‘Davidson’s grand design is in ruins’ (Foster 1976: 23). In response, Davidson avowed that it was never his ambition to avoid intensional notions, but only semantic notions like that of meaning (ITI 175–6). However, entailment is not just intensional but also semantic. Furthermore, Davidson’s paratactic account of entailment (ITI 176–8) ends up explaining it in terms of synonymy. He promises that synonymy will in turn be spelled out by the theory of meaning. But this defence is circular. The theory will spell out meaning and synonymy only if the problem of

235

236

Quine and Davidson on Language, Thought and Reality

extensionality is overcome; therefore, one cannot invoke synonymy to overcome that problem. Compositionalism is also impotent to solve the problem of rogue T-sentences. While it can weed out cases like (2), it is incapable of discriminating between terms which are co-extensive without being synonymous, like ‘cordate’ and ‘renate’. As a result, it allows rogue T-sentences like (4) ‘Tiger haben ein Herz’ is true iff tigers have a kidney.

Compositionalism also permits rogue T-sentences which are generated out of interpretative ones by adding a logical truth, as in (5) ‘Schnee ist weiss’ is true iff snow is white and grass is green or grass is not green.

The derivation of (5) relies not on the axioms for the object language but on the logical apparatus which the metalanguage requires in order to generate any T-sentences at all. If a theory yields (1), it automatically yields any of its logical consequences such as (5). Davidson has tried to plug this hole by insisting that an acceptable theory must derive T-sentences by way of a ‘canonical proof’ (ITI 138). Such a proof allows one to move from one biconditional to the next only by appeal to the basic axioms of the theory. (5) can be ruled out because it exploits additional logical apparatus. However, one can construct axioms from which (5) can be derived by canonical proof (Foster 1976: 7–16; Fodor and LePore 1992: 63–70), such as (II
) an object x satisfies ‘ist weiss’ iff x is white and the earth moves

(II
) does not generate false T-sentences. All the T-sentences concerning ‘white’ remain true when we add a true conjunct on the right-hand side, as in (6) ‘Schnee ist weiss’ is true iff snow is white and the earth moves.

One might disallow such counterfeit theories by insisting that the axioms should be no less canonical than the proof procedures (Williams 1998; Lycan 2000: 143). But while formal logic may furnish independent standards for what counts as a canonical proof, there are no such independent standards for canonical axioms of a theory of meaning. Davidson cannot furnish such standards by saying that canonical axioms state the meaning of sentence components correctly, or that they incorporate only what competent speakers are supposed to know about these components. For both

Meaning and understanding

replies presuppose an understanding of what the words of L mean, thereby violating the requirement of Non-Circularity. It emerges that compositionalism cannot solve the extensionality problem. At this point radical interpretation comes into play. A Davidsonian theory of meaning must be verifiable by behavioural evidence, without any prior knowledge of what speakers mean or think (ITI 27, 125–7). When applied to natural languages, such a theory delivers theorems of the form (T∗ ) s is true in L as spoken by x at t iff p.

These relativized T-sentences do not provide the evidence for radical interpretation, because at the outset of such interpretation we cannot divine under what conditions sentences of L are true. Instead, we must start with observations about what sentences speakers of the other language hold true under what circumstances. In the first instance, such observations concern individual speakers, as in (7) ‘Es regnet’ is held true by Kurt at t1 and it is raining near Kurt at t1 .

Such observations are then generalized to cover whole speech communities, as in (8) (x) (t) (if x belongs to the German speech community then (x holds true ‘Es regnet’ at t iff it is raining near x at t)).

If, relying on the principle of charity, we assume that German speakers hold true sentences that are by and large true, we can pass on to a T∗ sentence like (9) (x)(t) (‘Es regnet’ is true in German as uttered by x at t iff it rains in the vicinity of x at t).

Can this procedure rule out rogue T-sentences? Suppose that in the vicinity of Germanophones rain and high winds always coincide. In that case, there is no means for deciding between (9) and (10) (x)(t) (‘Es regnet’ is true in German as uttered by x at t iff it is windy in the vicinity of x at t).

Nonetheless, radical interpretation promises two new solutions of the extensionality problem. The first is (B). It trades on the causal or nomological status of T-sentences. Davidson maintains that the T-sentences of an adequate theory are empirical generalizations about the speakers of L and hence must not only be true but also lawlike (ITI 26 n11; SCT 313 n57). By

237

238

Quine and Davidson on Language, Thought and Reality

contrast, rogue T-sentences do not support appropriate counterfactuals. The truth-value of ‘Schnee ist weiss’ depends on the colour of snow rather than the colour of grass; hence it would retain that truth-value even if the colour of grass were to change. An initial objection to this solution is that T-sentences cannot be natural laws because of the conventional nature of language: it is not a law that ‘Es regnet’ has certain truth-conditions because it is not a law that this form of words means what it does in German (Fodor and LePore 1992: 84–6). But this does not hold of sentences like (8). Although these do not express laws of nature, they are causal statements which support counterfactuals, namely that under certain conditions speakers would assent to certain forms of words. Kurt’s assent to ‘Es regnet’ is caused by noticing rain in his vicinity, not by noticing wind. Consequently the truthconditions of that sentence are stated by (9) rather than (10). The radical interpreter establishes the truth-conditions of ‘Es regnet’ by noting the causal connection between the speaker’s assent to it and the perspicuous presence of rain (ITI 26 n10; SIO 29, 146–7; PCT 15). Unfortunately, this proposal does not distinguish between interpretative truth-theories and those which are merely extensionally adequate (Putnam 1975b: 261). An appeal to laws cannot slice truth-conditions more finely than nomological equivalence. But there are terms which are nomologically equivalent without being synonymous, such as ‘water’ and ‘H2 O’. Consequently, we are still lumbered with rogue T-sentences like (11) ‘Wasser ist nass’ is true iff H2 O is wet.

This leaves (C). A theory of meaning must allow us to interpret correctly all the sentences of a natural language under conditions of radical interpretation. Such interpretation, Davidson avers, is possible only if we lay down certain constraints on what counts as a correct interpretation, notably the principle of charity. These constraints solve the extensionality problem because they can allegedly be spelled out without presupposing the notion of translation; instead, they yield that notion, since they are constitutive of it (ITI 224; Miller 1998: 266). It appears, however, that Quine is right in treating charity as merely a heuristic device that will help us to come up with interpretations that we can recognize as correct by independent standards. Charity is not constitutive of the concepts of translation or interpretation, since one cannot define a correct translation/interpretation as one that ascribes predominantly

Meaning and understanding

rational beliefs or desires. A correct interpretation is simply one that preserves meaning. Wiggins (1997: 18–21) tries to avoid the looming circularity by spelling out the idea of a correct interpretation without reference to meaning. Successful interpretation need not be explained as one which respects relations of synonymy. We can explain it instead as interpretation that facilitates interaction to an optimal degree. Accordingly, a theory of meaning will be interpretative if the T-sentences it implies are not just true, but advance unimprovably the effort of making total sense of the speakers of L. Now, ‘making total sense’ is itself a semantic notion of the proscribed kind. But Wiggins retorts that making sense is wider than these undesired notions, and that it can be demonstrated non-linguistically. It is a matter of interpreter and native succeeding at joint enterprises in a shareable form of life. This strategy subjects semantic notions like making sense and understanding to an instrumentalist reduction. There is no a priori reason why false interpretations should not facilitate joint enterprise, notably if interpreter and subject pursue conflicting interests. By the same token, the problem of rogue T-sentences also remains unsolved. There may be T-sentences which unimprovably advance joint enterprise, without preserving meaning. For example, in coping with pious Germanophones we might fare best with T-sentences like ¨ (12) ‘Es gibt ein Jungstes Gericht’ is true iff the way humans conduct their lives has moral significance over and above their personal happiness.

But only a crude Wittgensteinian fideist could accept that (13) is meaninggiving. Davidson has spawned several ingenious attempts to bridge the gap between truth-conditions and meaning. Ultimately, however, the gap remains. Whatever hopes we invest in Davidsonian theories of meaning, they should not include the extraction of sentence meaning from truthconditions. By the same token, the theory of sense can no more be reduced to a theory of reference than the theory of force. 2

The privations of modesty

In homophonic theories of interpretation, the object language is part of the metalanguage; in heterophonic theories it is not. Leaving aside sentences involving indexicals, a homophonic theory uses disquotational axioms like

239

240

Quine and Davidson on Language, Thought and Reality

(III) ‘snow’ denotes snow

and (IV) an object x satisfies ‘is white’ iff x is white

to yield disquotational theorems like (13) ‘Snow is white’ is true iff snow is white.

Such T-sentences are notorious for being trivial. In chapter 4, section 2 I argued that this triviality is actually desirable in a definition of truth. But can it be tolerated in an empirical theory of meaning for a natural language L? This question is at the centre of a quarrel between Davidsonians on the one hand, Dummett on the other. According to Dummett, a constructive theory of meaning should be ‘full-blooded’ rather than ‘modest’ (1993b: ch. 1). That is to say, it should explain the concepts of L, and thereby the meaning of its terms, to someone who does not yet possess them. By contrast, advocates of modesty like McDowell insist that a homophonic theory is all we can hope for. The meanings or truth-conditions of our sentences are directly perceptible: ‘the thought (say) that some table-tops are square can be heard or seen in the words “Some table-tops are square”’, if only to those who already understand English (1998: 99–100). Against the complaint that such a theory is uninformative, they point out that both its axioms and its theorems state contingent facts about English (QVT 439; Evans and McDowell 1976: xi). Nevertheless, disquotational axioms and theorems do not satisfy a key requirement on a Davidsonian theory of meaning, namely that they ‘give the meaning’ or explain the expressions quoted on the left-hand side. What they state is not a tautology but contingent; yet the way it is stated is uninformative, and not just for contingent reasons. To be sure, on the lefthand side of (13) the sentence ‘snow is white’ is mentioned, whereas on the right-hand side it is used. But one cannot understand the sentence used in the explanans unless one also understands the sentence mentioned in the explanandum. Homophonic theories are reminiscent of those Anglophones abroad who try to make themselves understood by repeating English phrases in an increasingly loud voice (‘I want a shandy! Do you understand me? A shandy!!’). According to Davidson, knowledge of a theory of meaning should enable one to communicate with speakers of an entirely alien language. But homophonic theories explain the expressions of L only to those who already understand them. Consequently, they are

Meaning and understanding

ill-suited to the task of radical interpretation. When we encounter speakers of an alien tongue, we are ex hypothesis unable to hear or see their thoughts in their words; no immediate linguistic understanding is yet possible. In defence, proponents of modesty have made two points: a. To be able to communicate, what one needs to know is not an individual disquotational T-sentence, or even all of the infinitely many disquotational T-sentences, but how to derive these T-sentences from the axioms of the theory. b. Homophonic translation applies not to sentences in vacuo but to utterances, and it has empirical content because it relates these utterances to the conditions under which they are made (Alvarez 1994: 355–6; ITI 44, 130–3).

These replies do not rehabilitate homophonic theories. As regards (a), mastery of the axiomatic system of a homophonic theory is no more necessary or sufficient for understanding than being able to provide single disquotations. It is not necessary, since it is a skill lacked by competent speakers of English ignorant of the Tarskian apparatus. It is not sufficient for the following reason. I can report the lecture of a deconstructivist, provided that I have mastered indirect speech. I can also provide a truth-theory for such a lecture. All I need is a grasp of the standard axioms for terms like ‘and’, ‘not’ or ‘all’, and the ability to construct disquotational axioms like ‘“differance” denotes differance’ and ‘an object x satisfies “is logocentric” iff x is logocentric’. With their help, I can derive theorems that would grace any homophonic theory, for example (14) ‘Differance is logocentric’ is true iff differance is logocentric.

Nor does my theory miss the empirical content invoked in (b). Theoretical statements like ‘differance is logocentric’ neither gain nor lose intelligibility by being related to the speaker’s environment. In short, my ability to construct a homophonic theory does not guarantee that I understand any part of the deconstructivist lecture. Even the ability to provide homophonic translations like (3
) ‘This is white’ is true in English as spoken by x at t iff the object or stuff demonstrated by x at t is white

guarantees only that I’ve understood the indexical, not that I understand the quoted sentence as a whole.

241

242

Quine and Davidson on Language, Thought and Reality

It may appear that matters are strikingly different for heterophonic theories. A truth-theory for German couched in English will include the following two axioms: (V) ‘Schnee’ denotes snow. (VI) an object x satisfies ‘ist weiss’ iff x is white.

And these will allow one to derive the meaning-giving theorem (1) ‘Schnee ist weiss’ is true iff snow is white.

Unlike (13), (1) provides novel information. Of course, it is informative only to someone who is ignorant of the meaning of ‘Schnee ist weiss’ yet understands ‘snow is white’. But this is precisely the kind of ignorance a Davidsonian theory addresses: its theorems specify what the speakers of an unknown language say by uttering certain sounds. What is debatable is whether a theory of meaning should address itself exclusively to this kind of ignorance. The worry is that a heterophonically modest theory fails to explain the concepts expressed by the primitive terms of the object language, because it presupposes the concepts expressed by the primitive terms of the metalanguage. By the same token, the difference between such a theory and a Quinean translation manual that simply correlates sentences from different languages is less significant than Davidson makes out (Dummett 1993b: 6). To this last point, Davidson has a rejoinder (ITI 174–5; Evans and McDowell 1976: vii–xi; McDowell 1998: 98). Someone who knows the truth expressed by a theorem from a Quinean translation manual like (15) ‘Schnee ist weiss’ means the same as ‘La neige est blanche’

understands ‘Schnee ist weiss’ only on condition that she understands ‘La neige est blanche’. But one can know the truth expressed by (1) without understanding the language used on the right-hand side: we can ascribe such knowledge to a speaker of German who is ignorant of English. For although (1) uses a language to specify what ‘Schnee ist weiss’ means, it does so not by correlating it with another quoted sentence, but by stating the conditions under which it is true. This difference between a Davidsonian theory of meaning and a Quinean translation manual does not, however, allay Dummett’s scruples. Although understanding the truth expressed by (1) does not require an understanding of its specific metalanguage (English), it requires an understanding of some metalanguage or other, namely of the language in which

Meaning and understanding

the interpreter would state the content of (1). The theory does not explain the concepts expressed by the terms of that metalanguage. Davidson would argue that this is an advantage. He draws a sharp distinction between questions of logical form and the analysis of individual concepts. By his own admission, in a truth-theory for sentences involving ‘good’, for example, ‘what is special to evaluative words is simply not touched: the mystery is transferred from the term “good” in the object language to its translation in the metalanguage’ (ITI 31; see Evnine 1991: 93, 131). Even a heterophonic truth-theory does not provide a non-circular explanation of the concepts of L; what it furnishes instead is an account of logical form. The way in which T-sentences are derived from the axioms of the theory conveys information about how the quoted sentence is composed in truthrelevant ways (ITI 61). This position is problematic. To begin with, it is difficult to keep the elucidation of logical form and the analysis of concepts apart. Both are concerned with inferential powers. For instance, ‘p’ is entailed by ‘A knows that p’ no less than by ‘p and q’. What underlies the entailment in the second case is the concept of conjunction, what underlies it in the first is the concept of knowledge. Davidson has to demarcate those inferential powers which he regards as a matter of logical form, namely those associated with the logical constants of the predicate calculus, from those associated with so-called ‘material concepts’. A similar problem confronts Quine’s explanation of logical truths as those in which only the logical particles occur essentially, namely of specifying what particles should count as logical (Pap 1958: 130–3). Even if a non-circular demarcation of logical from conceptual truths is possible, a modest theory will account not for meaning but merely for logical form (Dummett 1987: 265). Its theorems will not be meaning-giving, they will only specify one kind of truth-relevant structure. Fortunately, Davidson does not follow his own advice. His work is replete with analyses of sentences involving concepts like belief, desire or causation. In that case, however, it is unclear why we should not be entitled to an explanation of concepts like that of whiteness, to say nothing of central philosophical concepts like knowledge and goodness. It emerges that explanatory modesty is undesirable. But is it avoidable? Initially, Dummett suggested that a full-blooded theory of meaning would involve ‘no presupposition’ at all: it conveys every concept of L to someone who possesses no concepts whatsoever, that is, someone who does not understand any expressions from any language (1993b: 6).

243

244

Quine and Davidson on Language, Thought and Reality

McDowell pointed out that such a theory is impossible. No theory could explain all concepts from scratch, since any theory has to employ some concepts of its own (1998: 194–5). The basis of linguistic understanding cannot be provided by a presuppositionless theory. It is a conceptual truth that any theoretical instruction, and any genuine explanation of meaning, presupposes certain prior knowledge and abilities on the part of the learner. And it is an empirical truth that human beings are not born with all of these prerequisites, but acquire at least some of them through training, a training which in turn builds on our innate abilities and propensities. Dummett later conceded that any theory must take some concepts for granted. He continued to insist, however, that an adequate theory of meaning must involve ‘conceptual analysis’, an informative explanation of the individual concepts of the language and of their interconnections (1993b: viii–ix). Dummett does not provide any adequate examples of such analysis. But it is clear that it cannot be disquotational. Instead, it would have to involve axioms that genuinely elucidate the meaning of terms, and deliver theorems that spell out the sense of sentences. Such analysis, however, need not amount to ‘conceptual breakdown’ (Evans and McDowell 1976: xi), that is, terminate with semantic simples as assumed by reductive analysis. Rather, it can take the form of Strawson’s connective analysis, of explicating concepts and paraphrasing sentences in an illuminating way. Admittedly, all explanations of meaning will eventually move in a circle. But this does not entail that they must all boil down to uninformative disquotations. For there are more or less illuminating circles. 3

Psychological modesty

Dummett is right, therefore, to extol the virtue of theories that are fullblooded in that they explain or analyse the terms of L. He also favours theories that are full-blooded in a psychological sense: they are supposed to make explicit the practical knowledge in fact possessed by competent speakers of a natural language (1993b: 13–15; 36–7). The archetype of such a theory is generative grammar. According to Chomsky, we have unconscious knowledge of a complex system of formation and derivation rules. Meaning and understanding a sentence involve operating a calculus the workings of which ‘are far beyond the level of actual or even potential consciousness’ (1965: 8). In his early essays, Davidson aspired to do for semantics what Chomsky had done for syntax. A theory of meaning was supposed to be an ‘empirical

Meaning and understanding

theory’ that tells us not just ‘what there is to be learned’, but also how it is learned. It accounts ‘for the workings of a natural language’ and yields at least ‘a feeble insight into the design of the machinery of our linguistic accomplishments’ (ITI 8, 24–5). Unlike Chomsky, however, Davidson felt that this insight is partly independent of experience: ‘we are entitled to consider in advance of empirical study what we shall count as knowing a language, how we shall describe the skill or ability of a person who has learned speaking a language’ (ITI 7–8). A priori reflections on the form of a constructive theory of meaning should explain ‘how finite resources suffice to explain the infinite semantic capacities of language’ (ITI 55, see 35, ch. 1). As competent speakers of a natural language we can produce and understand sentences which we have never encountered before. Furthermore, the number of these sentences is potentially infinite. The early Davidson drew two conclusions from this so-called ‘creativity of language’ or ‘semantic productivity’. The first is a compositionalist account of linguistic communication: ordinary speakers and hearers must calculate the semantic properties of sentences from the semantic properties of a finite number of constituents and from their mode of combination. Understanding a sentence is a matter of recursively deriving its truth-conditions from the relevant axioms of the truth-theory (Dummett 1993b: 13–5, 36–7). The second is that a learnable language must be finitely axiomatizable, and hence recursive in the style of a truth-theory. If a language L contained an infinite number of ‘semantic primitives’, its theory would have to contain an infinite number of axioms; yet such a language could not be mastered or acquired by finite creatures like ourselves (ITI ch. 1). It is a contingent fact about natural languages that they are learned. It follows that natural languages are learnable. Furthermore, there is no objection in principle to the idea that to be learnable, natural languages must possess certain features. But such transcendental arguments are precarious, and Davidson’s is no exception. For one thing, there are mathematical theories which are not finitely axiomatizable, yet can be learned (R. J. Haack 1978; cf. Matthews 1986). This holds for mathematical theories which employ axiom schemata rather than axioms. Such a schema is a representation of an infinite number of axioms by means of an expression containing syntactic variables and having axioms as values. Of course, to learn a theory that is not finitely axiomatized cannot consist in learning an infinite number of axioms. But it can consist in learning a schema for the construction of axioms. The point carries over to semantics.

245

246

Quine and Davidson on Language, Thought and Reality

A truth-theory can deal with the predicates of a language L in two different ways. Either it includes axioms that specify for each individual predicate under what conditions it is satisfied, as in: (VII) a sequence of objects X satisfies ‘xi is square’ iff the i-th object in X is square.

Or it includes a schema that specifies under what conditions any predicate of L is satisfied, as in: (VIII) a sequence X of objects satisfies Fxi, . . . xn iff the i-th to n-th objects in X stand in relation F.

(VIII) has the advantage of being applicable even when L contains a (denumerably) infinite number of sentence components. English arguably does so on account of containing infinitely many arithmetical predicates of the form ‘ . . . is divisible by n’. It is further arguable that speakers of a natural language learn not a specifiable and finite list of rules; instead, they master certain linguistic techniques, including that of predication alluded to in (VIII). That is why, unlike a theory listing a finite number of axioms, they have no problem coping with the fact that natural languages constantly gain and lose both singular terms and predicates. In any event, there is no problem about learning a theory of meaning that incorporates a schema like (VIII) instead of a finite number of axioms. Our ability to grasp a general recipe for the construction of specific axioms is no more mysterious than our ability to apply the specifications of a recursive theory. This means that the inference from learnability to finite axiomatization fails. Even the inference from learnability to a process of derivation is fallacious. There can be no a priori argument to the effect that certain abilities must be based on certain derivations, processes or mechanisms. For such claims are genetic in character. But statements about the factors operative in bringing about a certain kind of ability, understanding or knowledge do not follow a priori from a description of the result of that genetic process, namely the ability, understanding and knowledge in question. Compositionalism may be right to insist that, by and large, we would not credit someone with understanding a sentence unless we could also credit him with understanding the way it is composed out of words. But this does not show that the understanding of the sentence must be the result of tacit calculations. There is no conceptual necessity about our ability to understand an unlimited number of sentences issuing from the kind of derivation postulated by psychologically full-blooded theories.

Meaning and understanding

If someone could use and explain English sentences and their components, as well as react appropriately to their use by others, one could not deny that she understands them, even if she had not gone through such derivations. Nor is there anything unintelligible about the suggestion that our normal understanding of utterances is based not on tacit derivations, but the result of immediate, unreflective perception, albeit one which is the prerogative of those that are proficient in the language concerned (see chapter 7, section 3). In fact, the idea that normal speakers follow complex rules of derivation even though they can never become aware of this fact has been attacked from diverse quarters (Baker and Hacker 1984: chs. 8–9; Searle 1997). Its central shortcoming is that it ignores the difference between following a rule and merely acting in accordance with it. The latter presupposes that the rule provides the agent’s reason for acting as she does. But while the physiological causes of speech and understanding may be completely unknown to speakers, this cannot hold of their reasons.1 People often follow rules without explicitly consulting them, as in the case of proficient chess-players. In most cases, they will be able to specify these rules when prompted. But there are exceptions. For example, competent speakers may be incapable of explaining the difference between terms like ‘automatically’ and ‘inadvertently’, or between ‘bottle’ and ‘jar’ (Rundle 1990: ch. 4). However, they are nevertheless capable of recognizing the correct formulations of the relevant rules. As Dummett for one acknowledges, this is a minimum requirement for explaining a certain form of behaviour as guided by a rule (1993b: xi, 95–6). Alas, he ignores that even this potentiality is absent in the case of many of the rules featuring in constructive theories of meaning. Indeed their arcane apparatus is unintelligible to many competent speakers. Consider Davidson’s treatment of adverbial modifications (EAE ch. 6). He formalizes a valid inference like (I1 ) Mary kissed John in the garden at midnight; therefore Mary kissed John

through quantifying over events. (I2 ) ∃x (Kissing (x, Mary, John) & Occurred-in (x, the garden) & Occurred-at (x, midnight)); therefore ∃x (Kissing (x, Mary, John)). 1. Even though I have formulated it in terms of a contrast between reasons and causes, this argument is unaffected by Davidson’s celebrated thesis that reasons are causes. That thesis only implies that all reasons are causes, not that all causes are reasons. In his critique of Quine’s proximal theory, Davidson himself denies that neural events provide evidence or reasons for our beliefs.

247

248

Quine and Davidson on Language, Thought and Reality

But we cannot explain the fact that ordinary speakers recognize the validity of (I1 ) by reference to their implicit grasp of (I2 ). For (I1 ) is accepted by many people who are incapable of as much as learning the rules thereby imputed to them, even when these are expressed in a less formal way. This means that there is not even a minimal sense in which such rules guide their linguistic behaviour. Furthermore, paraphrases like (I2 ) depend on nominalizations which are derived from verbs (‘kisses’ – ‘kissing’); consequently they presuppose rather than explain our understanding of primary constructions like (I1 ) (Strawson 1992: ch. 8). More generally, our understanding of English sentences cannot be explained by the fact that we can learn to translate them into an interpreted formal language. For this is no more plausible than explaining our grasp of English by our ability to translate English sentences into Latin. Being able to translate from English into either of these two languages is a necessary condition of understanding English only for people who already know them. These points might be acceptable to the later Davidson. From 1973 onwards, he has explicitly disavowed the ambition of describing what is involved in actual linguistic training or communication, either as a matter of empirical fact or as a matter of conceptual necessity. A Davidsonian theory is ‘a model of the interpreter’s linguistic competence’, but it does not provide a psychological description of ‘the propositional knowledge of an interpreter’, or a neurophysiological description of ‘the inner workings of some part of the brain’. The claim is not that in everyday communication we rely on knowledge of a Tarskian truth-theory, but that such knowledge ‘would suffice for interpreting utterances’ even of a completely unknown language (NDE 438; ITI 171). We may insist that there must be some mechanism or other which allows ordinary speakers to understand, but this trivial claim adds nothing to Davidson’s account. The main point is a negative one: ‘It is obviously grotesque to suppose that in our everyday encounters we interpret the words and thoughts of others by working through the Byzantine sequence I have outlined’ (RD 264). On this modest construal, Davidson’s paraphrase cannot be faulted. Someone who understands the rules behind (I2 ) is capable of recognizing (I1 ) as valid, even though her route to that destination is very different from that of ordinary speakers. Nevertheless, even Davidson’s later position faces problems. The first is that it is unclear what purpose a consistently modest account serves. Davidson’s theory addresses a Kantian problem, namely

Meaning and understanding

how understanding and communication are possible. But whereas Kant sought to identify necessary conditions of our cognitive achievements, a project close to that of Dummett and the early Davidson, the later Davidson is officially concerned with sufficient conditions, namely with specifying knowledge which would suffice for understanding. What is the interest of this project? Why not stick to the common-or-garden explanations of words and sentences that ordinary speakers are capable of proffering, as urged by Baker and Hacker (1984: 232–7)? One possible motive is the Quinean aspiration of describing language in non-semantic terms. If we can show that meaning and understanding are possible on an austere extensional and physicalist basis, the threat which these higher level phenomena pose to a naturalistic world-view seems to be defused. We could get by on a physicalist shoe-string, even if we actually rely on something more lavish. However, even if one grants that the austere project is desirable, it is difficult to see how it is promoted by psychological modesty. Davidson concedes that meaningful discourse in fact relies on phenomena (thought and intentional action) that cannot be analysed in physical terms. His only consolation for physicalism is his token-token identity theory, according to which individual beliefs and actions are nonetheless identical with individual physical phenomena. That consolation, however, in no way depends on the feasibility of a modest theory of meaning. The second problem is that even the later Davidson does not stick to psychological modesty. Officially he disavows the ambition to describe actual psychological processes. In spite of disclaimers, however (SAL 3), he has continued to issue claims (some apodictic, others conjectural) about ‘what goes on’ in actual communication (NDE 441; see ITI 143, 277). According to these claims, everyday communication is, and indeed must be, based on constructing truth-theories on the basis of non-semantic evidence. Some commentators have suggested that such passages are simply a slip. But sticking to psychological modesty would leave gaping holes in Davidson’s edifice. It undermines his proof of the existence of events (see chapter 1, section 3). Even if one accepts that ‘our most common talk’ is substantially correct, modesty would prevent Davidson from claiming that we cannot ‘make sense’ of such talk without assuming that there are events (EAE 162). For he could not insist that those who accept (I1 ) must be quantifying over events, but only that they might be.

249

250

Quine and Davidson on Language, Thought and Reality

4

Language, norms and conventions

Psychological modesty is also at odds with Davidson’s recent account of linguistic communication. That account is dominated by an attack on a cluster of notions that include rules, conventions and a common language. Many readers have felt that this attack marks a sharp break with Davidson’s previous ideas, and even some of his followers have recoiled from it (Hacking 1986; Wiggins 1997: 20–1). There is indeed a shift in Davidson’s position, but it was a long time in coming. As early as 1973, he deplored the ‘uncritical evocation of the concepts of convention, linguistic rule, linguistic practice, or language-games’ (ITI 171). And ‘Communication and Convention’ (1982) is devoted to rejecting various types of linguistic conventions (ITI 266; see chapter 5, section 4 above). I shall concentrate on semantic or lexical conventions, conventions that fix the meaning of expressions. Davidson admits that it is arbitrary that we use a particular soundpattern to mean something specific. But he denies, rightly, that it must therefore be a matter of conventions (ITI 265). It is arbitrary that before going out I start by tying my left rather than my right shoe, yet this does not indicate the existence of a convention. In this context, it is crucial to distinguish two issues (SIO 113–14; Gluer 1999), which I shall refer to ¨ as normativity and conventionality. The first concerns the question whether meaningful speech is guided by rules. The second concerns the question whether such rules must be shared between speakers. Goaded by Bilgrami (1993), Davidson writes that ‘the yearning for [linguistic] norms is a nostalgic hangover from the dependence on a Platonic conception of meaning’ (RD 145). This statement is surprising, for three reasons. The first is that Davidson accounts for linguistic competence through a truth-theory. Such a theory involves linguistic norms, even if one leaves aside the fact that it incorporates the standard rules of inference. The axioms of the ‘scheme of reference’ are lexical rules for the singular terms and predicates, those of the ‘scheme of projection’ are recursive rules that specify how the satisfaction of complex expressions depends on that of simpler ones. Davidson cannot portray these axioms as mere empirical statements about the linguistic behaviour of speakers. For they license the move from one biconditional to the next in the canonical proof of T-sentences. In Davidson’s model of linguistic competence, speakers calculate the truth-conditions of actual or potential utterances; but such calculations proceed according to the axioms of the truth-theory.

Meaning and understanding

The second reason is that Davidson has recently embraced the Wittgensteinian idea that there is an irreducibly normative dimension to language. He accepts not only that there must be a ‘distinction between using words correctly and merely thinking one is using them correctly’, but also that this distinction requires ‘a norm, something that provides a speaker with a way of telling . . . that he has gone wrong’ (SAL 10–11; see RH 208, 359). There is only one way of reconciling these dicta. Davidson recognizes norms that evaluate uses of expressions as true or false, rational or irrational. What he repudiates are lexical norms of the kind invoked by the analytic/synthetic distinction, norms that distinguish between meaningful and nonsensical uses. We shall see instantly, however, that he cannot sustain this repudiation. The final reason for surprise about Davidson’s animadversions to norms is his ‘charitable’ insistence that we can interpret people only if we can make them out to speak and act in accordance with certain standards of rationality. Davidson tries to defuse the threat of inconsistency. Standards of rationality are essential to radical interpretation, he admits, but they create ‘no opportunity for judging the speaker’s use of language correct or incorrect’. For they tell us nothing about how people should or must speak to be understood. They impose no obligations on speakers; they are descriptive facts about what people do mean by what they say, about what is involved in (correctly) understanding them. (RD 146)

Up to a point, Davidson’s disclaimer is well-taken. If an interpretation ascribes absurd beliefs to the natives, his account implies that the interpretation rather than the natives is at fault. And if there is no interpretation that makes the natives out to be by-and-large rational, this is taken to show that they do not even speak a language or have thoughts, which would exempt them from judgements of correctness or incorrectness. The principle of charity imposes no obligations on the natives, because the latter have no obligation to engage in an interpretable form of behaviour. At this point, however, Davidson’s position starts to unravel. His account makes certain standards of rationality constitutive of concepts like interpretation, language or intentional action. Forms of behaviour that cannot be interpreted so as to conform to such standards simply do not count as linguistic or even intentional (see chapter 6, section 4). This has three embarrassing consequences. For one thing, it undermines

251

252

Quine and Davidson on Language, Thought and Reality

Davidson’s insistence that Quine’s rejection of the analytic/synthetic distinction ‘has saved the philosophy of language as a serious subject’ (SIO 145). As Lewis put it to him (1983: 112), on Davidson’s account it must be analytic that anybody to whom we can ascribe thoughts must satisfy the standards of rationality enshrined in the principle of charity. Initially, Davidson rejected this imputation (EAE 272–3). Later, however, he was compelled to admit that it ‘cannot be a factual question’ whether a creature with thoughts is approximately rational (RE 245). Davidson needs to treat his own pronouncements on rationality as conceptual in precisely the sense he officially denounces. By the same token, these standards of rationality indeed tell us something ‘about how people should or must speak to be understood’, not because they themselves are obliged to abide by them, but because failure to do so disqualifies them from engaging in intelligible behaviour. Davidson’s insistence that we are merely dealing with ‘descriptive facts’ ignores the definitory aspect of rules. It is precisely one of the functions of constitutive, as opposed to regulative, rules to lay down what counts as a such-such, what a thing must be like to satisfy a certain description (Searle 1969: ch. 2.5; Baker and Hacker 1984: ch. 7.3). Davidson cannot dispense with rules. The question is whether these rules are shared. This is not the problem debated by readers of Wittgenstein’s private language argument, namely whether there could be a language that is not or cannot be communicated to others. Davidson is ab initio concerned with cases of communication. His question is whether an appeal to conventions can elucidate the nature of linguistic communication. He adduces two basic reasons for a negative answer. The first is that explaining language by appeal to conventions is circular. Beliefs, desires, and intentions are a condition of language, but language is also a condition for them. On the other hand, being able to attribute beliefs and desires to a creature is certainly a condition of sharing a convention with that creature; while . . . convention is not a condition of language. I suggest, then, that philosophers who make convention a necessary element in language have the matter backwards. The truth is rather that language is a condition for having conventions. (ITI 280, see 27)

Davidson here argues for a Quinean conclusion, but in a distinctive fashion. It is prima facie plausible to hold that conventions presuppose language, since they must be expressed linguistically. Unlike Quine,

Meaning and understanding

Davidson does not reckon with the possibility that there might be implicit conventions that guide our behaviour without having been enunciated (see SIO 113–14). His argument runs as follows: (P1 ) All conventions (implicit ones included) presuppose beliefs and intentions about the beliefs and intentions of others. (P2 ) Beliefs and intentions about the beliefs and intentions of others presuppose language. (C) All conventions presuppose language.

The merits of (P1 ) depend on the concept of convention. It may be plausible for the demanding conception of Lewis (1969) that Davidson has in mind. But it is implausible given a weaker conception, according to which a convention is simply a shared arbitrary rule. If members of a community behave in a regular fashion, react to deviations not just with surprise but with disapproval, corrections or sanctions, and if these adverse reactions are generally accepted, then they share a rule (Hart 1961: 54–6; von Savigny 1988: ch. 2). And if that rule is arbitrary in the way in which Davidson acknowledges meaning to be, then they share a convention. Even if (P1 ) holds for conventions in this unassuming sense, (P2 ) will turn out to be unwarranted in the next chapter. Non-linguistic creatures can have not just beliefs and intentions, but also beliefs and intentions concerning the beliefs and intentions of others. This defuses Davidson’s circularity objection against explaining language and communication by appeal to conventions. Davidson’s second objection is that a knowledge of shared rules is ‘neither necessary nor sufficient for successful linguistic communication’ (SAL 2). According to a standard account inspired by Davidson’s own writings, communication is successful only if speaker and hearer share a truththeory that provides a recursive characterization of the truth-conditions of all potential utterances of a natural language L. Speaker and hearer approach each occasion of utterance armed with such a theory, and both assume that such a theory is shared by others, and assumed to be shared (NDE 436). Davidson claims that this account cannot make sense of the fact that malapropisms and similar foibles are not only quite common, but also commonly understood (NDE 440–1). In the appropriate circumstances, we have no difficulty in interpreting Mrs Malaprop’s remark ‘a nice derangement of epitaphs’ as meaning ‘a nice arrangement of epithets’. However, no antecedent truth-theory for English would deliver a T-sentence like

253

254

Quine and Davidson on Language, Thought and Reality

(16) ‘This is a nice derangement of epitaphs’ is true iff this is a nice arrangement of epithets.

We understand Mrs Malaprop, although her truth-theory differs from ours. Communication does not demand that speaker and hearer speak the same language or ‘mean the same thing by the same words’ (ITI 276–7). Davidson’s alternative account of successful communication runs as follows (ITI 277–8; NDE 441–2). We approach each communicative situation with a ‘prior theory’. But the prior theories which members of a linguistic community adopt differ substantially, not just from speaker to speaker but also for different interlocutors. Furthermore, speakers constantly have to react to unexpected utterances by passing from their prior theory to a ‘passing theory’, an ad hoc hypothesis which reinterprets the utterance in the light of the principle of charity. For the hearer, the prior theory expresses how he is prepared in advance to interpret an utterance of the speaker, while the passing theory is how he does interpret the utterance. For the speaker, the prior theory is what he believes the interpreter’s prior theory to be, while his passing theory is the one he intends the interpreter to use. (NDE 442)

Neither prior nor passing theory amount to a body of conventions which speaker and hearer use to communicate. The passing theory is shared between speaker and hearer, but it does not correspond to an interpreter’s general linguistic competence, since ‘knowing a passing theory is only knowing how to interpret a particular utterance on a particular occasion’ (NDE 443). A prior theory to some extent equips us to deal with a variety of different speakers and situations. But it is not shared between speaker and hearer, since it contains many idiosyncratic features. Driven by such considerations, Davidson reaches a notorious conclusion: ‘there is no such thing as language, not if a language is anything like what many philosophers and linguists have supposed. There is therefore no such thing to be learned, mastered or born with’ (NDE 446). He assures us that this is not to deny the existence of natural languages like Swahili or Basque. At stake is only the idea that such a language amounts to a set of precise and specifiable rules, shared knowledge of which is both necessary and sufficient for communication between its speakers (SAL 1–2; RD 117). Davidson is right to think that infelicities like malapropisms pose a threat to the standard account. He is also right to claim that in the roughand-tumble of everyday communication we often rely on non-linguistic

Meaning and understanding

pieces of information, whether they be general or context-dependent, and sometimes even on imagination. For this reason, knowledge of a truththeory, or of any other set of linguistic rules, would not always be sufficient for understanding. This is important, since it casts doubt on Davidson’s own compositionalism, the idea that for understanding an utterance it is not just necessary but also sufficient to understand its components and their mode of composition.2 Davidson is also right in holding that communication can succeed between individuals who do not share a language, even passively. In interactions with foreigners, we sometimes communicate by way of facial expressions, gestures and through intonation. The moot question is whether there is ‘communication by language’ without semantic conventions, as Davidson maintains (ITI 265). He supports this claim by several considerations. The first concerns malapropisms. When these are understood, it clearly amounts to linguistic communication. The question is whether such understanding relies on a background of shared conventions. We are often, albeit by no means always, capable of understanding linguistic infelicities, but only because they are recognizable and in some way intelligible deviations from a shared norm. We would not understand Mrs Malaprop had she said, for example, ‘a nice uldmob of delgolph’. Davidson’s second argument seems to avoid this comeback. In radical interpretation we manage to communicate, even though we do not share any conventions with the natives. He opines that ‘radical interpretation is so commonplace’, and concludes that the same holds for communication without convention (ITI 276). But this line of reasoning violates his professed psychological modesty. For it amounts to the claim that domestic communication is actually based on the construction of truth-theories under conditions of radical interpretation. More importantly, that claim is problematic, given Davidson’s own reservations about saddling ordinary speakers with tacit knowledge of a truth-theory. On the other hand, a consistently modest understanding of radical interpretation cannot sustain any claims about what linguistic communication must involve or cannot involve, whether as a matter of empirical fact (‘all cases of linguistic communication are based on calculations XYZ’) or because of the concept of a language (‘nothing counts as linguistic communication unless it is based on XYZ’). Consequently, Davidson cannot 2. Other worries about compositionalism are aired by Hintikka 1980; Baker and Hacker 1984: 211–18, 337; Schiffer 1987.

255

256

Quine and Davidson on Language, Thought and Reality

invoke a modest theory to support his conclusion ‘that there is no such thing’ as language governed by conventions. At most he could conclude that there need not be such a thing. Communication is not necessarily based on shared rules, since there is an alternative procedure – the construction of ‘passing’ truth-theories – that is capable of delivering the same result. Davidson’s tendency to regard radical interpretation as ubiquitous is connected to another idea, namely that ‘no two people actually speak the same language’ (SIO 115). He observes that ‘no two speakers speak in exactly the same way’, and concludes that ‘each speaker has, as Chomsky has maintained, a personal idiolect’. Indeed, because of the ever-changing nature of prior theories, these idiolects are transitory; strictly speaking, an individual’s language changes with every linguistic interaction (RD 117; see ITI 276–7; SCT 311). This reasoning is based on an untenable conception of what constitutes a (separate) language. From the fact that most people do not speak in precisely the same way it does not follow that they speak different languages. For not every divergence in linguistic use constitutes a distinct idiolect. Someone who persistently says that it is raining whenever it is raining diverges from common use. Yet he does not speak an idiolect, he is simply an idiot. Similarly, the fact that Americans do not use the sentence ‘God save the Queen!’ does not show that their language is impoverished, since its rules allow for the construction of that sentence (it shows rather that in this respect their political system is superior to the British). Nor did the British alter their language when they stopped using ‘God save the King!’ in 1952. For a related reason, Quine is wrong to identify a language with a theory, on the grounds that both are simply ‘a fabric of sentences’ (WO 11; RDH 308–11). A language like English is not a theory. For one thing, the identity of a language is determined not by sentences, but by the principles for the formation of meaningful sentences, that is by syntactic and semantic rules. For another, unlike a theory, a language does not state or predict anything. Finally, even if a language were a set of sentences rather than of rules for the construction of sentences, it could not be a theory. For a language must contain both sentences and their negations, which a coherent theory cannot. For the most part, Davidson sticks to a more modest position. He acknowledges that ‘society bends linguistic habits to a public norm’ and that knowledge of linguistic conventions is ‘a practical crutch of interpretation’, one that we cannot dispense with ‘in practice’. At the same time, he

Meaning and understanding

insists that we could do without it ‘in theory’ (ITI 278–9). Furthermore, he bluntly rejects the idea that speakers hold themselves responsible to an established use. It is absurd to be obligated to a language; so far as the point of language is concerned, our only obligation, if that is the word, is to speak in such a way as to accomplish our purpose by being understood as we expect and intend. It is an accident, though a likely one, if this requires that we speak as others in our community do.

Davidson asks, ‘what magic ingredient does holding oneself responsible to the usual way of speaking add to the usual way of speaking?’, and he maintains that any such ingredient ‘has nothing to do with meaning or successful communication’, but only with attaining some ‘social status’ (SAL 8–9; RD 118). This attack has a pleasant egalitarian ring to it, but it is uncompelling. The rhetorical question is easily answered. Speakers do not just ‘tend to use the same words to mean the same thing’, as Davidson grants (SAL 3). Another entirely mundane ‘ingredient’ of our linguistic practice is that aberrations are corrected – persistently in the case of children and learners, subject to custom and politeness in other cases. Furthermore, such corrections are standardly accepted by deviant speakers (Dummett 1986; Hacker 1988; Schroeder 1998: §17). Finally, these normative ingredients have everything to do with meaning. What they do not concern is speaker’s meaning. Davidson confuses the two. He accuses Dummett’s claim that words have a meaning independently of individual speakers of invoking ‘an elitist norm by implying that people not in the right social swim don’t really know what they mean’ (SAL 12). However, proponents of linguistic conventions have never suggested that speakers might be ignorant of what they mean by their words, in the sense of knowing what they meant to say, but rather that they might be ignorant of what these words mean, i.e. of what they actually said. Davidson writes: ‘An utterance has certain truth-conditions only if the speaker intends it to be interpreted as having those truth-conditions. . . . A malapropism or slip of the tongue, if it means anything, means what its promulgator intends it to mean’ (SCT 310). But although Mrs Malaprop may have intended to use a phrase meaning ‘a nice arrangement of epithets’, this is not what the phrase she actually used means. Davidson’s position smacks of the Humpty-Dumpty theory of meaning. He anticipated this charge: ‘Humpty-Dumpty is out of it. He cannot mean what he says he means because he knows that “There’s

257

258

Quine and Davidson on Language, Thought and Reality

glory for you” cannot be interpreted by Alice as meaning “There’s a nice knockdown argument for you”’ (NDE 440). It is only subject to constraints of interpretability that ‘what a speaker’s words mean’ is ‘what he intends them to mean’. He must have provided sufficient ‘clues’ to be understood (SIO 13–14, 28). Unfortunately, one cannot avoid Humpty-Dumpty semantics simply by meaning to do so. No matter how many clues a speaker drops, what his utterance literally means can differ from how he intends it to be understood. This is simply part and parcel of the notion of literal meaning. Surprisingly, Davidson professes allegiance to Grice’s distinction between speaker’s meaning and literal meaning (NDE 434–7). He tries to prise apart what is literal from what is conventional. His new conception of literal meaning goes by the name of ‘first meaning’. The first meaning of an utterance is determined by the speaker’s ‘semantic intention’, the intention to utter words that the hearer interprets in a certain way (LL 293). It is how she intends it to be interpreted in the first instance, before phenomena like metaphor or pragmatic implicatures come into play. But it remains unclear how to single out this first instance, without an appeal to conventional meaning. More seriously, ‘first meaning’ is in no way equivalent to literal meaning. Imagine that an adviser to President Bush has the task of reporting that US exports are down. She knows, however, that Bush persistently confuses the terms ‘export’ and ‘import’, and she prefers him to believe that imports are down. In that case she could say ‘Our exports are down’, with the intention of inducing the belief that US imports are down, thereby giving it the first meaning that US imports are down. But of course this in no way affects the literal meaning of her utterance. On the one hand, Davidson insists that communication succeeds when there is a ‘fit between how speakers intend to be interpreted and how their interpreters understand them’ (SCT 311). On the other, he grants that ‘if communication succeeds, speaker and hearer must assign the same meaning to the speaker’s words’ (ITI 277). The Bush case shows that these two specifications do not coincide. What is more, the second specification is superior, since it captures the idea of mutual understanding. The first captures only the idea of successfully pursuing ‘ulterior purposes’ with the help of speech acts, something which is semantically irrelevant by the lights of Davidson’s own insistence on the ‘autonomy of meaning’ (chapter 5, section 4 above).

Meaning and understanding

In other passages Davidson seems intent to simply replace the concept of the literal meaning of a word by that of speaker’s meaning. The notion of literal meaning is bound up with the notion of a linguistic mistake as a deviation from a shared norm. Davidson acknowledges the existence of this notion, but only to dismiss it as ‘shallow’. His reason is that we could do without conventions, and that conventions are not ‘essential to verbal communication’ (SAL 2, 12). What is essential, Davidson maintains, is the grasp of semantic intentions in passing theories. It is not easy to assess this claim. The possibility of interpretation from scratch does not substantiate it without further ado. For even at an early stage, such interpretation builds, if tentatively and precariously, on rules that interpreter and native have begun to share, e.g. concerning assent. On the other hand, it is conceivable that humans should communicate through noises, yet aided by nothing other than intimate acquaintance, luck and ingenuity. But verbal communication without conventions, and hence, as we have seen, without literal meaning, would definitely be incapable of expressing the fine-grained thoughts we can express by language. It would therefore come dangerously close to animal communication by gestures and expressive behaviour, a form of communication that is not linguistic. It is only through linguistic conventions that we can communicate fine-grained thoughts even to total strangers, people who are not in tune either with the speaker’s thoughts or with the circumstances of utterance. Even if Davidson’s scenario does not fall foul of the essence of linguistic communication, it is oblivious to the profound anthropological function that a shared language has for our species of social primates. 5

Thought, causation and first-person authority

Davidson has increasingly included a causal element in his semantics, partly for the sake of solving the extensionality problem, partly as a result of his exchange with Quine on the location of stimulus, and partly in reaction to externalism. The meaning of simple sentences is determined by ‘prompted assent’, and hence by the causal relation between a speaker’s disposition to assent and the causes of that disposition. Because of the close link between thought and language, what imbues sentences with meaning also imbues thought with content: ‘causality plays an indispensable role in determining the content of what we say and believe’ (SIO 150;

259

260

Quine and Davidson on Language, Thought and Reality

see IA 59). This causal theory is not meant to replace his truth-conditional account, but to specify what determines the truth-conditions: ‘the situations which normally cause a belief determine the conditions in which it is true’ (SIO 196–7). Davidson’s causal theory is distal rather than proximal: the meaninggiving causes are macroscopic phenomena external to the speaker. But where precisely should we locate the relevant cause or stimulus? Davidson’s answer appeals to ‘triangulation’ – a three-way relation between a speaker, a hearer and their shared environment. Davidson maintains that only creatures that communicate about a shared environment can have the idea of objective truth; from this he infers, first, that the capacity for linguistic communication is a precondition for having thoughts, and, secondly, that language is intrinsically social, that is, tied to communication (see chapter 9, section 4). Triangulation is also supposed to solve the problem of locating the stimulus (SIO 83, 117–19, 130). Imagine a learner and a teacher. The learner notices various similarities in her environment, and she will find her verbal responses to these situations reinforced by the teacher, provided that the latter judges her to respond similarly to similar situations. This process presupposes that teacher and learner have similar standards of similarity, both as regards the situations (objects and events) and as regards the verbal responses. It also implies that the relevant stimuli cannot be patterns of neural stimulation, since these proximal causes are not noticed by either teacher or learner. Rather, the causes that determine the content of our beliefs and utterances lie at the intersection of the causal chains that extend, respectively, from the speaker (learner or linguist) and hearer (native or teacher) to the events and objects in question. The meaning of a sentence is determined by the ‘common cause’ of different speaker’s dispositions to assent to it. ‘Communication begins where causes converge: your utterance means the same as mine does if belief in its truth is systematically caused by the same events and objects’ (SIO 151; see MTE 77). As the example of Quine shows, Davidson is not alone in propounding a causal theory of meaning. At the same time, such theories have faced taxing criticisms by authors as diverse as Wittgenstein, Chomsky and Kripke. According to Davidson, ‘a sentence that someone is inspired (caused) to hold true by and only by sightings of the moon is apt to mean something like “There’s the moon”’ (SIO 29; see 44–5; 150–2). One difficulty for causal theories is to explain more complex sentences, including empirical

Meaning and understanding

sentences about the past and future, or even the negation of simple cases (‘The moon is not there’). Davidson cannot solve this problem by insisting that such sentences are caused by facts of the appropriate kind, since he insists that causal relations obtain only between events (EAE ch. 7). His solution is to combine the causal theory with holistic elements close to conceptual role semantics. The meaning of our simplest sentences – observation sentences – is determined by their causes. The meaning of more complex sentences derives from their logical and probabilistic relations to other sentences (SIO 147; ITI 225). Yet ‘what ultimately ties language to the world is that the conditions that typically cause us to hold sentences true constitute the truth-conditions, and hence the meanings, of our sentences’ (FDT 275; see SIO 196–7, 201; PCT 15; CT 195; RI 21). Rumours about his rationalist affiliations notwithstanding, Davidson joins hands with Quine in affirming a ‘commonplace of the empirical tradition’, namely that language is anchored to reality through causal conditioning (SIO 43, see 29). But can a causal theory account even for simple sentences? The meaning of a sentence s is not determined simply by the conditions that cause natives to employ s, since there are non-semantic motives for using or withholding s. Like Quine, Davidson avoids this difficulty by focusing on the conditions under which natives hold true or assent to s. Nevertheless, several problems remain. For one thing, meaning/truth-conditions are not determined by causal relations between individual events. The meaning of Jane’s utterance ‘The moon is coming out’ is not captured by (17) Jane’s holding true ‘The moon is coming out’ is caused by the event of the moon coming out.

For Jane’s holding true might well have been caused instead by her children starting to howl like wolves, or by any number of other events. Yet this would leave the meaning/truth-conditions of her utterance unaffected. Davidson recognizes this problem. Contrary to his official theory of causation, he describes the semantically relevant causes of linguistic behaviour not as individual events but as situations, circumstances, episodes and conditions. He insists that meanings are constituted by ‘the conditions that typically cause us to hold sentences true’ (FDT 275, my emphasis). Unfortunately, this does not solve the problem of deviant causal chains. For what makes the causation of assent ‘typical’ or ‘normal’? Davidson provides no answer to this question. The obvious solution is that the causal chain is normal when the assent to s is caused by those conditions which

261

262

Quine and Davidson on Language, Thought and Reality

license the utterance of s (the coming out of the moon in our case). But far from explaining the truth-conditions of sentences by reference to the causes of linguistic dispositions, this would explain the typical causes of the latter by reference to assertion or truth-conditions. And any solution which avoids this circularity leaves open the possibility that the utterance of s is caused by conditions that have nothing to do with what s means. This problem is linked to another one, namely the failure of causal theories to capture the normative nature of meaning. Compare (18) Speakers hold true the sentence ‘The moon is coming out’ always and exclusively as a causal result of the moon coming out. (19) ‘The moon is coming out’ means the same as ‘The natural satellite of the earth is appearing from behind the clouds’.

(19) can serve as an explanation of what ‘The moon is coming out’ means, an explanation by reference to which utterances can be assessed as correct or incorrect. By contrast, (18) is an empirical hypothesis. It implies that an utterance of ‘the moon is coming out’ without the moon coming out is a statistical exception, but not that it is a mistake (Hacker 1998a). Davidson himself insists on this normative dimension in his critique of dispositional accounts of concepts (chapter 9, section 2 below), but it is at odds with his causal theory. A final problem concerns the genetic aspect of Davidson’s theory. He accepts the externalist view that what speakers can refer to and mean by their expressions is constrained by the original causal nexus between these expressions and reality in the learning situation. Similarly, propositional attitudes ‘are identified in part by the social and historical context in which they were acquired; in this respect they are like other states that are identified by their causes, such as suffering from snow blindness’ (SIO 51). Consider ‘Swampman’, a miraculously created molecule-formolecule replica of Davidson. Swampman could not mean what Davidson does by his words, even if he used them in the same way; indeed, he could not mean anything at all, because he has not acquired these words in the appropriate manner (SIO 18–19). For the triangular learning process is ‘not just a story about how we learn to use words; it must also be an essential part of an adequate account of what refers to and what they mean’ (SIO 43–4). However, in his attack on Quine’s proximal theory Davidson rightly notes that grasp of meaning depends only on ‘the end product’ of the

Meaning and understanding

learning process: ‘use of words geared to appropriate objects and situations . . . two speakers who “mean the same thing” need have no more in common than their dispositions to appropriate verbal behaviour’ (SIO 44, see 29). Whether someone understands a term depends on her ability to use it appropriately, including the ability to explain it and to react appropriately to its use by others. Ex hypothesis, Swampman possesses these proficiencies. How he acquired them is without semantic significance. Consequently, Swampman is a competent speaker of English, and Davidson’s emphasis on the causal genesis of speech is misguided. At the same time, Davidson must be applauded for his refusal to sacrifice first-person authority on the altar of his causal theory. Unlike Quine and certain versions of externalism, he does not make meaning and understanding depend ‘on causal relations of which speakers may well be ignorant’ (SIO 151n, see 197–202). For this would be incompatible with the ‘presumption’ that speakers are ‘not mistaken’ in their ‘firstperson present tense claims’ about what they think and mean, a ‘unique sort of authority’ that does not attach to their claims about the material world or the minds of others (SIO 3, 16, xiii). Davidson is also to be applauded for recognizing that this authority is not due to our having infallible introspective access to a mental realm. First-person presenttense psychological claims are not normally based on evidence or observation of any kind; indeed, ‘there is no way we know’ what we think (SIO 4–5, 66). Davidson explains first-person authority about thought by reference to first-person authority about meaning (SIO 3, 12): (P1 ) I know authoritatively that I hold true the uttered sentence ‘p’. (P2 ) I know authoritatively what the uttered sentence ‘p’ means. (C) I know authoritatively what I hold true and hence what I believe.

But as Davidson recognizes, ‘normally I know what I think before I speak or act’ (SIO 15), and his explanation does not account for this knowledge. Moreover, (P1 ) assumes rather than explains a speaker’s authority about what sentences he holds true (Hacker 1997: 297–8). Other problems arise with respect to Davidson’s defence of (P2 ). He derives first-person authority concerning speaker’s meaning as a precondition of interpretation. We know that we communicate, i.e. interpret one another. But whereas the hearer’s interpretations are always precarious and liable to diverge from homophonic translation,

263

264

Quine and Davidson on Language, Thought and Reality

the speaker, after bending whatever knowledge and craft he can to the task of saying what his words mean, cannot improve on the following sort of statement: ‘My utterance of “Wagner died happy” is true if and only if Wagner died happy’. (SIO 13, see 66)

Why? Because ‘unless there is a presumption that the speaker knows what she means, i.e. is getting her own language right, there would be nothing for an interpreter to interpret’. And ‘whatever she regularly does apply them to gives her words the meaning they have and her thoughts the content they have’ (SIO 37–8). I agree that without the presumption that the speaker knows what she means by her words there would be nothing to interpret. But as we have seen, how an individual A uses an expression e does not determine what e means, unless he is defining or redefining e. It may determine what A means by e. However, unless A can do better than provide a homophonic T-sentence, unless he can explain his utterance, A understands disquotation, not the words he uttered. Consequently, Davidson’s explanation of first-person authority about meaning is at least incomplete. At the same time, Davidson supplies an inventive argument why firstperson authority is compatible with what he regards as an externalist insight, namely that the content of utterances and thoughts depends at least partly on ‘causal interactions between people and parts and aspects of the world’ (SIO 24–5, 29). In a first step, he questions the inference from the premise that thoughts are identified by reference to something outside the head to the conclusion that thoughts ‘ain’t in the head’ (Putnam 1975b: 227). From the fact that my skin is sunburnt only if it stood in a certain causal relation to the sun it does not follow that the sunburn is not a condition of my skin. Consequently, externalism is compatible with anomalous monism, the idea that thoughts are ‘inner in the sense of being identical with states of the body’ (SIO 20, see 31–2, 48). However, while being sunburnt is a state of my skin, it cannot be identified as such by simply looking at my skin. By the same token, even if my thoughts are inner states of mine, I may not know what I think, if this requires knowledge of external factors that identify my thoughts, factors that may be beyond my ken. Davidson believes that this worry is due to the idea that thoughts are mental representations and that ‘to have a thought is to have an object “before the mind”’ (SIO 37, see 33–4, 46, 52, 58–9, 184). Although he denies that having a thought is to stand in a relation to a proposition, Davidson sees no alternative to treating belief sentences like

Meaning and understanding

(20) Paul believes that the earth moves

as relational (SIO 57–8). This is in line not just with his paratactic account of indirect speech, but also with Quine’s quotational account of propositional attitudes (ITI ch. 7 and WP ch. 17; WO 211–13; PT 67–73; see LePore and Loewer 1995; Orenstein 2002: 165–71). As with modality, Quine despairs of accommodating de re cases such as ‘There are some whom Ralph believes to be spies’. But other types of propositional attitudes can be construed as ‘relations between people and sentences’ (SS 93). In spite of creating an opaque context, (20) can be incorporated into canonical notation by treating it as a quotation: (20
) Paul believes ‘The earth moves’.

Similarly, Davidson incorporates (20) into a purely extensional truththeory. As in the case of non-declaratives, he achieves this through a paratactic analysis which links two independent sentences, each amenable to a truth-theory: (20∗ ) Paul believes that: The earth moves.

In (20∗ ), ‘that’ functions as a cataphoric demonstrative which picks out the attributor’s next utterance of ‘The earth moves’. In uttering (20), I am saying that ‘Paul believes what I would believe if I were sincerely to assert what I say next’. (20) relates Paul to an ‘object which in some regular way indicates Paul’s state of mind’ (SIO 63). The object in question is an utterance or sentence with the same truth-conditions as the belief it is used to identify. And if Paul says ‘I believe that the earth moves’, he expresses an attitude towards the sentence ‘the earth moves’, namely that of holding it true (SIO 210–11). As regards Davidson’s account of first-person authority, the punch line is this. In ascribing ‘propositional attitudes to men and beasts by quotation’, Quine does not ascribe ‘an understanding of the quoted language, or of any. A cat can believe “A mouse is in there”. The language is that of the ascriber of the attitude . . . The cat is purportedly in a state of mind in which the ascriber would say “A mouse is in there”’ (PT 68–9). As we shall see in the next chapter, Davidson contests the attribution of thoughts to beasts. But he acknowledges Quine’s reason for regarding it as compatible with the quotational and paratactic accounts (ITI 167). In order to know what he thinks, Paul need not ‘be directly acquainted with, or be able in some special way to identify . . . the objects that define (give the

265

266

Quine and Davidson on Language, Thought and Reality

contents of ) his thought’ (SIO 58, see 48–9). We can identify a belief of Paul by using a sentence that he is not familiar with, as when we use English sentences to ascribe thoughts to people who do not speak English. We specify what Paul thinks by reference to an object, yet he need not be related to that object. Accordingly, having a belief is not a relation to either a proposition or a sentence, but the ‘modification of a person’, and, more specifically, a dispositional mental state (DD 231–2; SIO 64). In talking about the beliefs of people we no more need suppose that ‘there are such entities as beliefs’ than in talking about the weights of objects we need suppose that there are weights for objects to have. To say that x weighs 10kg is to relate x not to ‘a weight’, but to other material objects, according to one of their properties. Similarly, talking about beliefs is a way ‘to keep track of the relevant properties of and relations among the various psychological states’ of people, for the purpose of explaining their behaviour (SIO 60, see 75, 214–18). The claim that beliefs are dispositional mental states is contentious, and so is the idea that (20) relates Paul to a sentence (Hacker 1998b). We identify his belief by means of a sentence like ‘The earth moves’. But the belief is identified by reference to what he believes – namely that the earth moves – rather than to sentences expressing what he believes, because what he believes can be expressed by different sentences. In one crucial respect, however, our previous discussions (chapter 3, section 1 and chapter 4, section 3) support Davidson. Admittedly, thatclauses and other noun-clauses such as ‘what Paul believes’ refer to what we believe. (20) entails, for example, that there is something Paul believes. But this ‘something’ is merely an object of discourse, and the noun-clauses are what White (1972) calls intentional rather than object-accusatives. ‘Paul believes Byers’ entails that there is an object x such that Paul believes x. Here the psychological verb expresses a genuine relation between two relata. By contrast, ‘Paul believes that the earth moves’ does not entail that there is an object x such that Paul believes x. Similarly, ‘I know Paul’ entails that there is an object x such that I know x. But ‘I know what Paul believes’ no more entails that there is an x such that Paul believes x and I know x than ‘I know what Paul weighs’ entails that there is an x such that Paul weighs x and I know x. ‘What Paul believes’ and ‘what Paul weighs’ incorporate interrogative rather than relative pronouns. Consequently, believing that p is no more a genuine relation than weighing n kilograms.

Meaning and understanding

By the same token, knowing what Paul believes is not to be acquainted with an object but, roughly speaking, the ability to answer the question ‘What does Paul believe?’ And to have first-person authority, Paul need not be intimately acquainted with a mental representation, he only needs the authority to answer that question.

267

9

Thought and language

Quine and Davidson adopt a third-person perspective not just on meaning, but also on the mind. Mental phenomena must be manifestable, at least in principle. The manifestability of the mental has a distinguished pedigree in the pragmatists, Wittgenstein and Ryle. But it is stronger and more controversial than the manifestability of meaning. Like our two protagonists, I shall concentrate on what Davidson calls ‘thought’ (SIO 98–9). By this he means all the propositional attitudes, but this generic term is less misleading, since neither Quine nor Davidson regards beliefs, desires, etc. as attitudes towards propositions. In Quine, there are three lines of reasoning to support the manifestability of the mental. The first two are connected to his naturalistic preoccupation with science. Naturalized epistemology elucidates the relation between theory and evidence through a scientific account of the relation between theoretical talk and observational talk. It can do so, because both scientific theories and scientific observations must be capable of being expressed in words (WO 26; OR 89; RR 34–9; NNK 74–5). This claim seems incontestable. Even if there are ineffable truths and observations that cannot be expressed in language, such as the ineffable insights and experiences alleged by mystics, they will not qualify as scientific, since science is a cooperative attempt to establish the nature of reality. Furthermore, the requirement of manifestability carries over to nonscientific thoughts, in so far as they are studied by science. The proper perspective for psychology is methodological behaviourism. Even if they exist, essentially private mental phenomena like ideas or sensations do not provide us with the kind of ‘intersubjective checkpoints’ required by a scientific explanation of human behaviour (RHS 74; SS 5; Gibson 1981: 203–4). It is legitimate to postulate mental phenomena that are not directly

[268]

Thought and language

observable, just as physics postulates elementary particles to explain observable events in nebula-chambers and accelerators. But these postulations remain tied to the explanation of observable phenomena, and to the possibility of identifying entities of the postulated kind at least indirectly. Once more, Quine’s position is difficult to contest. Even his mentalist opponents like Chomsky and Fodor would not deny that the postulation of unobservable phenomena remains ultimately tied to the explanation of observable phenomena. They merely claim that mental phenomena denied by Quine and Davidson – such as mental representations and tacit knowledge – are scientifically respectable because of the part they play in the explanation of, for instance, our linguistic competence. It may be possible to postulate unobservable mental phenomena in a scientific spirit, but only if one accepts that they are indirectly observable. It may be possible to cling to ineffability, but only if one sheds the pretence of science. One cannot coherently be both a scientist and a mystic on the same issue. Quine’s third reason for the manifestability of thought is independent of scientific methodology. The capacity for thought requires the capacity to manifest these thoughts. People may be unwilling to divulge their beliefs, or be temporarily prevented from expressing them. But beliefs cannot be ascribed to creatures or things that by their very nature are incapable of manifesting them. For ascribing thoughts makes sense only in cases where we have criteria for identifying thoughts (TT 2; Glock 2000). Although thoughts can be more or less definite, more or less specific, it makes sense to ascribe a specified thought that p to a creature a only if something counts as a thinking that p rather than that q. In a similar vein, Davidson denies ‘that thinking can be reduced to linguistic activity’, while insisting that there are ‘conceptual ties’ between thought and behaviour which ensure that thoughts can be correctly ascribed on the basis of behaviour. We cannot make sense of the notion of a belief as a private attitude completely detached from behaviour and its explanation (SIO 99–100; see ITI 159, 170). In this respect, Davidson agrees with Quine and Wittgenstein. However, there is a crucial difference. They hold that non-linguistic creatures like neonates and animals are capable of having thoughts of a simple kind, namely those that can be expressed in non-linguistic behaviour. Thus Wittgenstein famously suggested that a dog can believe that its master is at the door, but not that its master will return the day after tomorrow (1967a: 174). Davidson adopts a more radical form of lingualism, which is reminiscent of the pragmatists. He does not impute any kind of priority to language over thought (ITI 155–7; STL 17).

269

270

Quine and Davidson on Language, Thought and Reality

But he holds that thought and language necessarily go together: ‘a creature cannot have a thought unless it has language’ (SIO 100; see ITI 163). This claim is famous among philosophers, infamous among zoologists and pet-owners. In this chapter, I shall discuss four arguments Davidson advances in its support:

r the argument from the intensional nature of thought (section 1); r the idea that thoughts involve concepts (section 2); r the argument from the holistic nature of thought (section 3); r the claim that belief requires the concept of belief (section 4). According to Davidson, the first three arguments point in the right direction but are inconclusive, which is why he favours the last one. I am inclined to reverse this assessment. While Davidson’s favourite argument fails, his other arguments shed valuable light on the extent to which thought requires language, because they point to three respects in which our practice of thought-attribution is richer than the animal behaviour we explain by it. These arguments do not preclude us from ascribing simple thoughts to animals. But they show that the rich idiom we employ in such ascriptions has conceptual connections that go beyond the phenomena to which it is applied. 1

The intensional nature of thought

Davidson takes his cue from Malcolm, who relates the following tale: Suppose our dog is chasing the neighbour’s cat. The latter runs full tilt toward the oak tree, but suddenly swerves at the last moment and disappears up a nearby maple. The dog doesn’t see this maneuver and on arriving at the oak tree he rears up on his hind feet, paws at the trunk as if trying to scale it, and barks excitedly into the branches above. We who observe this whole episode from a window say, ‘He thinks that the cat went up that oak tree’. (1977: 49)

Against Malcolm, Davidson insists that strictly speaking the dog cannot believe anything, because he does not possess a language (SIO 96–100; ITI 155). Davidson would be rightly unimpressed by the charge that this denial is deplorably anthropocentric or insufficiently naturalistic. There is both biochemical similarity and evolutionary continuity between us and certain non-linguistic animals. But it does not follow that they must approximate to our mental life. Although it is probable that our closest ancestors without language shared many of our other mental capacities, these ancestors are extinct; and there is no guarantee that the biologically

Thought and language

closest extant species is mentally close to us. It so happens that chimpanzees share 98 per cent of our DNA. However, this does not show that they share 98 per cent of our mental life, because small biochemical differences may lead to significant differences in terms of our mental vocabulary. That vocabulary captures neither genetic nor neurophysiological differences, but differences in the kinds of capacities we humans are interested in. To that extent, our mental concepts may be anthropocentric, yet the view that these concepts preclude application to non-linguistic creatures is not (see SIO 96). A more serious objection is this: animals must be capable of having thoughts because we have no better way of explaining and predicting their behaviour than by attributing thoughts to them (Bennett 1976: §§7–8). According to Davidson, this provides a pragmatic justification for our attribution of thoughts to animals, but it does not show that animals can have thoughts. In attributing thoughts to animals, we merely treat them as if they were capable of acting for reasons (beliefs and desires), just as one might explain the movements of a heat-seeking missile by ascribing to it the desire to destroy an airplane. In this way, we can continue to explain the behavior of speechless creatures by attributing propositional attitudes to them while at the same time recognizing that such creatures do not actually have propositional attitudes. We will be bound to acknowledge that we are applying a pattern of explanation that is far stronger than the observed behavior requires, and to which the observed behavior is not subtle enough to give point. (SIO 102)

This account treats thought-attributions to animals as useful fictions. However, Davidson concedes that animal behaviour is much closer to human behaviour than the movements of heat-seeking missiles, and that we know of no better way of explaining the behaviour of animals than by attributing thoughts to them. These concessions invite an objection Davidson ignores. We regard attributing thoughts to animals not just as convenient, as he would have it, but as entirely justified. For, unlike attributing desires and beliefs to heat-seeking missiles, such attributions are not based on technological ignorance but on a biological insight, namely that the life and behaviour of animals shows them to have both wants and perceptual capacities. Davidson might reply that the alleged insight is merely an illusion of Aristotelian folk-biology, since animal behaviour could be fully explained by physiological processes, if our knowledge were sufficiently advanced. However, this invites the question of why human behaviour

271

272

Quine and Davidson on Language, Thought and Reality

should resist such explanation. Davidson has a well-known response: human action defies psychophysical laws because it is based on propositional attitudes forming a holistic web. But he could not invoke that response in the present context without presupposing that animals, unlike humans, lack propositional attitudes, which would beg the question at issue. Nevertheless, there may be a kernel of truth in Davidson’s suggestion that the pattern of explanation we employ with respect to animal behaviour in some sense outstrips the explanandum, because it is originally tailored to the explanation of the more complex behaviour of linguistic creatures like ourselves. But we need compelling arguments for the claim that strictu sensu non-linguistic creatures cannot even have simple beliefs. One problem with attributing thoughts to animals is this. Without verbal responses we cannot make the fine distinctions between different thoughts expressed in the same non-verbal behaviour. Thoughtattributions to humans create intensional contexts: if we substitute co-referential terms within the content-clause, this may lead from a true attribution (e.g. ‘Sarah believes that Cicero was Roman’) to a false one (e.g. ‘Sarah believes that Tully was Roman’). In the case of animals, by contrast, substitution of co-referential expressions often leads from attributions which we commonly regard as true to attributions which are absurd or unintelligible. The tree that the cat went up also happens to be the oldest tree in sight and the same tree the cat went up last time the dog chased it. But does Malcolm’s dog believe that the cat went up the oldest tree in sight, or the one it went up last time? Equally, a dog can know that its master is at the door. But does it also know that the president of the bank is at the door? ‘We have no real idea how to settle, or make sense of, these questions’ (ITI 163; see SIO 97). The reason, one might add, is that the dog can think neither that its master is the president, nor that he is not. One response to this failure of intensionality is to hold that in the sentence (1) The dog thinks that the cat went up that oak tree

the expression ‘that oak tree’ occurs transparently (in Quine’s terminology). Accordingly, (1) is paraphrased so as to avoid problems of intensionality, e.g. as (1
) The dog thinks, with respect to that oak tree, that the cat went up it.

Thought and language

Davidson retorts (SIO 98): But such constructions, while they may relieve the attributer of the need to produce a description of the object that the believer would accept, nevertheless imply that there is some such description; the de re description picks out an object the believer could somehow pick out. In a popular if misleading idiom, the dog must believe, under some description of the tree, that the cat went up that tree. But what kind of description would suit the dog?

In this passage, Davidson initially insists that there must be some description of the object ‘that the believer would accept’. Trivially, that requirement cannot be met by a creature without language; but to insist on it begs the question in favour of lingualism. Davidson seems to recognize this. In the sequel, he does not insist that there be a description which would be accepted by the believer, but merely that there be a description which ‘would suit’ him in the sense that ‘it picks out an object the believer could somehow pick out’. This weaker requirement is reasonable. If Malcolm’s dog could not distinguish the oak tree from among other objects (e.g. the pine tree or the garden fence), we might still explain his behaviour by reference to the oak tree, just as we might explain the convulsions of an oyster by reference to its being pricked with a needle. But (1
) would no longer be appropriate. For transparent constructions like ‘with respect to’ or ‘of’ require an anaphoric referent in the subsequent content-clause, an ‘it’ which the now disabled dog could not distinguish from other things. It remains an open question, however, whether that weaker requirement might not be met by non-linguistic creatures on account of their possessing certain discriminatory capacities. The dog believes something of the tree ‘under some description’, namely one that expresses those features by which the dog recognizes the tree and distinguishes it from other objects. Presumably to exclude this kind of response, Davidson insists that the dog cannot believe of an object that it is a tree unless it has many general beliefs about trees: that they are growing things, that they need soil and water, that they have leaves or needles, that they burn. There is no fixed list of things someone with the concept of a tree must believe, but without many general beliefs there would be no reason to identify a belief as a belief about a tree, much less an oak tree. (SIO 98)

This passage raises two objections: a general one concerning the holistic connections between thoughts, discussed in section 4, and a more specific

273

274

Quine and Davidson on Language, Thought and Reality

one concerning concepts to which I turn now. The original argument from intensionality questions whether there is a that-clause which would capture the content of an animal’s thought. This question now gives way to the question of whether such clauses would not be implying too much conceptual sophistication and to the question whether they would not drag in too many other thoughts. 2

Animals, thoughts and concepts

The obstacle which concepts create for animal thought is this. The thoughts we ascribe to animals in common parlance involve concepts with which the animal cannot be credited. That objection is explicit in Dummett, who denies that animals possess genuine concepts, and by implication genuine thoughts (1993b: chs. 12–13). Davidson has recently sharpened the link between concepts and thoughts along similar lines. He insists, first, that concept-possession and the ability to have thoughts amount to one and the same thing; and, secondly, that both are confined to language-users (STL 24–5; SIO 123–4). In fact, the first claim provides the rationale for the second: attributing thoughts to animals on the basis of non-linguistic behaviour is misguided, since these thoughts involve concepts which themselves cannot be attributed on such a basis. This raises two problems. Firstly, can animals possess concepts at all, and, if so, what kind of concepts? Secondly, if they cannot have concepts of any kind, does that really preclude them from having thoughts? One will have to answer the second question in the affirmative if one accepts the idea that thoughts are abstract entities which have concepts as their components, because that implies that one cannot have or grasp the thought without having or grasping its constituent concepts. But as we have seen, Davidson rejects the idea that thinking is a relation between a subject and an abstract entity, and regards it simply as the modification of a person. If this is correct, for a to believe that p, a need not stand in a relation to an object (a proposition) which would involve standing in a relation to components of that object (concepts). This blocks the obvious rationale for making thought dependent on concept possession, namely that concepts are the building blocks of thoughts. Davidson’s approach to human belief is ‘holophrastic’: we ascribe beliefs to linguistic creatures on the basis of their assenting to sentences as a whole (ITI 4, 22, 220–5; STL 25). The possibility he ignores is that rejecting the

Thought and language

building-block picture in the linguistic case invites an analogous move in the case of animals, a holodoxastic approach that starts out from the whole belief. According to such an approach, what matters is precisely a ‘modification’: if a creature can be correct or mistaken as to how things are, it can have beliefs. Although the sentences we use in ascribing thoughts have components, our ascriptions are not based on a prior ascription of these components. Instead, it is based on the subject manifesting certain perceptual capacities, attitudes and emotions. In the non-linguistic case, these manifestations will obviously not include assent to sentences. But they will include forms of behaviour, postures and facial expressions which higher animals share with human beings. When we say that Malcolm’s dog believes that the cat went up that oak tree, we do not do so on the grounds that it picks out objects and classifies them in a way that corresponds to the singular and general terms we use in the attribution – that is why it may be more accurate to rephrase (1) in the transparent manner of (1
). Rather, we simply note the dog’s reaction to its environment. We regard these reactions as directed towards particular objects, creatures and events, because we know that dogs have certain perceptual capacities, wants and dislikes. Because of its reliance on behavioural reactions, the holodoxastic move is confined to simple beliefs, notably about perceptible features of the subject’s environment. But it suffices to blunt the force of the line ‘No thoughts without concepts!’ It may seem that a problem remains nonetheless. Granting thoughts to animals while denying them concepts suggests that there is an incongruity between ascribing thoughts to animals and ascribing thoughts to linguistic creatures. In the second case, our ascriptions impute to the believer a grasp of the concepts involved, whereas in the first they do not. This creates a pressure for holding that intensional verbs like ‘believes’ are ambiguous, referring either to a holodoxastic, behavioural phenomenon or to a conceptual, linguistic one. In this vein, Malcolm suggests that while the dog can ‘believe’ that the cat went up the oak tree, only humans can ‘have the thought’ that it went up the oak tree. Similarly, Dummett maintains that while humans can have thoughts consisting of concepts, animals have mere ‘protothoughts’ consisting of spatial representations. But this type of distinction seems to count against ascribing one and the same belief to humans and animals. It suggests that ‘Both Sarah and the dog believe that p’ is not so much a falsehood as a zeugma. For ‘Sarah believes that p’ comes out as ‘Sarah has the

275

276

Quine and Davidson on Language, Thought and Reality

thought that p’ while ‘The dog believes that p’ comes out as ‘The dog has the protothought that p’. However, we can grant that there are important differences between the beliefs of conceptual and non-conceptual creatures, yet resist the pressure towards postulating distinct objects and hence distinct attitudes. As Quine and Davidson recognize, a certain disparity between the terms used in a belief report and those that could be used by the subject is present even in the linguistic case, without constituting a fundamental incongruity. The terms which occur in the content clause are in general dictated not so much by the creature whose belief we report, but by the concerns of speaker and audience. Thus, ‘Sarah thinks that the awful charlatan you introduced me to is about to give her a biscuit’ can be in order, whether Sarah is an adult, a child that lacks the concept of a charlatan, or a dog (Rundle 1997: 83). By Davidson’s own lights, therefore, the attribution of beliefs does not require that of concepts. Still, the question whether animals can possess concepts remains relevant, for two reasons. When it comes to attributing beliefs to animals, some terms appear inapposite, which suggests that animals can be credited with some concepts but not others. Furthermore, if non-linguistic creatures can have conceptual capacities, the lingualist argument from concepts fails even if it is right to tie beliefs to concepts. In line with Kant, Frege and Dummett, Davidson holds that animals can perceive, but lack concepts of any kind. At the other end of the spectrum are mentalists who have no qualms about ascribing complex concepts to animals. An intermediate position is occupied, for example, by Kenny, who maintains that animals can possess some concepts, namely those that can be manifested in non-linguistic behaviour (1989: 36–7). Proponents of this position have to concede that the concepts animals can have need not be the ones we use in ascribing thoughts to them. The discriminations which underlie animal behaviour may not coincide either extensionally or intensionally with our verbal classifications. The dog might group cats together with hamsters or distinguish black cats from all others; and even if it groups all and only cats together, it might recognize them by smell rather than visually. But this by itself is no obstacle to ascribing to animals concepts that differ from ours. For example, when chimpanzees distinguish foodstuffs and tools, the operative difference seems to be simply that between the edible and the inedible (Savage-Rumbaugh 1986: 87). Accordingly, what kind of concepts we should ascribe to animals depends on the parameters governing their

Thought and language

discriminatory behaviour. Such considerations are likely to indicate that our ordinary ascriptions require qualification, but not that they involve the kind of convenient pretence Davidson diagnoses. Whether this criticism holds water naturally depends on what one makes of concepts and concept-possession. One construal is that concepts are principles of discrimination, and that to possess a concept is to have the ability to recognize or discriminate different types of things (Price 1953: 355). On that account, animals certainly possess concepts: it is evident from their behaviour in the wild and in the laboratory that they can distinguish between a host of different colours, tastes, sounds, shapes, stuffs, types of creatures, etc. Moreover, many of these capacities are learnt rather than innate. Davidson admits this, but nevertheless resists the conclusion that animals can have concepts. He adduces several arguments to this effect. The first is a reductio ad absurdum. ‘Unless we want to attribute concepts to butterflies and olive trees, we should not count mere ability to discriminate between red and green or moist and dry as having a concept, not even if such selective behaviour is learned’ (STL 25). I agree that it would be absurd to credit olive trees with concepts. But I do not accept that this absurdity follows from treating concepts as powers of discrimination. Olive trees do not discriminate between moist and dry soil, since discrimination is a prerogative of sentient creatures. We must distinguish between mere differential reaction to causal inputs, which is a universal feature of physical phenomena, and discrimination, which is tied to creatures with perceptual capacities. Davidson’s second argument is that there is a fundamental difference between classification and discrimination: the former is required for conceptpossession, but only the latter is available to non-linguistic creatures. ‘To have a concept is to classify objects or properties or events or situations’, or, more accurately, to be able to do so. Powers of discrimination are mere ‘dispositions’, and therefore, ‘as Wittgenstein emphasized, have no normative force’. Such dispositions do not involve the ability to recognize a mistake, and hence no knowledge of the difference between correct and incorrect behaviour (STL 24–5; see SIO 104–5; RH 208). Davidson is right to hold that there is a type of classification which differs from mere discrimination in its normative dimension. He is also right to suggest that it is the absence of such classification which makes us reluctant to credit butterflies with concepts. Finally, he is right to maintain that the normativity required for such classification presupposes that the

277

278

Quine and Davidson on Language, Thought and Reality

classifier can make a mistake which she is capable of recognizing as such. To be capable of classifying or misclassifying, a creature a must not just have a disposition to behave in accordance with a rule – as butterflies do when they land only on red petals – but of following a rule. That is to say, the principle which distinguishes Fs from non-Fs must be part of a’s reason for differentiating between Fs and non-Fs, not just a law to which its differential behaviour conforms. If a can classify things into those which are F and those which are not, it must be possible that a should be mistaken, for instance in taking something as being F which in fact is not. But a can be accused of confusing Fs and non-Fs only if a is also capable in principle of recognizing the confusion. Only given that possibility can a be said to diverge from a rule which it was trying to follow, i.e. to have acted contrary to its own intentions. Otherwise, a is merely diverging from our expectations or a statistical norm. As Davidson points out, a slippery road may be a danger or a nuisance, but it does not commit a mistake. Mutatis mutandis, a butterfly that fails to discriminate between red and green may reduce its biological fitness, but it does not violate a principle to which it has committed itself. However, why should such normative behaviour be the prerogative of linguistic creatures? This is where Davidson’s third argument comes in. In line with his assimilation of mice to missiles, he contends that animals and children cannot be genuinely taught, but only causally manipulated. Because they are sentient, their behaviour can be altered by means of inflicting pleasure or pain. But the point remains: we improve the road, from our point of view, by spreading sand or salt; we improve the child, from our point of view, by causing pleasure or pain. In neither case does this process, by itself, teach road or child the distinction between correct and incorrect behaviour. To correct behaviour is not, in itself, to teach that the behaviour is incorrect. Toilet training a child is like fixing a bathtub so it will not overflow; neither apparatus nor organism masters a concept in the process. (STL 25)

The argument rightly assumes that only intentional discriminations can be corrected in the relevant sense, because only intentional behaviour can be accused of misapplying a principle of classification. It also intimates, again rightly, that such behaviour must be voluntary in the sense that the agent could have done otherwise. Mechanical behaviour – whether unconditioned or conditioned – cannot be accused of failing to live up to a principle, roughly because ought implies can. This is why

Thought and language

classification is not the exercise of a mere disposition to react differently to distinct external influences, even if, as with animals, these influences are stimuli perceived by a sentient creature. Rather, classification is the exercise of an ability. Unlike dispositions, abilities are not automatically exercised under specified conditions; the agent can intentionally exercise or refrain from exercising them (Kenny 1989: ch. 5). Nevertheless, Davidson is arguably wrong to hold that non-linguistic creatures have only dispositions and lack abilities. The behaviour of nonlinguistic creatures is not always explicable solely by reference to immediate biological imperatives. Both pre-linguistic infants and the great apes are capable of voluntary action, because they can refrain from a particular action, either by pursuing their goals in a different way or by forsaking them, at least temporarily. By the same token, in a particular situation they are capable of either heeding or disregarding a difference. Unlike bathtubs, some animals are capable of distinguishing objects of types F and G in one situation, and of ignoring the difference in another. It would be wrong to ascribe such classification to butterflies. But it seems equally wrong to deny classification to chimpanzees capable of selecting or making tools in advance of attempting the task. For these creatures deliberately distinguish between different kinds of objects (e.g. leaves and blades of grass) in some situations, but may disregard the difference in others, or if they are not in the mood (Byrne 1995: ch. 7). And here the question of whether a discrimination has been learnt is important, pace Davidson. An unconditioned reflex cannot be the exercise of an ability, but a learnt pattern of response can, because it is not necessarily automatic. This response may confine non-linguistic concept-possession, in the normative sense invoked by Davidson, to infants and the great apes. However, it does not make concept-possession dependent on language, but on discriminatory behaviour that is sufficiently complex and flexible. Once more, Davidson could not reply that the appearance of flexible behaviour among animals is deceptive, without inviting the same challenge concerning humans. He could, however, argue as follows (see Rundle 1997: ch. 4). Even if animals are capable of acting voluntarily, in the sense of doing otherwise, and of acting intentionally, in the sense of acting for a purpose, they are incapable of acting intentionally in the stronger sense of acting for a reason. We explain the behaviour of animals by reference to reasons (e.g. ‘the dog runs to the oak tree because he wants to catch the cat’). But in doing so we indicate only what their purposes or goals are, not how they have reasoned,

279

280

Quine and Davidson on Language, Thought and Reality

i.e. what their justification is for acting as they do. For that would presuppose that they are in principle capable of stating such reasons. If this is correct, we can rule out animal concepts by ruling out animal reasoning. Animals might discriminate for a purpose (e.g. to reap certain rewards), but they cannot reason. Even though their discriminations may be voluntary, they do not follow rules: they do not distinguish Fs from non-Fs for the reason that Fs possess certain distinctive features. Is this line of argument sound? Can a creature only act for a reason if it is capable of communicating this reason? Take a chimpanzee that has learnt to use different tools in the pursuit of dorylus ants and macrotermes termites. It is plausible to maintain that its reason for matching tool and prey is that they possess certain features. This impression is strengthened by the fact that chimpanzees display non-linguistic forms of behaviour that go together with the correction of error among humans, such as hesitation, displeasure, discarding one type of tool in favour of another, etc. More generally, it is far from obvious that animals are incapable of reasoning. Chimpanzees seem to do just that in their construction and employment of tools in advance of feedback from the task itself. Finally, consider the story of Chrysippus’ hunting-dog. In chasing a prey of which it has lost the scent, this dog reaches a crossroads; it sniffs down the first path, then sniffs down the second path, then it immediately follows the third without sniffing. In the case of dogs, perhaps such behaviour could only be a rigid conditioned reflex. But I can see no reason for denying that this is an intelligible form of behaviour for a non-linguistic creature capable of voluntary action. And if it is, what is wrong with the explanation that the behaviour evinces a disjunctive inference (‘p or q or r; neither p nor q; ergo r’)? We might grant that there is a difficulty in describing such a creature as silently consulting a principle. But as Ryle has convincingly argued, even the intelligent performances of humans are rarely accompanied by conscious consultations of this kind. However, although humans need not actually verbalize their reasoning, they are capable of doing so. In the absence of this capacity, the question arises of what in an animal’s behaviour could correspond to the ‘ergo’ of linguistic reasoning. This point is unanswerable with respect to creatures like dogs. But in the case of chimpanzees there can be an analogue, however thin, to our ‘ergo’. In the context of encountering and pondering a problem, certain gestures and grimaces, followed by renewed activity, can naturally be interpreted as marking the point when the shilling dropped. Even if this is an anthropomorphic interpretation in

Thought and language

the case of chimpanzees, we can easily imagine a non-linguistic hominoid whose facial expressions and gestures are so close to ours as to make such a description inevitable. Furthermore, even without the ergo, as regards context (problem-solving), demeanour (e.g. head-scratching) and result (problem-solution), the deliberations of chimpanzees are close enough to those of humans to qualify as instrumental reasoning. Consequently, there is no compelling case for denying that animals can possess concepts. Moreover, even if there is, it does not suffice to deny them thoughts, because of the possibility of holodoxastic belief. Davidson is right to insist that ‘concept-formation is not a way station between mere dispositions . . . and judgements’ (STL 25). Concepts and judgements remain on a par. A chimpanzee capable of classifying things into sticks and knives, and hence of concepts like ‘stick’, can also believe that the object it confronts is a knife, or wish that it were a stick. My point is rather that it has yet to be shown either that concept-formation and judgement require linguistic judgement or that holodoxastic belief cannot be a way station between dispositions and judgement involving concepts.

3

The holistic nature of thought

Davidson’s last argument against animal concepts is that the concepts which feature in the beliefs we commonly ascribe to animals require certain ‘general beliefs’ with which we cannot credit them (SIO 98). This is part of a wider qualm, namely that attributing thoughts to animals is incompatible with ‘the intrinsically holistic character of the propositional attitudes’, the alleged fact that ‘to have one is to have a full complement’ (SIO 96). Since at least some members of that complement are definitely beyond their pale, animals cannot even have the simple beliefs commonly ascribed to them. We identify thoughts, distinguish among them, describe them for what they are, only as they can be located within a dense network of related beliefs. If we really can intelligibly ascribe single beliefs to a dog, we must be able to imagine how we would decide whether the dog has many other beliefs of the kind necessary for making sense of the first. It seems to me that no matter where we start, we very soon come to beliefs such that we have no idea at all how to tell whether a dog has them, and yet such that, without them, our confident first attribution looks shaky. (SIO 98)

281

282

Quine and Davidson on Language, Thought and Reality

According to Davidson, there are three types of beliefs ‘on which any particular thought depends’: particular beliefs (e.g. that the cat seen running a moment ago is still in the neighbourhood), general but empirical beliefs (e.g. that cats can scratch or climb trees), and logical beliefs. The first type does not cast novel doubts on animal thought. Having recognized his original error, Malcolm’s dog might engage in searching behaviour that indicates his belief that the cat is still in the vicinity. The other two types give rise to three new objections. One is that beliefs must display a degree of rationality which requires a large web of consistent thoughts. The other concerns the identity conditions of thought. The logical relations between thoughts are partly constitutive of their identity, because the content of a thought cannot be divorced from what it entails and what is entailed by it. The thought that p could not lack the logical connection with the thoughts it entails ‘without becoming a different thought’ (SIO 98–9, see 123–5; DD 232). Accordingly, if p entails q, and if a believed that p without believing that q, a would be at least partly ignorant of the content of its own belief. Such ignorance seems incompatible with the fact that believing that p is not simply a mental occurrence, but something with a specifiable content. To the extent to which a is ignorant of the content of his alleged thought, a’s thought cannot be the same as that of b, who is cognizant of that content. Ergo, if b’s belief is that p, a’s belief cannot be that p. The claim that thoughts are individuated through their content and hence through their logical connections is correct in principle. But does it hold for all types of thought and without qualification? To tackle this question one must do what neither proponents nor opponents of holism have done so far, namely spell out and assess various holistic principles. I shall argue that these principles are either too strong for Davidson’s case, because they preclude plausible cases of human thought, or too weak, because they include some forms of animal thought. The strongest holistic principle lingualists could invoke runs as follows: (A) (aBp & (p ⇒ q)) ⇒aBq

This principle is excessively restrictive. Human beings can believe, for example, the axioms of Euclidean geometry without believing all the theorems entailed by them. At this point, Davidson might claim that to ‘intelligibly ascribe’ (SIO 98) a thought, a need not actually believe that q, but

Thought and language

only be capable of believing (learning, understanding) that q: (B) (aBp & (p ⇒ q)) ⇒ aBq.

Even that principle can be challenged, however. Why shouldn’t there be people who believe the Euclidean axioms without even being capable of learning the theorems? One must not be too bloody-minded about the identity conditions of thoughts. Grasping a thought is not an all-or-nothing affair. Why should we have to refrain from using the same that-clause as soon as there is even a single consequence of the thought that p which a but not b is capable of understanding? After all, many humans have a habit (nasty according to some, endearing according to others) of rejecting at least some of the logical consequences of their beliefs, even if these are pointed out to them. However, Davidson need not subscribe to any general closure principle, not even a modally qualified one like (B). He states that ‘there is no fixed list of things someone with the concept of a tree must believe’ (SIO 98). This may simply mean that there is no fixed list of things entailed by thoughts about trees; but it may also mean that among these entailments there is no fixed list which a must be capable of appreciating, and hence that a need be capable only of appreciating some of the things entailed by p. Schematically, this would amount to something like this: (C) aBp ⇒ ∃q((p ⇒ q) & aBq)

According to this principle, if a human being is incapable of even understanding any of the theorems entailed by Euclid’s axioms, his beliefs in the axioms do not have the same content as the belief of a human being who is capable of understanding these theorems; and to that extent, the two have different beliefs. At this juncture, an opponent of lingualism is left with two replies. The first is to reject (C). Thus it might be claimed that logical relations are neither the only nor the most basic feature by which we identify a belief. After all, we have no better way of describing illogical humans than this: they are unable to grasp the consequences of their beliefs. But that very formulation presupposes that we can distinguish between having a belief and being able to grasp its consequences. Simple perceptual beliefs are ascribed even to humans primarily on the basis of behavioural responses. When Mowgli flees from Shera Khan we are not worried about what consequences of ‘A tiger is chasing me’ he is capable of appreciating.

283

284

Quine and Davidson on Language, Thought and Reality

But perhaps we should be worried if Mowgli were incapable of grasping any of these consequences. For this reason, it is best to accept (C), while maintaining that there are animals capable of appreciating some consequences of simple perceptual beliefs. For example, by consistently barking up the oak and completely ignoring the pine even when prompted by us to do otherwise, Malcolm’s dog could display the belief that the cat is not on the pine tree. Less contentiously, if non-linguistic creatures are in principle capable of reasoning and if chimpanzees do in fact reason, as argued in section 2, then such creatures are capable of appreciating certain consequences of things they have learned about their environment. This point is connected with the holistic argument about rationality. Davidson maintains that having even a single thought is to have a ‘largely correct logic’ (SIO 99), i.e. a whole pattern of thoughts that cohere logically and do not display radical irrationality. This argument is compatible with animal thought. The great apes can be rational not just in the sense of behaving in accordance with evolutionary design or their own wants, but also in the sense of acting according to what they have learnt with respect to a given task. What’s more, like humans they can also, on occasion, be irrational in this sense and improve their performance as a result of trial and error or instruction. At this point, the lingualist might continue his attack by maintaining that thought requires not just the capacities to act rationally and to have certain beliefs entailed by an original belief, but the capacity to have beliefs in the entailments. This argument is not in Davidson, nor is it plausible, since it would also rule out children and many adults. It is unsuccessful even if the holistic requirement is restricted both modally and in scope, as in (D) aBp ⇒ ∃q((p ⇒ q) & aB(p ⇒ q))

On one understanding of (D), some animals qualify. They can learn that p entails q, in the sense of consistently reasoning from p to q in solving cognitive tasks. What they cannot display is an awareness of the difference between an empirical and a logical consequence. It makes no sense to wonder whether Chrysippus’ dog acts on an inductive generalization (‘Whenever p or q and not-p, it turns out that q’) or on a logical inference (‘p or q; not p; ergo q’). But we ascribe disjunctive reasoning to human beings irrespective of whether they are capable of recognizing this difference; indeed, Quine is committed to the idea that ultimately this difference is more apparent than real.

Thought and language

This leaves a final argument, from conceptual holism. That argument does not threaten the possibility of holodoxastic belief. To assess its consequences for conceptual belief, one should distinguish between two types of general beliefs, namely analytic and empirical. Davidson insists that beliefs presuppose certain general empirical beliefs on account of their constituent concepts. This is implausible for many of his examples (SIO 98–9; ITI 200). It would seem that one can believe that the cat went up the oak tree without knowing that trees burn or that one can believe that a cloud is passing before the sun without possessing the meteorological knowledge that clouds are made of water vapour. Such a radical conceptual holism also creates a general problem, namely that any empirical discovery amounts to a conceptual change, with the possible consequence that scientific theories talk about different things as soon as some of their empirical claims are at odds (Fodor and LePore 1992: ch. 2). Davidson seems on firmer ground when he invokes general beliefs which are clearly analytic, e.g. that cats are animals or physical objects that move in certain ways. Animals cannot recognize the analytic status of such beliefs, nor do they need to in order to have thoughts. But as Davidson himself recognizes, they can ‘generalize’ in the sense of reacting similarly to similar stimuli (SIO 104–5). By this token, they can also in principle distinguish not just between, e.g. mice and cats, but between animals, plants and inanimate objects. The notion of an enduring physical object is more problematic, since it is in many respects a result of philosophical reflection that exceeds the requirements of ordinary human thought. But if it is spelled out in a pedestrian way, it is clear that chimpanzees can learn to distinguish, for example, between physical objects on the one hand, mirror reflections or TV images on the other. Finally, we must remember that a creature could have concepts without having our concepts. This means that there is no holistic argument against animal concepts. More generally, holism does not provide a compelling objection against the possibility of animal thought. There are plausible holistic principles, notably (C), which exclude the possibility of a creature having just a single belief. In any event, the complexity required for conceptual belief is incompatible with a behavioural repertoire capable of exhibiting just a single belief. But these reflections do not establish that the web of which any belief must be part need extend as far as the web of sophisticated human thought. There may be larger and smaller webs. What kind of web is required may depend on the belief and the creature concerned. From

285

286

Quine and Davidson on Language, Thought and Reality

the fact that an animal lacks our web of beliefs and our concepts, it does not follow that it has no beliefs and no concepts. 4

Belief and the concept of belief

By Davidson’s own admission, the considerations mentioned so far provide no compelling argument for his thesis that beliefs require the capacity for language. The line of thought he himself favours proceeds in two steps (SIO 102): I. to have a belief, one must have the concept of belief; II. to have the concept of belief one must have language.

Davidson admits that his argument can be challenged at several points. In my view, however, matters are worse: (I) is mistaken, and although (II) is correct, it cannot be defended the way Davidson does. Davidson’s argument for (I) revolves around the following statement: ‘Someone cannot have a belief unless he understands the possibility of being mistaken, and this requires grasping the contrast between truth and error – true belief and false belief’ (ITI 170). His reasoning can be reconstructed as proceeding along the following steps: i. ii. iii. iv.

A belief is something that ‘can be true or false’ (SIO 104). aBp ⇒ a can be mistaken in believing that p; aBp ⇒ a can recognize that he is mistaken in believing that p; aBp ⇒ a has the concept of a mistake.

(iv) entails (I), since the concept of a mistake at issue is that of a mistaken belief. (i) is incontrovertible, and (ii) holds at least for standard empirical beliefs. Troubles start with (iii). Davidson supports it by holding that one cannot have beliefs without also having ‘reflective thoughts’ (SIO 103), beliefs about one’s own beliefs, which in turn presupposes the concept of a belief. Davidson is not claiming that all beliefs are self-conscious, i.e. that one can only believe that p if one also believes that one believes that p. But he maintains that one cannot have beliefs without having some beliefs about one’s beliefs. The reason is that having a belief entails the possibility of surprise. If I believe that there is a coin in my pocket, it must be possible that something should happen that would change my mind. Finally, it is not enough that I should first believe that there is a coin in my pocket and then no longer have this belief after emptying my pockets:

Thought and language

Surprise requires that I be aware of a contrast between what I did believe and what I come to believe. Such awareness, however, is a belief about a belief: if I am surprised, then among other things I come to believe that my original belief was false. (SIO 104)

In one respect, Davidson’s focus on surprise is promising. The vast majority of our beliefs are implicit expectations of which we become aware only when they are disappointed. But Davidson’s conception of surprise is unusual. We do not ordinarily speak of surprise whenever somebody realizes that a prior belief was wrong. If I find that there is no coin in my pocket after all, I may simply shrug my shoulders, mindful of the fact that I am a notorious scatterbrain. In ordinary parlance, ‘surprise’ is the name of a particular kind of reaction to things being otherwise than one expected or believed (or of the fact or event which produces that reaction). Such reactions are displayed not just linguistically, but in forms of behaviour and facial expressions which are not the prerogative of language-users. We would not hesitate to speak of a chimpanzee as being surprised by finding that what looked like a banana is merely a decoy, provided that its behaviour or facial expressions show signs of disorientation or disappointment (DeGrazia 1994: 149). Davidson approaches matters from a different angle. He distinguishes between being startled and being surprised. Perhaps he has in mind the difference between being surprised by an object or event and being surprised that things are thus-and-so. In any event, he seems to conceive of being surprised independently of any specific behavioural manifestations, and as involving beliefs about beliefs by definition. Given this stipulation, the soundness of (iii) hinges straightforwardly on the question of whether a creature can only have beliefs if it is capable of also believing that a prior belief was false. One should grant that a creature can believe that p only if it is also capable of believing something that is incompatible with that belief, notably that not-p. But Davidson has no argument to rule out the possibility of a simply switching from a belief that p to a belief that q, without that switch involving a believing that its original belief was false. For example, Malcolm’s dog first believes that the cat is in the oak tree, as witnessed by its barking up the oak, and then that it went up the pine tree, as witnessed by its barking up the pine. Even the step from (iii) to (iv) is more problematic than might appear. We can distinguish between cases in which an animal simply acquires a new belief and cases in which it corrects a previously held belief. The latter are

287

288

Quine and Davidson on Language, Thought and Reality

marked by behavioural reactions and facial expressions which show that the animal considers a previous course of action to be mistaken. For example, after barking up the oak tree for a while, Malcolm’s dog shows signs of disappointment and frustration; it then turns round and looks for the cat somewhere else; finally, on spotting the cat, it starts barking up the pine tree with renewed vigour. It has not simply spotted the cat on the pine tree, but recognized that the cat went up the pine instead of the oak. However, that is not necessarily to recognize that its prior belief that it went up the oak tree was mistaken, even though the fact that the cat did not go up the oak tree does entail that that prior belief was false. As we have just seen, one can believe that p, in our case, that the cat did not go up the oak tree, without necessarily believing all the things that are entailed by that belief. Accordingly, even if a is capable of recognizing a mistake, and, in that sense, of understanding the possibility of being mistaken, it remains to be shown why a needs to have the concept of a mistake, as Davidson has it. What a must be able to recognize is that an object x which it initially took to be F, is not F after all. However, there is as yet no need for a’s recognition to display a grasp of the general concept of a mistake, a concept that covers not just a’s own misapplication of the specific concept F, but also anybody else’s misapplication of any other concept, including concepts that a lacks. For a chimpanzee to recognize that it is dealing with dorylus rather than macrotermes, it does not need to recognize that its own previous belief falls under the same concept – that of a mistake – which also applies to all other mistakes, e.g. Columbus’ belief that he had reached India. Neither does it need to recognize that its previous classification falls under the same concept – that of misclassification – as that of someone who classifies whales as fish. If I am right, belief does not require the concept of belief. But does the latter require language? Davidson insists that the concept of a belief is the concept of something which can be true or false, correct or incorrect. From this he rightly infers that to have the concept of belief is to have the concept of ‘objective truth’, the idea that there is a difference between my believing that p and it being true or the case that p, or between ‘belief and truth’. To grasp the concept of objective truth is also to grasp the ‘subjective– objective contrast’ and the concept of an ‘objective reality’ (SIO 104; ITI 169–70; STL 26–7). Davidson goes on to claim that complex behaviour, including the ability to learn or generalize about the environment, is no guarantee for grasping the contrast between ‘belief and truth’. What does suffice is

Thought and language

communication. To communicate with someone else, I need not agree with her in all matters, but I must ‘share her world’, that is, entertain the same propositions with ‘the same subject-matter, and the same concept of truth’ (SIO 105). But the concept of intersubjective truth and of an intersubjective world is the concept of an objective world about which different communicators can have beliefs. Davidson’s claim that communication suffices for the notion of truth is problematic, since he also suggests that the latter requires the notion of error. Linguistic communication is possible with young or autistic children that cannot ascribe errors to others and hence lack the concept of error (Wimmer and Perner 1983). At the same time, he is correct to hold that the concept of intersubjective truth is sufficient for the concept of belief: an intersubjective truth can be understood as one that all rational creatures have reason to believe. However, the crucial question is whether the concept of intersubjective truth, and with it the notion of communication, is necessary for the notion of objective truth and hence for the concept of belief. Davidson claims it is, because ‘the only way one could come to have the belief-truth contrast is through having the concept of intersubjective truth’ (SIO 105). He admits that he does not have a compelling argument for confining the concept of objective truth to communicators. Instead he presents his opponent with a challenge to show how else ‘one could arrive at the concept of an objective truth’. He also appeals to triangulation. I would have no way of determining the spatial distance between me and many objects without triangulation, i.e. changing my position with respect to them. Grasping the notion of objective truth depends on a different sort of triangle, namely one involving two creatures. Although each of them interacts with an object, what provides them with ‘the concept of the way things are objectively is the base line formed between the creatures by language’, i.e. the fact that they share a concept of truth. In recent papers, Davidson uses triangulation no longer merely as an analogy but as a genetic model for ‘the initial phase of ostensive learning’ in both ontogenesis and phylogenesis. Two creatures a and b not only react similarly to similar stimuli, but also notice the fact that they react similarly. In this way they set up a triangle, the corners being a, b, ‘and the objects, events or situations to which they mutually respond’. The notion of an error arises if the correlation of reactions to the third corner of the triangle breaks down. a notices that b no longer reacts to a situation in the previously shared way. It can therefore judge that b has erred. But in order

289

290

Quine and Davidson on Language, Thought and Reality

to grasp the concept of truth, a needs to take a further step, namely ‘to communicate the contents – the propositional contents of the shared experience, and this requires language’ (STL 27). This is supposed to show that a can have the concept of belief, and hence, according to Davidson, beliefs only if it communicates with others by way of language. It is equally supposed to show, however, that a can have a language only if it communicates with others. For just as thought presupposes language, language presupposes thought (ITI 155). Accordingly, triangulation provides an argument against the possibility of a private language. Language is essentially social not in that communication requires a shared language, but in that a creature can have a language only if it actually communicates with others. Language, thought and rationality are all confined to communicators (SIO 100, 117–21). Davidson is right to hold that triangulation makes the notion of an error intelligible. He is also right in holding that we pick up the notion of objective truth through communication: the distinction between believing that p and it being true that p is learnt through linguistic interaction. But this does not provide the kind of a priori argument Davidson is after, an argument which shows that it is impossible to acquire or explain the concept of truth other than through communication. The analogy of triangulation establishes nothing of the kind. Even in the case of measuring distances, triangulation is only one of many possible ways of achieving objectivity. Moreover, the analogy shows at best that to grasp the notion of objectivity, we need to understand the possibility of observing one and the same object from different perspectives. But it remains to be shown that such an understanding presupposes a recognition of error in others. At least prima facie, different perspectives can be conceived simply as perspectives that one and the same individual can occupy. To be able to occupy these different perspectives, and to form an objective picture of the world as a result, all a creature would seem to need is the ability to move, as well as perceptual capacities that are adjusted to such movement. Finally, even if the ‘triangular’ recognition of error in others is essential, it remains mysterious why this amounts to a conception of truth only if that recognition can be communicated to the wayward individual. Perhaps there is an essential intersubjective element which my picture omits. For example, G. H. Mead may have been right to suggest that the idea of a different perspective can only be explained by reference to the idea of taking the role of the other. Unfortunately, even this would not show that a must communicate with b. Conversely, the tactical deceptions

Thought and language

practised by chimpanzees provide grounds for holding that non-linguistic creatures are capable of having beliefs about the beliefs of others, and, more specifically, that they can recognize mistaken beliefs in others (Byrne 1995: chs. 8–9). However, these grounds need not be conclusive. Consider the statement: (2) Chimpanzee a believes that chimpanzee b believes that p.

Perhaps (2) can be rephrased completely in terms of beliefs concerning what b does or is about to do, rather than what b believes. Admittedly, (2) may be in order without b doing anything just yet. But in the case of linguistic creatures we can distinguish (2
) a believes that b will 

from (2∗ ) a believes that b will  because b believes that p.

The challenge is to show how that distinction could be drawn in the nonlinguistic case. In my view, the challenge can be met. (2∗ ) rather than (2
) is appropriate, for example, in cases in which a displays behaviour anticipating b’s ing only in situations in which a believes that it can deceive b into thinking that p. Such a description seems appropriate, e.g. in the welldocumented cases of chimpanzees using a ‘hide-and-peek’ strategy to unmask the deception of a lower-rank individual that deliberately withholds its gaze from a source of food. Be that as it may, Davidson’s triangulation argument fails to show that having the concept of truth requires being a linguistic communicator. There may be another argument to this effect. One might insist that second-order beliefs as such do not guarantee possession of the concept of truth. One must also understand of beliefs in general what it is for them to be true and what it is for them to be false. This presupposes, for example, a grasp of the equivalence between ‘It is true that p’ and ‘p’ or ‘It is a fact that p’. Such a grasp is confined to creatures that are capable of manifesting such conceptual operations; and it is difficult to see how this could include non-linguistic creatures. This may furnish a defence of (II), the claim that the concept of belief requires language. But it does not reinstate Davidson’s favoured argument for lingualism, given the failure of (I) – the claim that belief requires the concept of belief. Nor does it reinstate his argument for the claim that language is the prerogative of creatures that actually communicate with others.

291

292

Quine and Davidson on Language, Thought and Reality

According to Davidson, intensionality, concepts and holism ‘point in the direction of language, but they do not amount to a demonstration that language is necessary to thought. Indeed, what these considerations suggest is only that there probably can’t be much thought without language’ (SIO 101). This verdict is puzzling. Davidson’s radical holism according to which to have one thought is ‘to have a full complement’ is incompatible with the conclusion of merely restricting the scope of animal thoughts in the direction of which it is supposed to point. By contrast, my modest holism according to which thoughts come in larger or smaller packages favours such a restriction, since packages that include beliefs manifestable only in linguistic behaviour are the preserve of language-users. Elsewhere (ITI 163–4), Davidson expresses a separate reservation about the argument from intensionality, namely that it merely shows our attributions of thoughts to animals to be ‘seriously underdetermined’. That gloss is at odds with his simultaneous observation that we ‘cannot make sense’ of the question whether the dog believes that the president of the bank is at home. In my view, the observation is correct and the reservation problematic. In so far as attributions of thought are seriously underdetermined, they are vacuous. But in the case of simple thoughts, non-verbal behaviour suffices for distinguishing a creature believing one thing from it believing something else. If this is on the right track, the problem with the arguments concerning intensionality, concepts and holism is not that they are uncompelling, but that all they compel us to is confining animal thought to simple cases. Furthermore, our discussion lends support to a heretical idea. Pet lovers and zoologists attribute a greater variety of thoughts to animals not because they suffer from anthropomorphism, as philosophers are fond of intimating, but because they are better acquainted with their behaviour and capacities. At the same time, Davidson’s reflections point to ways in which attributions of even simple thoughts to animals are not mere extrapolations from the human case. Animals do not just have thoughts of a simpler kind, their having these thoughts amounts to something simpler, because it is part of a smaller logical space. In the case of animals, there is at most an analogue of intensionality. In so far as thought-ascriptions to animals are holodoxastic, they are not only restricted to thoughts about perceptible features of the environment, they also lack conceptual connections which apply in the human case: we cannot infer from the fact that the dog thinks that something is F that the dog grasps the concept F. Furthermore, even if

Thought and language

animals can have concepts, these are not just confined to concepts of a perceptual kind. Animals are also incapable of satisfying one of the two criteria which we standardly use in attributing concepts. They may be able to apply principles of classification, but not to explain them. In fact, the two restrictions are linked. A chimpanzee may discriminate between its keeper and other humans just as deliberately as it does between red and black ants. But we are more inclined to ascribe to it the concept of redness than the concept of a keeper, because there is so much more to explain with respect to the latter. Finally, Davidson’s holism indicates that even those thoughts animals might be credited with lack the kind of context which characterizes sophisticated linguistic thought. Accordingly, attributions of simple thoughts to animals are neither intensional, nor conceptual, nor holistic in the way thought-attributions to humans are. However, the best analogy is not the anthropomorphic explanation of missiles, but one Davidson has suggested in discussion. Attributing thoughts to animals is like using numerals for the purpose of labelling members of a football team. Although natural numbers stand in complex relations of order and numerical difference, these relations are ignored in this context. What matters here is not the numerical difference between two numbers, nor even which one is greater, but only that no two numbers should be used for the same player. The analogy is illuminating. Thought-attributions to animals employ a rich conceptual apparatus to an area in which many of the logical connections constituting that apparatus do not apply. But it breaks down in one important respect. Attributing thoughts to animals is not simply an impoverished application of a rich technique. For that richer technique evolves around a central core of cases in which creatures believe, know or desire things on account of their wants and perceptual capacities. These biological basics of belief are shared by humans and animals. At the same time, when we move from this core area in the direction of conceptual thought, we also move in the direction of linguistic thought. The features which non-linguistic creatures must possess to be capable of conceptual thought – intentionality, complexity, flexibility – correspond to those features by which theorists from Descartes to Chomsky have distinguished language from more basic systems of communication. In this respect, at least, our reflections tend to confirm rather than to negate the connection between thought and language.

293

References

Alston, W. P. 1958 ‘Ontological Commitments’, Philosophical Studies 9, 8–17. 1996 A Realist Conception of Truth (Ithaca: Cornell University Press). Alvarez, M. 1994 ‘Radical Interpretation and Semantic Nihilism: Reply to Glock’, Philosophical Quarterly 44, 354–60. Arrington, R. L. 1996 ‘Ontological Commitment’, in Arrington and Glock (eds.), 196–211. Arrington, R. L. and H. J. Glock (eds.) 1996 Wittgenstein and Quine (London: Routledge). Austin, J. L. 1962 Sense and Sensibilia (Oxford University Press). 1970 Philosophical Papers (Oxford University Press). Ayer, A. J. 1971 Language, Truth and Logic (Harmondsworth: Penguin). Baker, G. P. and P. M. S. Hacker 1984 Language, Sense and Nonsense (Oxford: Blackwell). Barrett, R. and R. Gibson (eds.) 1990 Perspectives on Quine (Oxford: Blackwell). Bealer, G. 1998 ‘Analyticity’, in E. Craig (ed.), 234–9. Beckermann, A. 2001 Analytische Einfuhrung in die Philosophie des Geistes (Berlin: de Gruyter). ¨ Beldoyen, R. van 1984 Characteristics and Recognizability of Vocal Expressions of Emotions (Dordrecht: Foris). Bennett, J. 1976 Linguistic Behaviour (Cambridge University Press). ¨ Bergstrom, L. 1994 ‘Quine’s Truth’, Inquiry 37, 421–35. Bilgrami, A. 1993 ‘Norms and Meaning’, in Stoecker (ed.), 121–44. Bird, G. 1995 ‘Carnap and Quine: Internal and External Questions’, Erkenntnis 42, 41–64. Black, M. 1948 ‘The Semantic Definition of Truth’, Analysis 8, 49–63. Boghossian, P. 1997 ‘Analyticity’, in Hale and Wright (eds.), 331–68. Brown, H. 2001 ‘Incommensurability and Reality’, in P. Hoyningen-Heune and H. Sankey (eds.), Incommensurability and Related Matters (Dordrecht: Kluwer), 23–142. Byrne, R. 1995 The Thinking Ape (Oxford University Press). Carnap, R. 1937 The Logical Syntax of Language (London: Routledge & Kegan Paul). 1956 Meaning and Necessity (University of Chicago Press). 1963 ‘Replies and Systematic Expositions’, in P. A. Schilpp (ed.), The Philosophy of Rudolf Carnap (La Salle: Open Court), 859–1013. 1990 ‘Quine on Analyticity’, in Creath (ed.), 427–32. Carroll, L. 1895 ‘What the Tortoise Said to Achilles’, Mind 4, 278–80. Cartwright, R. L. 1954 ‘Ontology and the Theory of Meaning’, Philosophy of Science 21, 316–25.

[294]

References Chomsky, N. 1965 Aspects of a Theory of Syntax (Cambridge, Mass.: MIT Press). 1969 ‘Quine’s Empirical Assumptions’, in Davidson and Hintikka (eds.), 53–68. Chomsky, N. and J. Katz 1974 ‘What the Linguist is Talking About’, Journal of Philosophy 71, 347–67. Churchland, P. 1986 Neurophilosophy (Cambridge, Mass.: MIT Press). Craig, E. (ed.) 1998 The Routledge Encyclopedia of Philosophy (London: Routledge). Crane, T. 1994 ‘Physicalism (2): Against Physicalism’, in Guttenplan (ed.), 479–84. Creath, R. (ed.) 1990 Dear Carnap, Dear Van: The Quine— Carnap Correspondence and Related Work (Berkeley: University of California Press). Dancy, R. M. (ed.) 1983 ‘Alien Concepts’, Synthese 56, 283–300. Davidson, D. and J. Hintikka (eds.) 1969 Words and Objections (Dordrecht: Reidel). DeGrazia, D. 1994 Taking Animals Seriously (Cambridge University Press). Dilman, I. 1984 Quine on Ontology, Necessity and Experience (London: Macmillan). 1996 ‘Existence and Theory: Quine’s Conception of Reality’, in Arrington and Glock (eds.), 173–95. Dummett, M. 1978 Truth and Other Enigmas (London: Duckworth). 1981 Frege: Philosophy of Language (London: Duckworth). 1986 ‘ “A Nice Derangement of Epitaphs”: Some Comments on Davidson and Hacking’, in LePore (ed.), 459–76. 1987 ‘Reply to John McDowell’, in B. Taylor (ed.), Michael Dummett: Contributions to Philosophy (Dordrecht: Nijhoff), 253–68. 1993a Origins of Analytical Philosophy (London: Duckworth). 1993b The Seas of Language (Oxford University Press). Dupr ´e, J. 1993 The Disorder of Things (Cambridge, Mass.: Harvard University Press). Evans, G. 1985 Collected Papers (Oxford University Press). Evans, G. and J. McDowell (eds.) 1976 Truth and Meaning (Oxford: Clarendon Press). 1976 ‘Introduction’, in Evans/McDowell (eds.), vii–xxiii. Evnine, S. 1991 Donald Davidson (Cambridge: Polity Press). Feldman, R. 1998 ‘Charity, Principle of’, in Craig (ed.), 282–5. Field, H. 1972 ‘Tarski’s Theory of Truth’, Journal of Philosophy 69, 347–75. 1980 Science Without Numbers (Oxford: Blackwell). Fodor, J. and E. LePore 1992 Holism: A Shopper’s Guide (Oxford: Blackwell). Føllesdal, D. 1973 ‘Indeterminacy of Translation and Under-Determination of the Theory of Nature’, Dialectica 27, 289–301. Forbes, G. 1985 The Metaphysics of Modality (Oxford: Clarendon). Foster, J. 1976 ‘Meaning and Truth Theory’, in Evans and McDowell (eds.), 1–32. Frege, G. 1893 Grundgesetze der Arithmetik, Vol. I (Jena: Pohle). 1953 The Foundations of Arithmetic (Oxford: Blackwell). 1979 Posthumous Writings (Oxford: Blackwell). Friedman, M. 1975 ‘Physicalism and the Indeterminacy of Translation’, Nous ˆ 9, 353–74. Gibson, R. F. 1982 The Philosophy of W. V. Quine (Tampa: University Presses of Florida). 1986 ‘Translation, Physics, and Facts of the Matter’, in Hahn and Schilpp (eds.), 139–54. 1988 Enlightened Empiricism: An Examination of W. V. Quine’s Theory of Knowledge (Tampa: University Presses of Florida). 1994 ‘Quine and Davidson: Two Naturalized Epistemologists’, Inquiry 37, 449–63. Glock, H. J. 1993 ‘The Indispensability of Translation in Quine and Davidson’, Philosophical Quarterly 43, 194–209.

295

296

References 1995 ‘A Radical Interpretation of Davidson: Reply to Alvarez’, Philosophical Quarterly 45, 206–12. 1996a ‘Abusing Use’, Dialectica 50, 205–23. 1996b ‘Necessity and Normativity’, in H. Sluga and D. Stern (eds.), The Cambridge Companion to Wittgenstein (Cambridge University Press), 198–225. 1996c A Wittgenstein Dictionary (Oxford: Blackwell). 2000 ‘Wittgenstein and Quine: Mind, Language and Behaviour’, in S. Schroeder (ed.), Wittgenstein and Contemporary Philosophy of Mind (Basingstoke: Palgrave), 3–23. 2003 ‘The Linguistic Doctrine Revisited’, in A. Coliva and E. Picardi (eds.), Wittgenstein Today (Dordrecht: Kluwer). Glock, H. J. and J. M. Preston 1995 ‘Externalism and First Person Authority’, The Monist 78, 515–34. ¨ Gluer, K. 1999 Sprache und Regeln (Berlin: Akademie Verlag). Goodman, N. and W. V. Quine 1947 ‘Steps Towards a Constructive Nominalism’, Journal of Symbolic Logic 12, 97–122. Grandy, R. 1973 ‘Reference, Meaning and Belief’, Journal of Philosophy 70, 439–52. Grayling, A. 1997 Philosophical Logic (Oxford: Blackwell). Grice, H. P. 1989 Studies in the Way of Words (Cambridge, Mass.: Harvard University Press). Grice, H. P. and P. F. Strawson 1956 ‘In Defense of a Dogma’, Philosophical Review 65, 141–58. Guttenplan, S. (ed.) 1975 Mind and Language (Oxford: Clarendon Press). (ed.) 1994 A Companion to the Philosophy of Mind (Oxford: Blackwell). Haack, R. J. 1978 ‘Davidson on Learnable Languages’, Mind 87, 230–49. Haack, S. 1978 Philosophy of Logics (Cambridge University Press). Hacker, P. M. S. 1988 ‘Language, Rules and Pseudo-Rules’, Language and Communication 8, 159–72. 1996a ‘On Davidson’s Idea of a Conceptual Scheme’, Philosophical Quarterly 46, 289–307. 1996b ‘Wittgenstein and Quine: Proximity at Great Distance’, in Arrington and Glock (eds.), 1–38. 1997 ‘Davidson on First-Person Authority’, Philosophical Quarterly 47, 285–304. 1998a ‘Davidson on Intentionality and Externalism’, Philosophy 73, 539–52. 1998b ‘Davidson on the Ontology and Logical Form of Belief’, Philosophy 73, 81–96. Hacking, I. 1975 Why Does Language Matter to Philosophy? (Cambridge University Press). 1986 ‘The Parody of Conversation’, in LePore (ed.), 447–58. Hahn, L. E. (ed.) 1999 The Philosophy of Donald Davidson (Open Court: La Salle). Hahn, L. E. and P. A. Schilpp (eds.) 1986 The Philosophy of W. V. Quine (La Salle: Open Court). Hale, B. 1987 Abstract Objects (Oxford: Blackwell). Hale, B. and C. Wright (eds.) 1997 A Companion to the Philosophy of Language (Oxford: Blackwell). Hanfling, O. 2000 Philosophy and Ordinary Language (London: Routledge). Hansen Soles, D. 1999 ‘Preferring True: Archimedean Point or Achilles Heel’, in Hahn (ed.), 311–29. Harman, G. 1999 Reasoning, Meaning and Mind (Oxford: Clarendon Press). Hart, H. L. A. 1961 The Concept of Law (Oxford: Clarendon Press). Heal, J. 1978 ‘On the Phrase “Theory of Meaning”’, Mind 87, 359–75. 1989 Fact and Meaning (Oxford: Blackwell). 1997 ‘Radical Interpretation’, in Hale and Wright (eds.), 175–96.

References Hintikka, J. 1980 ‘Theories of Truth and Learnable Languages’, in S. Kanger and ¨ S. Ohmann (eds.), Philosophy and Grammar (Dordrecht: Reidel), 37–57. Hookway, C. 1988 Quine (Cambridge: Polity Press). Horgan, T. 1994 ‘Physicalism (1)’, in Guttenplan (ed.), 471–9. Horwich, P. 1998 Truth (Oxford: Blackwell). Hugly, P. and C. Sayward 1996 Intensionality and Truth (Dordrecht: Kluwer). Isaacson, D. 1992 ‘Carnap, Quine and Logical Truth’, in D. Bell and W. Vossenkuhl (eds.), Subjectivity and Science (Berlin: Akademie Verlag), 100–30. Izard, C. E. 1969 The Face of Emotion (New York: Appleton-Century-Crofts). Jackson, F. 1998 From Metaphysics to Ethics: A Defence of Conceptual Analysis (Oxford: Clarendon). James, W. 1978 Pragmatism and the Meaning of Truth (Cambridge, Mass.: Harvard University Press). Jeffrey, R. 1983 The Logic of Decision (University of Chicago Press). Katz, J. 1990 ‘The Refutation of Indeterminacy’, in Barrett and Gibson (eds.), 177–97. Kenny, A. J. P 1989 The Metaphysics of Mind (Oxford University Press). Kirk, R. 1986 Translation Determined (Oxford University Press). Kirkham, R. L. 1992 Theories of Truth (Cambridge, Mass.: MIT Press). Koppelberg, D. 1987 Die Aufhebung der Analytischen Philosophie: Quine als Synthese von Carnap und Neurath (Frankfurt: Suhrkamp). Kripke, S. A. 1975 ‘Outline of a Theory of Truth’, Journal of Philosophy 72, 690–716. 1980 Naming and Necessity (Oxford: Blackwell). Kuhn, T. 1970 The Structure of Scientific Revolutions (University of Chicago Press). ¨ Kunne, W. 1983 Abstrakte Gegenst ¨ande (Frankfurt: Suhrkamp). 1993 ‘Truth, Meaning and Logical Form’, in Stoecker (ed.), 1–20. 2003 Conceptions of Truth (Oxford University Press). Larson, R. and G. Segal 1995 Knowledge of Meaning (Cambridge, Mass.: MIT Press). Laudan, L. 1996 Beyond Positivsm and Relativism (Boulder: Westview). Lauener, H. 1982 Willard V. Quine (Munich: Beck). 1990 ‘Holism and Naturalized Epistemology Confronted with the Problem of Truth’, in Barrett and Gibson (eds.), 213–28. Leonardi, P. and M. Santambrogio (eds.) 1995 On Quine (Cambridge University Press). LePore, E. (ed.) 1986 Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson (Oxford: Blackwell). LePore, E. and B. Loewer 1995 ‘Quine and the Attitudes’, in Leonardi and Santambrogio (eds.), 186–205. LePore, E. and B. McLaughlin (eds.) 1985 Actions and Events: Perspectives on the Philosophy of Donald Davidson (Oxford: Blackwell). Lewis, C. I. 1946 An Analysis of Knowledge and Valuation (La Salle: Open Court). Lewis, D. 1969 Convention: A Philosophical Study (Oxford: Blackwell). 1983 Philosophical Papers, vol. I (Oxford University Press). Loewer, B. 1997 ‘A Guide to Naturalizing Semantics’, in Hale and Wright (eds.), 108–26. Lorenz-Meyer, L. 1993 ‘The Architecture and Evidential Base of the Unified Theory’, in Stoecker (ed.), 251–62. Lowe, J. 1997 ‘Objects and Criteria of Identity’, in Hale and Wright (eds.), 613–33. Lukes, S. 1982 ‘Relativism in its Place’, in M. Hollis and S. Lukes (eds.), Rationality and Relativism (Oxford: Blackwell). Lycan, W. G. 2000 Philosophy of Language: a Contemporary Introduction (London: Routledge).

297

298

References McDowell, J. 1998 Meaning, Knowledge and Reality (Harvard University Press). Malachowski, A. 1988 ‘Searle on First Person Meaning and Indeterminacy’, Theoria 54, 25–30. Malcolm, N. 1977 Thought and Knowledge (Cornell University Press). Malpas, J. E. 1992 Donald Davidson and the Mirror of Meaning: Holism, Truth, Interpretation (New York: Cambridge University Press). Marcus, R. B. 1990 ‘A Backward Look at Quine’s Animadversions on Modalities’, in Barrett and Gibson (eds.), 230–43. Matthews, R. J. 1986 ‘Learnability of Semantic Theory’, in LePore (ed.), 49–58. Mellor, D. H. 1995 The Facts of Causation (London: Routledge). Miller, A. 1998 Philosophy of Language (University College London Press). Murphy, J. P. 1990 Pragmatism from Peirce to Davidson (Boulder: Westview Press). Neurath, O. 1944 Foundations of the Social Sciences (University of Chicago Press). Nimtz, C. 2002 W ¨orter, Dinge, Stellvertreter: Quine, Davidson und Putnam zur Unbestimmtheit der Referenz (Paderborn: Mentis). Orenstein, A. 2002 W. V. Quine (Chesham: Acumen). Pap, A. 1958 Semantics and Necessary Truth (New Haven: Yale University Press). Peacocke, C. 1997 ‘Holism’, in Hale and Wright (eds.), 227–47. Peirce, C. S. 1934 Collected Papers of Charles Sanders Peirce, vol. V (Cambridge, Mass.: Harvard University Press). Perry, J. 1979 ‘The Problem of the Essential Indexical’, Nous ˆ 13, 3–21. 1994 ‘Davidson’s Sentences and Wittgenstein’s Builders’, Proceedings of the American Philosophical Association 68, 23–37. Platts, M. 1979 Ways of Meaning (London: Routledge & Kegan Paul). Preston, J. M. (ed.) 1997, Thought and Language (Cambridge University Press). Price, H. H. 1953 Thinking and Experience (London: Hutchinson). Prior, A. 1967 ‘Existence’, in P. Edwards (ed.), The Encyclopedia of Philosophy (London: Macmillan), vol. V, 141–7. 1971 Objects of Thought (Oxford: Clarendon Press). Putnam, H. 1975a Mathematics, Matter and Method: Philosophical Papers Volume 1 (Cambridge University Press). 1975b Mind, Language and Reality: Philosophical Papers Volume 2 (Cambridge University Press). 1988 Representation and Reality (Cambridge, Mass.: MIT Press). Quinton, A. 1973 The Nature of Things (London: Routledge & Kegan Paul). Ramberg, B. T. 1989 Donald Davidson’s Philosophy of Language (Oxford: Blackwell). Ramsey, F. P. 1990 Philosophical Papers (Cambridge University Press). Romanos, G. D. 1983 Quine and Analytic Philosophy (London: MIT Press). Rorty, R. 1972 ‘Indeterminacy of Translation and of Truth’, Synthese 23, 443–62. 1979 Philosophy and the Mirror of Nature (Oxford: Blackwell). 1986 ‘Pragmatism, Davidson and Truth’, in LePore (ed.), 333–55. 1990 ‘Introduction: Pragmatism as Anti-Representationalism’, in Murphy, 1–6. Rumfitt, I. 1995 ‘Truth Conditions and Communication’, Mind 104, 827–62. Rundle, B. 1979 Grammar in Philosophy (Oxford University Press). 1990 Wittgenstein and Contemporary Philosophy of Language (Oxford: Blackwell). 1997 Mind in Action (Oxford University Press). Russell, B. 1935–6 ‘The Limits of Empiricism’, Proceedings of the Aristotelian Society, 36. Ryle, G. 1980 The Concept of Mind (Harmondsworth: Penguin).

References Savage-Rumbaugh, S. 1986 Ape Language: From Conditioned Response to Symbol (Oxford University Press). Savigny, E. von 1988 The Social Foundations of Meaning (Heidelberg: Springer). Scheffler, I. and N. Chomsky 1958 ‘What is Said to Be’, Proceedings of the Aristotelian Society 59, 71–82. Schiffer, S. 1987 Remnants of Meaning (Cambrige, Mass.: MIT Press). ¨ Schroeder, S. 1998 Das Privatsprachenargument (Paderborn: Schoningh). Searle, J. 1969 Speech Acts (Cambridge University Press). 1987 ‘Indeterminacy, Empiricism and the First Person’, Journal of Philosophy 84, 123–46. 1995 The Construction of Social Reality (New York: Free Press) . 1997 ‘The Explanation of Cognition’, in Preston (ed.), 103–26. Sidelle, A. 1989 Necessity, Essence, and Individuation: A Defence of Conventionalism (Ithaca: Cornell University Press). Smart, J. J. P. 1986 ‘How to Turn the Tractatus Wittgenstein into (Almost) Donald Davidson’, in LePore (ed.), 92–100. Smullyan, A. F. 1971 ‘Modality and Description’, in L. Linsky (ed.), Reference and Modality (Oxford University Press), 35–43. Sokal, A. and J. Bricmont 1998 Intellectual Impostures (London: Profile). Stainton, R. J. 1996 Philosophical Perspectives on Language (Peterborough : Broadview). Stoecker, R. 1993 (ed.) Reflecting Davidson (Berlin: de Gruyter). 1995 ‘Davidson, Donald’, in Metzler Philosophen Lexikon (Stuttgart: Metzler), 194–200. Strawson, P. F. 1952 Introduction to Logical Theory (London: Methuen). 1959 Individuals (London: Methuen). 1971 Logico-Linguistic Papers (London: Methuen). 1992 Analysis and Metaphysics (Oxford University Press). 1997 Entity and Identity (Oxford: Clarendon). Stroll, A. 1989 ‘What Water Is or Back to Thales’, Midwest Studies in Philosophy XIV, 258–74. Stroud, B. 1984 The Significance of Philosophical Scepticism (Oxford: Clarendon). Tarski, A. 1949 ‘The Semantic Conception of Truth’, in H. Feigl and W. Sellars (eds.), Readings in Philosophical Analysis (New York: Appleton-Century-Crofts), 52–84. 1969 ‘Truth and Proof’, Scientific American 220, 63–77. 1983 Logic, Semantics, Metamathematics (Indianapolis: Hackett). Teichmann, R. 1992 Abstract Entities (Basingstoke: Macmillan). Travis, C. 1997 ‘Pragmatics’, in Hale and Wright (eds.), 87–107. Tugendhat, E. 1982 Traditional and Analytical Philosophy (Cambridge University Press). Unger, P. 1984 Philosophical Relativity (Oxford: Blackwell). Vuillemin, J. 1986 ‘On Duhem’s and Quine’s Theses’, in Hahn and Schilpp (eds.), 595–618. Waismann, F. 1968 How I See Philosophy (New York: Macmillan). Welch, J. R. 1984 ‘Referential Inscrutability: Coming to Terms Without It’, Southern Journal of Philosophy 22, 263–73. White, A. 1970 Truth (London: Macmillan). 1972 ‘What We Believe’, in N. Rescher (ed.), Studies in the Philosophy of Mind (Oxford: Blackwell), 69–84. Wiggins, D. 1997 ‘Meaning and Truth Conditions: From Frege’s Grand Design to Davidson’s’, in Hale and Wright (eds.), 3–28. Williams, S. G. 1998 ‘Meaning and Truth’, in Craig (ed.), 219–26. Williamson, T. (1990) Identity and Discrimination (Oxford: Blackwell). Wimmer, H. and J. Perner 1983 ‘Beliefs about Beliefs’, Cognition 13, 103–28.

299

300

References Wittgenstein, L. 1922 Tractatus Logico-Philosophicus (London: Routledge & Kegan Paul). 1967a Philosophical Investigations (Oxford: Blackwell). 1967b Zettel (Oxford: Blackwell). 1975 Philosophical Remarks (Oxford: Blackwell). 1978 Remarks on the Foundations of Mathematics (Oxford: Blackwell). Wood, A. H. 1938 History and Geography of Tonga (Auckland: Wilson & Horton). Wright, C. 1980 Wittgenstein on the Foundations of Mathematics (London: Duckworth). 1982 Frege’s Conception of Numbers as Objects (Aberdeen University Press). 1997 ‘The Indeterminacy of Translation’, in Hale and Wright (eds.), 397–426. Yablo, S. 1998a ‘Does Ontology Rest on a Mistake?’, Proceedings of the Aristotelian Society Supplementary Volume 72, 229–61. 1998b ‘Essentialism’, in Craig (ed.), 417–22.

Index

ability 263, 279 abstract entity/object 16, 62, 63, 69, 230 see also nominalism; Platonism; singular term, abstract ‘acquiescing in our mother tongue’ (Quine) 203 action (intentional) 35–36, 249, 278–279 see also behaviour adverbial modification 158, 247 Alston, W. 68 ambiguity 61–62 analysis (conceptual, logical) 22, 101 connective vs. reductive 115, 244 see also explication; reduction; logical form analytic/synthetic distinction 24, 64, 96–97, 251, 285 borderline cases 74, 83, 89 Davidson on 31, 214–215, 252 a ‘dogma of empiricism’ or ‘pseudo’ 25, 81 dynamic conception of 87–92 explanatory value of 82, 92 intuitive conception of 74, 82 projective quality of 73 Quine’s attack on 1, 8, 18–21, 26, 64–70, 72–95, 167 presupposes indeterminacy of translation 77 Quine’s own version of 82 see also methodological monism analytic philosophy 3–7, 22, 23, 40 see also linguistic turn analytic truths definition of 73, 79 constitutive of meaning/concepts 84–86, 87, 88 merely obvious/firmly entrenched 81, 93 vs. logical truths 71

analytical hypothesis 172, 190, 215 ‘and’ 159, 178 animals 21, 39, 265, 269–293 anomalous monism 3, 33, 123, 264 Quine and 229 anthropocentrism 124–126, 270–271 anthropology 166 and interpretation from scratch 170, 190, 199, 218–219, 222 philosophical 2, 37 see also human beings; human nature Apel, K. O. 202 apparatus of individuation 174, 216, 220–223 a priori 8, 23, 26, 40, 71–72, 117, 290 analytic theory of 71–72 Quine’s attack on 86–87 see also gradualism argument from above 173, 224–232 argument from below 174, 224 Aristotle 23, 25, 41, 99, 102, 132, 159 artificial intelligence see cognitive science assent/dissent 33 intensional and intentional 9, 33, 180, 183, 184, 223 ‘surface’ 180–181 see also holding true assent-conditions 140, 172, 183, 185–188, 221, 238 preserved by proxy-functions? 222–224 vs. conditions of utterance 140, 171, 193, 261 see also distal vs. proximal theory assertion 162, 164 attribute 26, 46, 47, 50, 56, 67, 147 essential 98 Austin, J. L. 5 axiomatization 105 axiom schemata 245 finite 134, 144–145, 245–246

[301]

302

Index Baker, G. P. 155, 249 behaviour 143, 189 explanation of 34, 268, 271 linguistic (dispositions to) 171, 207, 208, 211, 212 non-linguistic 197, 239, 274, 276, 280, 287, 292 see also action; human beings; human nature behaviourism 1, 3, 20, 83, 93, 176, 268 and Davidson 183, 188 see also assent, surface; stimulus meaning; third-person perspective behaviourist approach to meaning/language 139–142, 171–172, 180–182, 209–214, 224, 233 ‘mandatory’ 181, 211 beliefs 192 about beliefs 286–288, 291 and the concept of belief 286–291 interplay with meaning 195, 197 not attitudes towards propositions but dispositions 264, 266, 268, 274 and surprise 286–287 and truth 286, 288–291 see also holism of the mental; propositional attitude; thought Bentham, J. 139 Bilgrami, A. 250 bilingual 166, 176, 177, 218 Black, M. 112 Boghossian, P. 84 Brentano thesis 200, 227, 229, 230 Burge, T. 212 Bushism see malapropism canonical notation 2, 29, 44, 54, 111, 171, 265 see also explication canonical proof 236–237 Carnap, R. 2, 5, 14–15, 17, 22, 24, 29, 31, 40, 63–65, 73, 74, 77, 89, 148, 167, 218 see also internal/external; meaning postulate Cartesianism 4, 12, 21, 186 see also mentalism Cartwright, R. L. 48 causal explanation vs. justification 32, 186, 189 causal theory of meaning and content 259–264 and truth-conditional theory 260 see also behaviourist approach to meaning causation 223 and assertion/truth-conditions 259, 260, 261, 262

deviant vs. typical causal chains 261–262 see also events; facts as causes centre of gravity 223 charity (principle of) 33, 37, 38, 141, 172, 194–199, 202, 226, 237 across the board 185, 196, 215 heuristic maxim or constitutive 184, 195, 238 maximizing agreement 195–196 optimizing agreement 196 a precondition of interpretation from scratch 195 principle of correspondence vs. principle of coherence 194 children 39, 269, 284, 289 Chomsky, N. 142, 171, 210, 225, 244–245, 256, 260, 269, 293 Churchland, P. 24 circularity charge against analyticity 73–77 deploys intensional concept 75 itself circular 77 class 46, 47, 50, 129 cognitive science 4, 24, 219 ‘A Coherence Theory of Truth and Knowledge’ (Davidson) 108 collateral information 171, 173, 177 communication 18, 33, 139, 203–207, 249, 250–259 Davidson’s model of 254 as precondition for language 290 as precondition for thought 260, 289–290 compositionalism 139, 145, 235–237, 245, 246, 255 and contextualism 145 concepts 124, 224, 274 holism of 285–286 and language 276–281, 288, 293 and thoughts 274–276 conceptual vs. factual/theoretical change 87–92 see analytic/synthetic; philosophy and science conceptual role semantics 84, 261 conceptual scheme/framework 24, 42, 64–70 vs. content 21, 186, 198–199 content 125, 184, 200, 259, 282 empirical 77–78, 86 vs. mood 160 see also what is said; propositional attitude contextualism 6, 139, 145–146, 221–223 conventions 1, 11, 72, 80, 122, 125 and communication 253–259 ‘deep need for’ (Wittgenstein) 93–95 definition of 253 implicit 92–95, 253

Index of individuation 101 and language 252–253 lexical 250 vs. norms 250 vs. speaker’s intention 162, 258 see also normativity of meaning; rules Convention T 104, 110, 117, 124, 149 relativized to speakers and times 150 conventionalism 72, 79 see also linguistic doctrine of analyticity conversational implicature 158, 258 cosmic complement 175, 222 criteria of identity 46, 49, 210 formal 50 vs. principles of individuation 51–52, 99 see also ontological admissibility; intensionality; thought; truth-bearers Czolbe, H. 20 Darwinism 18 Davidson, D. biographical data 13 developments in his views 107–108, 202, 233, 248, 250 compared to Quine 1, 2, 6, 26–33, 203–207, 214–216, 219–220 decision theory 21, 34–35, 192–193 de dicto vs. de re 97–98, 101, 265 definite descriptions 97 Russell’s theory of 44, 54 definition 79–81, 115, 139 contextual/implicit 84–85 discursive vs. legislative definitions/postulates 80 vs. empirical proposition 89–92 deflationism see ontological deflationism; truth, deflationary theory of demonstrative 163, 265 see also indexicals, paratactic Descartes, R. 25, 293 desire 192, 197 determinable 51 Dewey, J. 18, 21, 124–126, 139 discrimination 265, 273, 277 vs. classification/concepts 277–279 disquotation 105, 141, 202, 204, 239–241, 244, 264 see also homophonic vs. heterophonic translation; truth, disquotational account of; T-sentences distal vs. proximal theory of assent/meaning 33, 185–188, 247, 260 dualism 21 conceptual/property 18, 33–36, 188, 229

Kantian 21, 23 see also conceptual scheme vs. content Duhem, P. 77 Duhem–Quine hypothesis 77 Dummett, M. 10, 82, 113, 128, 143, 151–152, 240–249, 257, 274, 275, 276 elimination (of higher phenomena) vs. derivation/extraction 32, 169, 170, 182–183, 184, 188, 199, 200, 234 vs. reduction 1, 29–30 empathy 187 empiricism 1, 2, 17, 19, 22, 127, 185, 231 British 13–14 consistent/logical 13–14, 22 and Davidson 31, 128, 189, 261 epistemological vs. semantic 15 moderate 71 radical 23, 25, 71–72, 93 see also myth of the given entailment 49, 235, 284 paratactic account of 235 enumerability 50, 59 epistemic possibility/necessity 87, 96 epistemology 37, 180, 233 naturalized 29, 31, 38, 268 vs. ontology 209–210 essence/essentialism 40, 95–101 and naturalism 96 of origin 100 eternal sentence 121 Evans, G. 216–217, 218 events 33, 223, 230, 247, 249, 261 describable in different ways 34 ‘everything’ 41 evidence 185, 186, 192, 220, 268 Evnine, S. 3 existence (‘there is’) 40, 52 ambiguity of 60–62 common vs. philosophical sense of 60 philosophical vs. scientific questions/statements of 25, 42, 63, 69 and quantification 56, 58 see also internal vs. external; ontology existentialism 18 experience ‘recalcitrant’ 86, 91, 93 explanation (of meaning) 84, 87, 89, 155, 158, 244, 249, 262 vs. disquotation 204 speaker’s 206, 212 see also definition explication 27, 29, 46, 47, 68, 75, 181 see also canonical notation, regimentation

303

304

Index extensionalism 1, 18, 68, 70, 157 Davidson’s version of 148, 235, 249 driving force behind Quine’s attack on analyticity 73–74 unconditional 42, 46 see also intensionality extensionality problem 169, 234–239 externalism 212–213, 230, 262 and first-person authority 263, 264

general term see predicate gestures 198 Gibson, R. F. 3 Goodman, N. 46 gradualism 86 grammar see syntax Grandy, R. 195–197 Grice, H. P. 5, 73, 76, 167, 205, 258 see also conversational implicature

fact of the matter about meaning (semantic) 6, 167, 169, 209, 214, 225, 227, 229, 231 are all physical (Quine) 227 about reference 180, 203, 224 facts 130–131, 231 as causes 223, 261 don’t make sentences true 131 not in the world 130 vs. propositions 130 fallibilism 86 Feyerabend, P. 30 fictional objects 59, 61 vs. abstract objects 60 Field, H. 47, 112 first-person authority 169, 206–207, 213 and behaviourism 206 defended by Davidson 202, 206, 263–267 and indeterminacy 206–207 vs. introspection 263 precondition of interpretation 263–264 see also externalism ‘flight from intension’ (Quine) 11, 46, 76, 125, 167, 182 circular 227 depends on indeterminacy of translation 52, 69 presupposes physicalism 227 Fodor, J. 211, 269 force 148 and mood 161, 164 vs. sense 160 vs. speaker’s intention 161 theory of 160 formal correctness 103, 104, 106 Foster, J. 235 foundationalism 18, 20, 225 Frege, G. 4, 13–14, 23, 32, 62, 71, 145, 146, 157, 160, 224, 231, 276 sense/meaning distinction 26, 138 From a Logical Point of View (Quine) 14, 19 full-bloodedness see modesty

Haack, S. 20 Hacker, P. M. S. 155, 198, 249 Hacking, I. 36, 151 Harman, G. 74, 85, 208 Heal, J. 152, 197 Hegel, J. W. F. 21, 23, 36 Heidegger, M. 40 hermeneutics 17, 18, 22, 34–35, 198, 202 hermeneutic circle 177 higher phenomena see intensionality; meaning; naturalism; thought holding true 183, 191, 237, 261, 263 see also preferring true holism 19, 21, 77, 141, 146, 224–225, 261, 273 vs. building block approach 220–221, 274–275 epistemic (of confirmation) 6, 15, 77, 86, 127, 226 moderate 78 of the mental/thoughts 34–35, 273, 281–286, 292 principles of 282–285 semantic 77, 144, 218 see also contextualism; Duhem–Quine hypothesis; web of beliefs holophrastic/holodoxastic 171, 274–275, 281, 292 homophonic vs. heterophonic interpretation/translation 202, 206, 239–242, 263–264 Hookway, C. 3, 176, 217 Horwich, P. 133–134 human beings (capacities/practices) 12, 36, 63, 99, 108, 124, 189, 202, 231, 259 see also anthropocentrism human nature (‘common behaviour of humankind’) 197–198, 214, 215 humanity (principle of) 196–197 Hume, D. 15, 96 Humpty-Dumpty 257

Gadamer, H. G. 202 ‘gavagai’ 171, 173, 174, 216–219

ideal language philosophy see canonical notation; language, ordinary/natural vs. ideal; ordinary language philosophy

Index identifiability 51–52, 59 identity 55, 100, 175 law of 85 theoretical identifications 100 token-token vs. type-type 34, 229, 249 idiolect 256 ‘In Defense of a Dogma’ (Grice/Strawson) 73 indeterminacy of translation 1, 9–10, 27, 31, 51, 148, 167, 169, 172, 176, 199, 200–232 extent of 172–175, 181, 185, 215–216 ontological rather than epistemological 209 a philosophical point 218 a reductio of behaviourism 181–182 and underdetermination of theory by evidence 225–227, 231 vs. vagaries of translation 173 see also fact of the matter; inscrutability of reference; translation manuals indexicals 53, 121, 150, 239, 241 indirect speech paratactic account of 265–266 inextricability thesis 83, 93, 214 Inquiries into Truth and Interpretation (Davidson) 31, 142 inscrutability of reference 6, 27, 31, 174, 203, 214, 215–224 see also ontological relativity; proxy-functions instrumentalism 18, 19, 38, 53, 99, 218, 239, 271 intensionality/intensions 2, 60, 73, 116, 123, 147, 207–208 criteria of identity for 27, 62, 76, 167, 200, 207 vs. extension 26 nihilism about 32 see also analytic/synthetic distinction circularity charge against; Frege, sense/meaning distinction internal vs. external questions 64–65, 70, 73 relations 94 interpretation and cooperation 191, 214, 232, 239 from scratch 189, 190, 232, 259 interpretationalism 205 James, W. 18, 19, 20, 22 Jeffrey, R. 192 Kant, I. 5, 14, 23, 24, 32, 36, 37, 70, 95, 108, 249, 276 see also synthetic a priori

Kenny, A. J. P. 276 Kirk, R. 218 knowledge 21 see also epistemology; instrumentalism Kripke, S. A. 95, 99, 100, 260 Kuhn, Th. 91 ¨ Kunne, W. 62, 135 language 36, 293 aim of 113 as a form of behaviour/interaction 2, 31, 38, 118 ‘creativity of’ 145, 245 formal vs. pragmatist approach to 23, 124 natural/ordinary vs. artificial/ideal 3, 23, 34, 89, 103–105, 115, 142, 148, 150, 151, 157–165, 245 ‘no such thing as a shared language’ 6, 254, 259, 290 obligation to 257 semantically closed 104, 111 social nature of 22, 211, 260, 290 source vs. target/background 170, 202, 208 vs. theory 256 and truth 126 see also canonical notation; communication; conventions; ideal vs. ordinary language philosophy; meaning language-acquisition 20, 145, 155, 168, 189, 211, 244, 245–246 LePore, E. 211 Lewis, C. I. 13, 19 Lewis, D. 189, 195–196, 252–259 lingualism 269, 273, 283, 291 linguistic doctrine of analyticity/logical necessity 72, 81, 92, 101 linguistic vs. factual component of truth 78 Quine’s partial agreement with 85, 96 linguistic framework see conceptual scheme linguistic turn 4, 24, 38, 138 linguistics 143, 166–167 Locke, J. 25 logic 2, 13–14, 23, 34–35, 38, 78, 86 combinatory 53 formal vs. philosophical 3 illogical/prelogical mentality 87, 172 intuitionist 86 and physics 6 quantum 86 see also semantics, formal logic of science 14, 40 logical constants/particles 44, 71, 73, 80, 84, 92, 158, 177, 243

305

306

Index logical form/structure 34, 79, 114, 158, 194, 215 vs. individual concepts 243 see also quantification logical positivism 1, 14, 22, 23–24, 29, 37, 40, 71–72, 78, 83, 94–95 Davidson and 16–18, 148 pragmatism and 21–22 Quine and 15 logical pragmatism 1, 22 logical truths see analytic truths; necessity; tautologies McDowell, J. 240, 244 Mach, E. 14, 15 malapropism 82, 253–255, 257–258 Malcolm, N. 270, 275 Marxism 18 material adequacy 103–104, 106, 110 materialism eliminative 30, 33 vs. physicalism 29 see also reduction vs. elimination mathematical cyclist 98 mathematics 23, 71, 78, 83, 86 constructivist 86 maxim of minimum mutilation 86 Mead, G. H. 18, 21, 290 meaning 1, 76, 137–165, 211 arbitrary 250, 253 and assent-conditions 185 ‘autonomy of’ (Davidson) 162, 258 cognitive 140 and communication/understanding 1, 205, 210, 213 derived from more basic phenomena 188–190 ‘first’ (Davidson) 258 immanence of 154–155 intersubjective accessibility of 186–189 intuitive/traditional notion of 137, 168, 200, 206–208, 209 literal 162, 205, 212, 228, 257–259 manifestability of 211–214, 221, 268 natural 95 and naturalism 169 as obscure/problematic 33, 109, 184 vs. reference 106, 138 referential account of 138, 147 speaker’s 205, 212, 257–259 and translation 204–205 and truth-conditions 141, 147, 148, 151–157, 159, 234, 239 type-expressions, a feature of 153

see also communication; indeterminacy of translation; ‘meanings’; stimulus meaning; theory of meaning; use meaning postulate 75, 81 ‘meanings’ (reification of meaning) 37, 50, 76, 79, 138–139, 148, 209 see also meaning vs. reference; ‘myth of a museum’ Meinong, A. 42 mentalism 118–119, 123, 269 see also Cartesianism metalanguage vs. object language 104–105, 114 metaphor 3, 258 metaphysics 2–4, 14, 29–30, 40 descriptive vs. revisionary 30 see also ontology methodological monism 28 Mill, J. S. 14, 23, 71, 86, 231 mind 21, 31, 34, 189, 228 manifestability of 268–269 see also psychology; thought mistake 278, 286, 288 modal logic (quantified) 96–98, 101 modality 27, 48, 265 modesty 157 explanatory 240–244 incompatible with Davidson’s project 249, 255–256 psychological 244–250, 256 mood 160–165, vs. indicative core 163–165 indicator/setter 160, 164 see also force multiple realizability see supervenience myth of the given 22, 186, 188 ‘myth of a museum’ (Quine) 79, 84–85, 138 names 44, 100, 103, 121 natural kinds 99, 100, 122 naturalism 2, 3, 11, 12, 27–30, 33, 66, 69–70, 96, 169, 182, 209–210, 211, 231, 268, 270 anti-naturalism 5 epistemological 3–7, 27, 127 metaphilosophical 27 ontological 3–7, 27, 40, 230 origin of 20 supernaturalism 2–4, 12, 60 see also elimination; philosophy and science; scientism necessary a posteriori 100 necessity 8, 14 vs. a priority and analyticity 95 logical vs. physical/natural 71, 96 metaphysical 96, 100

Index see also conventionalism; de dicto/de re; modality; linguistic doctrine of analyticity/logical necessity Neumann, J. von 208 Neurath, O. 16, 21 ship metaphor 28 neurophysiology 31, 185, 186–188, 211, 228, 230, 271 Nimtz, C. 174 nominalism 16, 60 about truth-bearers 119, 120, 125 conditional 42, 46, 47 thoroughgoing 46 non-contradiction, law of 85, 87 nonsense 95, 123 vs. obvious/trivial falsehood 85 normativity of meaning and concepts 85, 251, 262, 277–279 see also convention; mistake; rules norms see rules norms of rationality (Davidson) 33, 34, 195, 251 constitutive of thought/language 195, 198, 251–252 vs. lexical norms/rules 251 numbers 46, 47, 62, 64 objects common sense (material object) vs. logical (object of reference/discourse) 62, 63, 125, 266 and existents 41 kinds of 42 physical 43, 46, 47, 50, 285 observation sentence 16, 78, 81, 86, 185; see also protocol sentence entering wedge of radical translation 171–172, 177, 179, 192, 194, 261 individualistic vs. social definition of 186–187 Occam’s razor 45 occasion sentence 120, 121, 171 ‘On what there is’ (Quine) 41 ontic decision see ontological reform ontological admissibility ‘no entity without identity’ 46, 49, 51, 207, 209 standard of 43, 49 ontological (existential) commitment 43–44, 67, 68, 204 appraisal/clarification vs. revision/reduction of 52, 67 criterion of 42, 43, 58 general idea of 43, 56 intensionality of Quine’s criterion 47–49

restricted to canonical notation 52–53 restricted to quantification over infinite domains 54–56 ‘to be is to be the value of a bound variable’ 45 ontological deflationism 56, 59, 60–63 existential 60–62 linguistic 63–69 objectual 62–63 ontological inflationism 56–58 ontological parsimony 42, 45, 49 see also Occam’s razor ontological problem 41, 43, 59 ontological reform 46, 66, 68 Ontological Relativity (Quine) 201 ontological relativity 201, 203–204 and inscrutability of reference 204, 215–216 ontology 25, 40–70 vs. ideology 219–220 instrumentalist conception of 38, 42 Quine’s preferred 40, 43, 47 opacity vs. transparency 26, 48–52, 97–98, 265, 272, 275 ordered pair/n-tuple 46 ordinary language philosophy 24, 37, 52, 72 vs. ideal language philosophy 16–17 Orenstein, A. 3 ostension 216, 222 deferred 222 paradigm 91 paraphrase 66–69 see also explication; regimentation paratactic theory of mood 163 Peirce, C. S. 18, 19, 21, 23, 31, 39 Perry, J. 53 phenomenalism 16 philosophy 23, 25, 69 ‘armchair’ 24, 68 first 28 and morality/politics 21–22 queen of the sciences 25 and science 2, 4, 11, 21, 24, 28, 31, 68, 70, 73, 96 systematic 36 underlabourer 25 see also analytic/synthetic; linguistic turn; naturalism; scientism philosophy of language, applied vs. pure 36 physicalism 16, 18, 227–231, 249 see also materialism physics 16, 30 ‘Platonic riddle of nonbeing’ (Quine) 44, 208

307

308

Index Platonism 4, 12, 14, 23, 47, 60, 63, 84–85, 250 about truth-bearers 119 posit 38, 42, 43, 47, 69 possible worlds 99 criteria for transworld identity 99 pragmatic considerations/questions 19, 22, 42, 64, 65, 73, 86, 218 see also maxim of minimum mutilation pragmatism 1, 4, 18–23, 39, 135, 268, 269 Davidson and 19–23, 124 Quine and 19–23, 38, 69, 73 predicate (open sentence) 26, 48, 54, 58, 67, 96, 105, 219–220, 245 sortal 49 preferring true 192–194 Prior, A. 58 prior vs. passing theory 252–259 private language 252, 290 pro-attitude see desire pronouns 54, 57 interrogative vs. relative 76, 125, 266 property see attribute propositional attitudes 27, 34 ascriptions of 158, 264–265 and indeterminacy 200, 219–220 quotational account of 265–266 see also beliefs; desires; thoughts propositions 26, 50, 62, 133–134, 147 and sentence meaning 134, 153–154 see also what is said protocol sentence 16 proxy-functions 175, 204, 219–224 psychological laws 35 psychological plausibility 196, 219 psychology 21, 184, 200 sui generis 36 psychophysical laws 34–35, 272 Putnam, H. 5, 20, 86, 89, 128, 212 quantification 54, 215–216 into modal contexts 97 non-nominal 58, 107, 134 objectual vs. substitutional 45, 49–51, 56–58, 97, 133 into predicate or sentence position see non-nominal quantification into quotation marks 105 see also existence; variables Quine, W. V. 13 and Davidson see Davidson and Quine developments in his views 19, 28, 81–82, 177, 186–188, 204 quotation 27, 103 radical interpretation (Davidson) 9–10, 32, 144, 148, 166–199, 233, 237, 241, 251

evidence for 33, 183, 188, 190, 213, 237 and radical translation 182, 194, 202 as sufficient but not necessary for understanding 213 radical translation (Quine) 1, 9–10, 27, 87, 166, 167, 170–182 ‘begins at home’ 168, 202 breakdowns of 177 evidence for/constraints on 168, 172, 226 a form of dialogue/interaction 176, 179 purports to be presuppositionless 177 tacitly violates behaviourism 175 radical translation/interpretation 213 a heuristic device 167 importance of 168 presuppositionless/radical 168, 169, 179, 180, 183, 193–194 as sufficient for interpretation from scratch 169, 195, 214, 219 task vs. procedure 168 Ramberg, B. T. 113 Ramsey, F. P. 136, 192 rationalism 22, 261 rationality/reasoning 21, 22, 34–35, 87, 185, 219, 279–281, 282, 284, 290 ‘real’ 43, 50, 59 realism 126 see also truth, anti-realist and correspondence theory of reasons/reasoning 223 and causes 34–35, 247 see also causal explanation recalcitrant idioms 158 recursion 105, 106, 111, 145, 152, 170, 245–246 reduction 14, 15, 17, 26, 112 analytic vs. scientific 228–229 vs. elimination 29–30 see also materialism, eliminative reference defusion of/instrumentalist conception of 117, 169, 214, 220–223 see also satisfaction regimentation 42, 43–46, 50, 52 relativism 37 alethic 127 conceptual 34 representationalism 20, 264–267 rigid designator 100 Rorty, R. 20, 21, 108, 166, 225 rules 79, 85, 101, 140, 154, 278 constitutive vs. regulative 252 Davidson’s attack on 250 of inference vs. axioms 94 individual or shared (conventions) 250 and truth 85 unconscious 218, 247–248

Index see also linguistic doctrine of analyticity/necessity; normativity of meaning and concepts Rundle, B. 154 Russell, B. 4, 13–14, 23, 85 Ryle, G. 5, 38, 60–61, 62, 268, 280 satisfaction 105–106, 128, 220 definition of 111, 112, 117 scepticism 3, 33, 37, 126 Schlick, M. 16, 24 ¨ Schonfinkel, M. 53 science 68, 77, 78, 127, 268–269 natural vs. Geisteswissenschaften 17 unity of 15, 17, 29 see also philosophy and science scientific progress scientism 15, 17, 28, 30, 31 see also naturalism Searle, J. 206 self-evidence 82, 83 semantic ascent 65, 106, 135 semantic intention 258, 259 semantic nihilism 207, 208–209 semantic paradox 103, 104, 111–112 semantic scepticism 208, 213, 231 semantically relevant facts 207, 210, 212, 213, 219, 220–221, 223, 233 semantics 103, 105, 143 antecedes ontology and epistemology 51 formal 2, 31, 158 intensional 148, 156, 223 vs. pragmatics 158–159 traditional 200 translational 155 sense-data 16, 185, 186 sentence 26–33, 105, 139 vs. names/words 140, 147, 178 non-declarative 158, 159–165, 178 limiting case of predicate 106 open see predicate unit of communication 139, 146 sentence meaning see proposition sententialism 119–124, 126, 132, 153 sequence 105 infinite 105, 129 set see class set-theory 14–15, 208 shadows 222 Sheffer, S. 13 Sidelle, A. 101 sincerity 191 singular terms 26, 44, 62, 111 abstract 56, 60 elimination of 44, 53–54, 100 mark of 55

see also apparatus of individuation; predicate Skinner, B. F. 21 slingshot 129–130, 223 Socrates 115 ‘something’ 58, 59 speech acts 120, 122, 125, 140, 178, 191, 193, 221–222 spelling 103 stimulation location of stimulus 259, 260 neural (‘triggerings’) 1, 16, 140, 172, 180 see also assent-conditions; distal vs. proximal theory; sense-data stimulus-analytic 81 stimulus meaning 27, 137, 224 behaviourist ersatz of meaning proper 181, 209 definition of 141 neural character of 185, 212 privacy of 186–188 stimulus-synonymous 177 Strawson, P. F. 5, 50, 51, 55, 62–63, 68, 73, 76, 110, 115, 130–131, 135, 143, 167 see also conceptual analysis structuralism, global see reference, instrumentalist conception of supervenience 228, 229, 230 ‘Swampman’ (Davidson) 262–263 synonymy 27, 47, 73, 75–76, 140, 148, 152, 156, 209, 224, 235 vs. material, nomological or logical equivalence 73, 132, 223, 234, 236, 238 Quine’s attack on 167 syntax 67, 105, 244 synthetic a priori 14, 24, 80, 83 system of the world (global theory) 127–128 tacit/unconscious knowledge 143, 244, 255–256 Tarski, A. 2, 14, 17, 31, 102, 118–119, 131, 136, 157 and Davidson 32, 107, 109, 234 and Quine 106 see also truth, semantic theory of tautologies 71, 73, 94, 156, 243 Teichmann, R. 66 that-clause 123, 125, 134, 266, 273 intentional rather than object accusative 266 see also pronoun, interrogative vs. relative theoretical sentence 173, 215, 241, 260–261 theory of meaning 17, 107, 110, 142–165, 168, 234 analytic vs. constructive 142, 151–152 empirical verifiability/status of 144–145, 151, 233, 240

309

310

Index theory of meaning (cont.) extensional adequacy (holism) of 144, 235 form of 142, 143, 147, 153, 245 interpretative/meaning-giving 142, 149, 151, 152, 158, 183, 234, 235, 240, 243 modest construal of 248–249 non-circularity of 144–145, 148, 234 sufficient but not necessary for interpretation/communication 143, 151, 159, 168, 240, 248–249 and theories of reference, sense and force 138, 148, 150, 160, 163, 209, 239 and theory of understanding 143–144 truth-conditional 32, 123, 146, 150–165, 182, 233–239 see also meaning; radical interpretation theory of truth vs. explicit definition of truth 111 and theory of meaning 103, 109–110, 114, 136 third-person perspective 21, 31, 183, 210–214, 268 thought/thoughts 249, 268 and behaviour 269 criteria of identity for 269, 282 in the head 264 intensionality of 272–273, 292 and language 2, 3, 11, 38, 259, 270–293 logical connections between 35–36, 93, 281–286 onto- and phylogenesis of 3, 38, 289 see also belief; desire; language-acquisition; propositional attitudes translation 37, 49, 107, 166, 167 as either presuppositionless or indeterminate 224, 232 between object- and metalanguage 124, 202, 234 indispensability of (Quine) 203, 204 normal vs. from scratch vs. radical 168 on a par with physics 225 translation manuals 170 behaviourally or physically equivalent but semantically incompatible 173, 206–208, 227 vs. theory of meaning 183, 242 triangulation 39, 260, 262 truth 8–9, 74, 102–136, 149, 184 anti-realist/epistemic theories of 19, 108, 124, 126–128 atemporal 123 classical Aristotelian conception of 102–103 coherence theory of 108 connections with other concepts 132, 135 correspondence theory of 20, 107, 128–131;

fact- vs. satisfaction-based versions of 107, 108, 128–130 definition of 103, 105, 106, 107, 114, 117, 131, 132, 133, 240 deflationary account of 20, 108, 115, 131–136; definition of 131–132; Quine and 132 disquotational account of (Quine) 106, 119, 123, 131–133, 135 ecumenical vs. sectarian conception of 127–128 hierarchy of truth-predicates 111–112 immanence of 126, 132 indefinable and basic (Davidson) 108, 111, 112, 115, 136 vs. justification 127 minimalism about 133, 153–154, 156 pragmatist theory of 20, 108, 218 as property 132–133 redundancy theory of 106, 108, 132 relativized to a language 113, 118, 124, 126 relativized to speakers and times 119–120, 123, 153 semantic theory of (Tarski) 3, 17, 32, 103–110, 123, 152; and the correspondence theory 128; and the definition of truth 109–118 vs. truth of 123 value of 132, 135 see also truth-theory truth-bearers (‘truths’) 31, 118–126 criteria of identity for 121, 122, 123, 125, 126 are sayables and thinkables 125–126 see also Platonism; nominalism; mentalism; sententialism; proposition truth-conditions 17, 32, 107, 155–156, 208, 221, 242 and assent-conditions 141, 194 statement of vs. definition of truth 116–117 see also theory of meaning, truth-conditional ‘Truth and Meaning’ (Davidson) 107, 147, 233 truth and meaning see meaning and truth; theory of meaning; truth-conditions truth by virtue of meaning/convention/ definition 14, 71, 73, 79–81, 84 truth-theory 157, 159, 163, 165, 215, 233, 265 requires rules 250 shared 253 vs. theory of truth 110, 118, 142 T-schema 104, 115, 149 T-sentence 32, 103, 113, 117, 122, 128, 149, 153, 155–157, 160, 221, 239 causal status of 237

Index empirical or conceptual 107, 109, 114, 115, 116, 240 meaning-giving vs. rogue 142, 234, 235, 236, 237–238 relativized to speakers and times 120, 150, 235 triviality of 114 see also disquotation ‘Two Dogmas of Empiricism’ (Quine) 14, 72–73 type vs. token 90, 119–124 underdetermination of theory by evidence 10, 127, 173, 225–226, 231–232 understanding 37, 143, 201, 233, 239 basis of 244, 249 defeasibility of 205 failure of vs. disbelief/factual ignorance 83, 84, 93 immediate or based on inferences 241, 244, 246 and interpretation/translation 202–205, 241, 255–256 requires shared language 258 as theory-construction 188, 249 see also communication unified theory of meaning, thought and action 33, 37, 184, 191, 194, 197, 215

use 69, 139–142, 205, 212, 263 arbitrary without rules 94 pragmatic vs. semantic aspects of 122 utilitarianism 18, 19 variable 44, 53, 56, 57, 62 see also pronouns verificationism 15, 141, 171 in Quine 77, 225–226, 231 Watson, J. B. 21 web of beliefs 26, 68, 77, 79, 86, 93, 285 what is said 122, 124, 125, 134 see also that-clauses White, A. 266 Whitehead, A. N. 13 Wiggins, D. 239 Wittgenstein, L. 4, 17, 23, 24, 62, 63–65, 85, 93, 124–126, 135, 152, 189, 197, 230, 239, 277 compared to Quine and/or Davidson 4–5, 38, 140, 146, 147, 268, 269 Tractatus 14, 17, 142–143, 145, 146, 157 see also normativism; private language; use Word and Object (Quine) 167, 186 Zermelo, E. 208

311